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_rb_node *nd,
  47                                      struct thread *th, bool lock);
  48 static int append_inlines(struct callchain_cursor *cursor, struct map_symbol *ms, u64 ip);
  49
  50 static struct dso *machine__kernel_dso(struct machine *machine)
  51 {
  52         return map__dso(machine->vmlinux_map);
  53 }
  54
  55 static void dsos__init(struct dsos *dsos)
  56 {
  57         INIT_LIST_HEAD(&dsos->head);
  58         dsos->root = RB_ROOT;
  59         init_rwsem(&dsos->lock);
  60 }
  61
  62 static void machine__threads_init(struct machine *machine)
  63 {
  64         int i;
  65
  66         for (i = 0; i < THREADS__TABLE_SIZE; i++) {
  67                 struct threads *threads = &machine->threads[i];
  68                 threads->entries = RB_ROOT_CACHED;
  69                 init_rwsem(&threads->lock);
  70                 threads->nr = 0;
  71                 INIT_LIST_HEAD(&threads->dead);
  72                 threads->last_match = NULL;
  73         }
  74 }
  75
  76 static int thread_rb_node__cmp_tid(const void *key, const struct rb_node *nd)
  77 {
  78         int to_find = (int) *((pid_t *)key);
  79
  80         return to_find - (int)thread__tid(rb_entry(nd, struct thread_rb_node, rb_node)->thread);
  81 }
  82
  83 static struct thread_rb_node *thread_rb_node__find(const struct thread *th,
  84                                                    struct rb_root *tree)
  85 {
  86         pid_t to_find = thread__tid(th);
  87         struct rb_node *nd = rb_find(&to_find, tree, thread_rb_node__cmp_tid);
  88
  89         return rb_entry(nd, struct thread_rb_node, rb_node);
  90 }
  91
  92 static int machine__set_mmap_name(struct machine *machine)
  93 {
  94         if (machine__is_host(machine))
  95                 machine->mmap_name = strdup("[kernel.kallsyms]");
  96         else if (machine__is_default_guest(machine))
  97                 machine->mmap_name = strdup("[guest.kernel.kallsyms]");
  98         else if (asprintf(&machine->mmap_name, "[guest.kernel.kallsyms.%d]",
  99                           machine->pid) < 0)
 100                 machine->mmap_name = NULL;
 101
 102         return machine->mmap_name ? 0 : -ENOMEM;
 103 }
 104
 105 static void thread__set_guest_comm(struct thread *thread, pid_t pid)
 106 {
 107         char comm[64];
 108
 109         snprintf(comm, sizeof(comm), "[guest/%d]", pid);
 110         thread__set_comm(thread, comm, 0);
 111 }
 112
 113 int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
 114 {
 115         int err = -ENOMEM;
 116
 117         memset(machine, 0, sizeof(*machine));
 118         machine->kmaps = maps__new(machine);
 119         if (machine->kmaps == NULL)
 120                 return -ENOMEM;
 121
 122         RB_CLEAR_NODE(&machine->rb_node);
 123         dsos__init(&machine->dsos);
 124
 125         machine__threads_init(machine);
 126
 127         machine->vdso_info = NULL;
 128         machine->env = NULL;
 129
 130         machine->pid = pid;
 131
 132         machine->id_hdr_size = 0;
 133         machine->kptr_restrict_warned = false;
 134         machine->comm_exec = false;
 135         machine->kernel_start = 0;
 136         machine->vmlinux_map = NULL;
 137
 138         machine->root_dir = strdup(root_dir);
 139         if (machine->root_dir == NULL)
 140                 goto out;
 141
 142         if (machine__set_mmap_name(machine))
 143                 goto out;
 144
 145         if (pid != HOST_KERNEL_ID) {
 146                 struct thread *thread = machine__findnew_thread(machine, -1,
 147                                                                 pid);
 148
 149                 if (thread == NULL)
 150                         goto out;
 151
 152                 thread__set_guest_comm(thread, pid);
 153                 thread__put(thread);
 154         }
 155
 156         machine->current_tid = NULL;
 157         err = 0;
 158
 159 out:
 160         if (err) {
 161                 zfree(&machine->kmaps);
 162                 zfree(&machine->root_dir);
 163                 zfree(&machine->mmap_name);
 164         }
 165         return 0;
 166 }
 167
 168 struct machine *machine__new_host(void)
 169 {
 170         struct machine *machine = malloc(sizeof(*machine));
 171
 172         if (machine != NULL) {
 173                 machine__init(machine, "", HOST_KERNEL_ID);
 174
 175                 if (machine__create_kernel_maps(machine) < 0)
 176                         goto out_delete;
 177         }
 178
 179         return machine;
 180 out_delete:
 181         free(machine);
 182         return NULL;
 183 }
 184
 185 struct machine *machine__new_kallsyms(void)
 186 {
 187         struct machine *machine = machine__new_host();
 188         /*
 189          * FIXME:
 190          * 1) We should switch to machine__load_kallsyms(), i.e. not explicitly
 191          *    ask for not using the kcore parsing code, once this one is fixed
 192          *    to create a map per module.
 193          */
 194         if (machine && machine__load_kallsyms(machine, "/proc/kallsyms") <= 0) {
 195                 machine__delete(machine);
 196                 machine = NULL;
 197         }
 198
 199         return machine;
 200 }
 201
 202 static void dsos__purge(struct dsos *dsos)
 203 {
 204         struct dso *pos, *n;
 205
 206         down_write(&dsos->lock);
 207
 208         list_for_each_entry_safe(pos, n, &dsos->head, node) {
 209                 RB_CLEAR_NODE(&pos->rb_node);
 210                 pos->root = NULL;
 211                 list_del_init(&pos->node);
 212                 dso__put(pos);
 213         }
 214
 215         up_write(&dsos->lock);
 216 }
 217
 218 static void dsos__exit(struct dsos *dsos)
 219 {
 220         dsos__purge(dsos);
 221         exit_rwsem(&dsos->lock);
 222 }
 223
 224 void machine__delete_threads(struct machine *machine)
 225 {
 226         struct rb_node *nd;
 227         int i;
 228
 229         for (i = 0; i < THREADS__TABLE_SIZE; i++) {
 230                 struct threads *threads = &machine->threads[i];
 231                 down_write(&threads->lock);
 232                 nd = rb_first_cached(&threads->entries);
 233                 while (nd) {
 234                         struct thread_rb_node *trb = rb_entry(nd, struct thread_rb_node, rb_node);
 235
 236                         nd = rb_next(nd);
 237                         __machine__remove_thread(machine, trb, trb->thread, false);
 238                 }
 239                 up_write(&threads->lock);
 240         }
 241 }
 242
 243 void machine__exit(struct machine *machine)
 244 {
 245         int i;
 246
 247         if (machine == NULL)
 248                 return;
 249
 250         machine__destroy_kernel_maps(machine);
 251         maps__zput(machine->kmaps);
 252         dsos__exit(&machine->dsos);
 253         machine__exit_vdso(machine);
 254         zfree(&machine->root_dir);
 255         zfree(&machine->mmap_name);
 256         zfree(&machine->current_tid);
 257         zfree(&machine->kallsyms_filename);
 258
 259         machine__delete_threads(machine);
 260         for (i = 0; i < THREADS__TABLE_SIZE; i++) {
 261                 struct threads *threads = &machine->threads[i];
 262
 263                 exit_rwsem(&threads->lock);
 264         }
 265 }
 266
 267 void machine__delete(struct machine *machine)
 268 {
 269         if (machine) {
 270                 machine__exit(machine);
 271                 free(machine);
 272         }
 273 }
 274
 275 void machines__init(struct machines *machines)
 276 {
 277         machine__init(&machines->host, "", HOST_KERNEL_ID);
 278         machines->guests = RB_ROOT_CACHED;
 279 }
 280
 281 void machines__exit(struct machines *machines)
 282 {
 283         machine__exit(&machines->host);
 284         /* XXX exit guest */
 285 }
 286
 287 struct machine *machines__add(struct machines *machines, pid_t pid,
 288                               const char *root_dir)
 289 {
 290         struct rb_node **p = &machines->guests.rb_root.rb_node;
 291         struct rb_node *parent = NULL;
 292         struct machine *pos, *machine = malloc(sizeof(*machine));
 293         bool leftmost = true;
 294
 295         if (machine == NULL)
 296                 return NULL;
 297
 298         if (machine__init(machine, root_dir, pid) != 0) {
 299                 free(machine);
 300                 return NULL;
 301         }
 302
 303         while (*p != NULL) {
 304                 parent = *p;
 305                 pos = rb_entry(parent, struct machine, rb_node);
 306                 if (pid < pos->pid)
 307                         p = &(*p)->rb_left;
 308                 else {
 309                         p = &(*p)->rb_right;
 310                         leftmost = false;
 311                 }
 312         }
 313
 314         rb_link_node(&machine->rb_node, parent, p);
 315         rb_insert_color_cached(&machine->rb_node, &machines->guests, leftmost);
 316
 317         machine->machines = machines;
 318
 319         return machine;
 320 }
 321
 322 void machines__set_comm_exec(struct machines *machines, bool comm_exec)
 323 {
 324         struct rb_node *nd;
 325
 326         machines->host.comm_exec = comm_exec;
 327
 328         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
 329                 struct machine *machine = rb_entry(nd, struct machine, rb_node);
 330
 331                 machine->comm_exec = comm_exec;
 332         }
 333 }
 334
 335 struct machine *machines__find(struct machines *machines, pid_t pid)
 336 {
 337         struct rb_node **p = &machines->guests.rb_root.rb_node;
 338         struct rb_node *parent = NULL;
 339         struct machine *machine;
 340         struct machine *default_machine = NULL;
 341
 342         if (pid == HOST_KERNEL_ID)
 343                 return &machines->host;
 344
 345         while (*p != NULL) {
 346                 parent = *p;
 347                 machine = rb_entry(parent, struct machine, rb_node);
 348                 if (pid < machine->pid)
 349                         p = &(*p)->rb_left;
 350                 else if (pid > machine->pid)
 351                         p = &(*p)->rb_right;
 352                 else
 353                         return machine;
 354                 if (!machine->pid)
 355                         default_machine = machine;
 356         }
 357
 358         return default_machine;
 359 }
 360
 361 struct machine *machines__findnew(struct machines *machines, pid_t pid)
 362 {
 363         char path[PATH_MAX];
 364         const char *root_dir = "";
 365         struct machine *machine = machines__find(machines, pid);
 366
 367         if (machine && (machine->pid == pid))
 368                 goto out;
 369
 370         if ((pid != HOST_KERNEL_ID) &&
 371             (pid != DEFAULT_GUEST_KERNEL_ID) &&
 372             (symbol_conf.guestmount)) {
 373                 sprintf(path, "%s/%d", symbol_conf.guestmount, pid);
 374                 if (access(path, R_OK)) {
 375                         static struct strlist *seen;
 376
 377                         if (!seen)
 378                                 seen = strlist__new(NULL, NULL);
 379
 380                         if (!strlist__has_entry(seen, path)) {
 381                                 pr_err("Can't access file %s\n", path);
 382                                 strlist__add(seen, path);
 383                         }
 384                         machine = NULL;
 385                         goto out;
 386                 }
 387                 root_dir = path;
 388         }
 389
 390         machine = machines__add(machines, pid, root_dir);
 391 out:
 392         return machine;
 393 }
 394
 395 struct machine *machines__find_guest(struct machines *machines, pid_t pid)
 396 {
 397         struct machine *machine = machines__find(machines, pid);
 398
 399         if (!machine)
 400                 machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
 401         return machine;
 402 }
 403
 404 /*
 405  * A common case for KVM test programs is that the test program acts as the
 406  * hypervisor, creating, running and destroying the virtual machine, and
 407  * providing the guest object code from its own object code. In this case,
 408  * the VM is not running an OS, but only the functions loaded into it by the
 409  * hypervisor test program, and conveniently, loaded at the same virtual
 410  * addresses.
 411  *
 412  * Normally to resolve addresses, MMAP events are needed to map addresses
 413  * back to the object code and debug symbols for that object code.
 414  *
 415  * Currently, there is no way to get such mapping information from guests
 416  * but, in the scenario described above, the guest has the same mappings
 417  * as the hypervisor, so support for that scenario can be achieved.
 418  *
 419  * To support that, copy the host thread's maps to the guest thread's maps.
 420  * Note, we do not discover the guest until we encounter a guest event,
 421  * which works well because it is not until then that we know that the host
 422  * thread's maps have been set up.
 423  *
 424  * This function returns the guest thread. Apart from keeping the data
 425  * structures sane, using a thread belonging to the guest machine, instead
 426  * of the host thread, allows it to have its own comm (refer
 427  * thread__set_guest_comm()).
 428  */
 429 static struct thread *findnew_guest_code(struct machine *machine,
 430                                          struct machine *host_machine,
 431                                          pid_t pid)
 432 {
 433         struct thread *host_thread;
 434         struct thread *thread;
 435         int err;
 436
 437         if (!machine)
 438                 return NULL;
 439
 440         thread = machine__findnew_thread(machine, -1, pid);
 441         if (!thread)
 442                 return NULL;
 443
 444         /* Assume maps are set up if there are any */
 445         if (maps__nr_maps(thread__maps(thread)))
 446                 return thread;
 447
 448         host_thread = machine__find_thread(host_machine, -1, pid);
 449         if (!host_thread)
 450                 goto out_err;
 451
 452         thread__set_guest_comm(thread, pid);
 453
 454         /*
 455          * Guest code can be found in hypervisor process at the same address
 456          * so copy host maps.
 457          */
 458         err = maps__clone(thread, thread__maps(host_thread));
 459         thread__put(host_thread);
 460         if (err)
 461                 goto out_err;
 462
 463         return thread;
 464
 465 out_err:
 466         thread__zput(thread);
 467         return NULL;
 468 }
 469
 470 struct thread *machines__findnew_guest_code(struct machines *machines, pid_t pid)
 471 {
 472         struct machine *host_machine = machines__find(machines, HOST_KERNEL_ID);
 473         struct machine *machine = machines__findnew(machines, pid);
 474
 475         return findnew_guest_code(machine, host_machine, pid);
 476 }
 477
 478 struct thread *machine__findnew_guest_code(struct machine *machine, pid_t pid)
 479 {
 480         struct machines *machines = machine->machines;
 481         struct machine *host_machine;
 482
 483         if (!machines)
 484                 return NULL;
 485
 486         host_machine = machines__find(machines, HOST_KERNEL_ID);
 487
 488         return findnew_guest_code(machine, host_machine, pid);
 489 }
 490
 491 void machines__process_guests(struct machines *machines,
 492                               machine__process_t process, void *data)
 493 {
 494         struct rb_node *nd;
 495
 496         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
 497                 struct machine *pos = rb_entry(nd, struct machine, rb_node);
 498                 process(pos, data);
 499         }
 500 }
 501
 502 void machines__set_id_hdr_size(struct machines *machines, u16 id_hdr_size)
 503 {
 504         struct rb_node *node;
 505         struct machine *machine;
 506
 507         machines->host.id_hdr_size = id_hdr_size;
 508
 509         for (node = rb_first_cached(&machines->guests); node;
 510              node = rb_next(node)) {
 511                 machine = rb_entry(node, struct machine, rb_node);
 512                 machine->id_hdr_size = id_hdr_size;
 513         }
 514
 515         return;
 516 }
 517
 518 static void machine__update_thread_pid(struct machine *machine,
 519                                        struct thread *th, pid_t pid)
 520 {
 521         struct thread *leader;
 522
 523         if (pid == thread__pid(th) || pid == -1 || thread__pid(th) != -1)
 524                 return;
 525
 526         thread__set_pid(th, pid);
 527
 528         if (thread__pid(th) == thread__tid(th))
 529                 return;
 530
 531         leader = __machine__findnew_thread(machine, thread__pid(th), thread__pid(th));
 532         if (!leader)
 533                 goto out_err;
 534
 535         if (!thread__maps(leader))
 536                 thread__set_maps(leader, maps__new(machine));
 537
 538         if (!thread__maps(leader))
 539                 goto out_err;
 540
 541         if (thread__maps(th) == thread__maps(leader))
 542                 goto out_put;
 543
 544         if (thread__maps(th)) {
 545                 /*
 546                  * Maps are created from MMAP events which provide the pid and
 547                  * tid.  Consequently there never should be any maps on a thread
 548                  * with an unknown pid.  Just print an error if there are.
 549                  */
 550                 if (!maps__empty(thread__maps(th)))
 551                         pr_err("Discarding thread maps for %d:%d\n",
 552                                 thread__pid(th), thread__tid(th));
 553                 maps__put(thread__maps(th));
 554         }
 555
 556         thread__set_maps(th, maps__get(thread__maps(leader)));
 557 out_put:
 558         thread__put(leader);
 559         return;
 560 out_err:
 561         pr_err("Failed to join map groups for %d:%d\n", thread__pid(th), thread__tid(th));
 562         goto out_put;
 563 }
 564
 565 /*
 566  * Front-end cache - TID lookups come in blocks,
 567  * so most of the time we dont have to look up
 568  * the full rbtree:
 569  */
 570 static struct thread*
 571 __threads__get_last_match(struct threads *threads, struct machine *machine,
 572                           int pid, int tid)
 573 {
 574         struct thread *th;
 575
 576         th = threads->last_match;
 577         if (th != NULL) {
 578                 if (thread__tid(th) == tid) {
 579                         machine__update_thread_pid(machine, th, pid);
 580                         return thread__get(th);
 581                 }
 582                 thread__put(threads->last_match);
 583                 threads->last_match = NULL;
 584         }
 585
 586         return NULL;
 587 }
 588
 589 static struct thread*
 590 threads__get_last_match(struct threads *threads, struct machine *machine,
 591                         int pid, int tid)
 592 {
 593         struct thread *th = NULL;
 594
 595         if (perf_singlethreaded)
 596                 th = __threads__get_last_match(threads, machine, pid, tid);
 597
 598         return th;
 599 }
 600
 601 static void
 602 __threads__set_last_match(struct threads *threads, struct thread *th)
 603 {
 604         thread__put(threads->last_match);
 605         threads->last_match = thread__get(th);
 606 }
 607
 608 static void
 609 threads__set_last_match(struct threads *threads, struct thread *th)
 610 {
 611         if (perf_singlethreaded)
 612                 __threads__set_last_match(threads, th);
 613 }
 614
 615 /*
 616  * Caller must eventually drop thread->refcnt returned with a successful
 617  * lookup/new thread inserted.
 618  */
 619 static struct thread *____machine__findnew_thread(struct machine *machine,
 620                                                   struct threads *threads,
 621                                                   pid_t pid, pid_t tid,
 622                                                   bool create)
 623 {
 624         struct rb_node **p = &threads->entries.rb_root.rb_node;
 625         struct rb_node *parent = NULL;
 626         struct thread *th;
 627         struct thread_rb_node *nd;
 628         bool leftmost = true;
 629
 630         th = threads__get_last_match(threads, machine, pid, tid);
 631         if (th)
 632                 return th;
 633
 634         while (*p != NULL) {
 635                 parent = *p;
 636                 th = rb_entry(parent, struct thread_rb_node, rb_node)->thread;
 637
 638                 if (thread__tid(th) == tid) {
 639                         threads__set_last_match(threads, th);
 640                         machine__update_thread_pid(machine, th, pid);
 641                         return thread__get(th);
 642                 }
 643
 644                 if (tid < thread__tid(th))
 645                         p = &(*p)->rb_left;
 646                 else {
 647                         p = &(*p)->rb_right;
 648                         leftmost = false;
 649                 }
 650         }
 651
 652         if (!create)
 653                 return NULL;
 654
 655         th = thread__new(pid, tid);
 656         if (th == NULL)
 657                 return NULL;
 658
 659         nd = malloc(sizeof(*nd));
 660         if (nd == NULL) {
 661                 thread__put(th);
 662                 return NULL;
 663         }
 664         nd->thread = th;
 665
 666         rb_link_node(&nd->rb_node, parent, p);
 667         rb_insert_color_cached(&nd->rb_node, &threads->entries, leftmost);
 668         /*
 669          * We have to initialize maps separately after rb tree is updated.
 670          *
 671          * The reason is that we call machine__findnew_thread within
 672          * thread__init_maps to find the thread leader and that would screwed
 673          * the rb tree.
 674          */
 675         if (thread__init_maps(th, machine)) {
 676                 pr_err("Thread init failed thread %d\n", pid);
 677                 rb_erase_cached(&nd->rb_node, &threads->entries);
 678                 RB_CLEAR_NODE(&nd->rb_node);
 679                 free(nd);
 680                 thread__put(th);
 681                 return NULL;
 682         }
 683         /*
 684          * It is now in the rbtree, get a ref
 685          */
 686         threads__set_last_match(threads, th);
 687         ++threads->nr;
 688
 689         return thread__get(th);
 690 }
 691
 692 struct thread *__machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid)
 693 {
 694         return ____machine__findnew_thread(machine, machine__threads(machine, tid), pid, tid, true);
 695 }
 696
 697 struct thread *machine__findnew_thread(struct machine *machine, pid_t pid,
 698                                        pid_t tid)
 699 {
 700         struct threads *threads = machine__threads(machine, tid);
 701         struct thread *th;
 702
 703         down_write(&threads->lock);
 704         th = __machine__findnew_thread(machine, pid, tid);
 705         up_write(&threads->lock);
 706         return th;
 707 }
 708
 709 struct thread *machine__find_thread(struct machine *machine, pid_t pid,
 710                                     pid_t tid)
 711 {
 712         struct threads *threads = machine__threads(machine, tid);
 713         struct thread *th;
 714
 715         down_read(&threads->lock);
 716         th =  ____machine__findnew_thread(machine, threads, pid, tid, false);
 717         up_read(&threads->lock);
 718         return th;
 719 }
 720
 721 /*
 722  * Threads are identified by pid and tid, and the idle task has pid == tid == 0.
 723  * So here a single thread is created for that, but actually there is a separate
 724  * idle task per cpu, so there should be one 'struct thread' per cpu, but there
 725  * is only 1. That causes problems for some tools, requiring workarounds. For
 726  * example get_idle_thread() in builtin-sched.c, or thread_stack__per_cpu().
 727  */
 728 struct thread *machine__idle_thread(struct machine *machine)
 729 {
 730         struct thread *thread = machine__findnew_thread(machine, 0, 0);
 731
 732         if (!thread || thread__set_comm(thread, "swapper", 0) ||
 733             thread__set_namespaces(thread, 0, NULL))
 734                 pr_err("problem inserting idle task for machine pid %d\n", machine->pid);
 735
 736         return thread;
 737 }
 738
 739 struct comm *machine__thread_exec_comm(struct machine *machine,
 740                                        struct thread *thread)
 741 {
 742         if (machine->comm_exec)
 743                 return thread__exec_comm(thread);
 744         else
 745                 return thread__comm(thread);
 746 }
 747
 748 int machine__process_comm_event(struct machine *machine, union perf_event *event,
 749                                 struct perf_sample *sample)
 750 {
 751         struct thread *thread = machine__findnew_thread(machine,
 752                                                         event->comm.pid,
 753                                                         event->comm.tid);
 754         bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
 755         int err = 0;
 756
 757         if (exec)
 758                 machine->comm_exec = true;
 759
 760         if (dump_trace)
 761                 perf_event__fprintf_comm(event, stdout);
 762
 763         if (thread == NULL ||
 764             __thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
 765                 dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
 766                 err = -1;
 767         }
 768
 769         thread__put(thread);
 770
 771         return err;
 772 }
 773
 774 int machine__process_namespaces_event(struct machine *machine __maybe_unused,
 775                                       union perf_event *event,
 776                                       struct perf_sample *sample __maybe_unused)
 777 {
 778         struct thread *thread = machine__findnew_thread(machine,
 779                                                         event->namespaces.pid,
 780                                                         event->namespaces.tid);
 781         int err = 0;
 782
 783         WARN_ONCE(event->namespaces.nr_namespaces > NR_NAMESPACES,
 784                   "\nWARNING: kernel seems to support more namespaces than perf"
 785                   " tool.\nTry updating the perf tool..\n\n");
 786
 787         WARN_ONCE(event->namespaces.nr_namespaces < NR_NAMESPACES,
 788                   "\nWARNING: perf tool seems to support more namespaces than"
 789                   " the kernel.\nTry updating the kernel..\n\n");
 790
 791         if (dump_trace)
 792                 perf_event__fprintf_namespaces(event, stdout);
 793
 794         if (thread == NULL ||
 795             thread__set_namespaces(thread, sample->time, &event->namespaces)) {
 796                 dump_printf("problem processing PERF_RECORD_NAMESPACES, skipping event.\n");
 797                 err = -1;
 798         }
 799
 800         thread__put(thread);
 801
 802         return err;
 803 }
 804
 805 int machine__process_cgroup_event(struct machine *machine,
 806                                   union perf_event *event,
 807                                   struct perf_sample *sample __maybe_unused)
 808 {
 809         struct cgroup *cgrp;
 810
 811         if (dump_trace)
 812                 perf_event__fprintf_cgroup(event, stdout);
 813
 814         cgrp = cgroup__findnew(machine->env, event->cgroup.id, event->cgroup.path);
 815         if (cgrp == NULL)
 816                 return -ENOMEM;
 817
 818         return 0;
 819 }
 820
 821 int machine__process_lost_event(struct machine *machine __maybe_unused,
 822                                 union perf_event *event, struct perf_sample *sample __maybe_unused)
 823 {
 824         dump_printf(": id:%" PRI_lu64 ": lost:%" PRI_lu64 "\n",
 825                     event->lost.id, event->lost.lost);
 826         return 0;
 827 }
 828
 829 int machine__process_lost_samples_event(struct machine *machine __maybe_unused,
 830                                         union perf_event *event, struct perf_sample *sample)
 831 {
 832         dump_printf(": id:%" PRIu64 ": lost samples :%" PRI_lu64 "\n",
 833                     sample->id, event->lost_samples.lost);
 834         return 0;
 835 }
 836
 837 static struct dso *machine__findnew_module_dso(struct machine *machine,
 838                                                struct kmod_path *m,
 839                                                const char *filename)
 840 {
 841         struct dso *dso;
 842
 843         down_write(&machine->dsos.lock);
 844
 845         dso = __dsos__find(&machine->dsos, m->name, true);
 846         if (!dso) {
 847                 dso = __dsos__addnew(&machine->dsos, m->name);
 848                 if (dso == NULL)
 849                         goto out_unlock;
 850
 851                 dso__set_module_info(dso, m, machine);
 852                 dso__set_long_name(dso, strdup(filename), true);
 853                 dso->kernel = DSO_SPACE__KERNEL;
 854         }
 855
 856         dso__get(dso);
 857 out_unlock:
 858         up_write(&machine->dsos.lock);
 859         return dso;
 860 }
 861
 862 int machine__process_aux_event(struct machine *machine __maybe_unused,
 863                                union perf_event *event)
 864 {
 865         if (dump_trace)
 866                 perf_event__fprintf_aux(event, stdout);
 867         return 0;
 868 }
 869
 870 int machine__process_itrace_start_event(struct machine *machine __maybe_unused,
 871                                         union perf_event *event)
 872 {
 873         if (dump_trace)
 874                 perf_event__fprintf_itrace_start(event, stdout);
 875         return 0;
 876 }
 877
 878 int machine__process_aux_output_hw_id_event(struct machine *machine __maybe_unused,
 879                                             union perf_event *event)
 880 {
 881         if (dump_trace)
 882                 perf_event__fprintf_aux_output_hw_id(event, stdout);
 883         return 0;
 884 }
 885
 886 int machine__process_switch_event(struct machine *machine __maybe_unused,
 887                                   union perf_event *event)
 888 {
 889         if (dump_trace)
 890                 perf_event__fprintf_switch(event, stdout);
 891         return 0;
 892 }
 893
 894 static int machine__process_ksymbol_register(struct machine *machine,
 895                                              union perf_event *event,
 896                                              struct perf_sample *sample __maybe_unused)
 897 {
 898         struct symbol *sym;
 899         struct dso *dso;
 900         struct map *map = maps__find(machine__kernel_maps(machine), event->ksymbol.addr);
 901         bool put_map = false;
 902         int err = 0;
 903
 904         if (!map) {
 905                 dso = dso__new(event->ksymbol.name);
 906
 907                 if (!dso) {
 908                         err = -ENOMEM;
 909                         goto out;
 910                 }
 911                 dso->kernel = DSO_SPACE__KERNEL;
 912                 map = map__new2(0, dso);
 913                 dso__put(dso);
 914                 if (!map) {
 915                         err = -ENOMEM;
 916                         goto out;
 917                 }
 918                 /*
 919                  * The inserted map has a get on it, we need to put to release
 920                  * the reference count here, but do it after all accesses are
 921                  * done.
 922                  */
 923                 put_map = true;
 924                 if (event->ksymbol.ksym_type == PERF_RECORD_KSYMBOL_TYPE_OOL) {
 925                         dso->binary_type = DSO_BINARY_TYPE__OOL;
 926                         dso->data.file_size = event->ksymbol.len;
 927                         dso__set_loaded(dso);
 928                 }
 929
 930                 map__set_start(map, event->ksymbol.addr);
 931                 map__set_end(map, map__start(map) + event->ksymbol.len);
 932                 err = maps__insert(machine__kernel_maps(machine), map);
 933                 if (err) {
 934                         err = -ENOMEM;
 935                         goto out;
 936                 }
 937
 938                 dso__set_loaded(dso);
 939
 940                 if (is_bpf_image(event->ksymbol.name)) {
 941                         dso->binary_type = DSO_BINARY_TYPE__BPF_IMAGE;
 942                         dso__set_long_name(dso, "", false);
 943                 }
 944         } else {
 945                 dso = map__dso(map);
 946         }
 947
 948         sym = symbol__new(map__map_ip(map, map__start(map)),
 949                           event->ksymbol.len,
 950                           0, 0, event->ksymbol.name);
 951         if (!sym) {
 952                 err = -ENOMEM;
 953                 goto out;
 954         }
 955         dso__insert_symbol(dso, sym);
 956 out:
 957         if (put_map)
 958                 map__put(map);
 959         return err;
 960 }
 961
 962 static int machine__process_ksymbol_unregister(struct machine *machine,
 963                                                union perf_event *event,
 964                                                struct perf_sample *sample __maybe_unused)
 965 {
 966         struct symbol *sym;
 967         struct map *map;
 968
 969         map = maps__find(machine__kernel_maps(machine), event->ksymbol.addr);
 970         if (!map)
 971                 return 0;
 972
 973         if (RC_CHK_ACCESS(map) != RC_CHK_ACCESS(machine->vmlinux_map))
 974                 maps__remove(machine__kernel_maps(machine), map);
 975         else {
 976                 struct dso *dso = map__dso(map);
 977
 978                 sym = dso__find_symbol(dso, map__map_ip(map, map__start(map)));
 979                 if (sym)
 980                         dso__delete_symbol(dso, sym);
 981         }
 982
 983         return 0;
 984 }
 985
 986 int machine__process_ksymbol(struct machine *machine __maybe_unused,
 987                              union perf_event *event,
 988                              struct perf_sample *sample)
 989 {
 990         if (dump_trace)
 991                 perf_event__fprintf_ksymbol(event, stdout);
 992
 993         if (event->ksymbol.flags & PERF_RECORD_KSYMBOL_FLAGS_UNREGISTER)
 994                 return machine__process_ksymbol_unregister(machine, event,
 995                                                            sample);
 996         return machine__process_ksymbol_register(machine, event, sample);
 997 }
 998
 999 int machine__process_text_poke(struct machine *machine, union perf_event *event,
1000                                struct perf_sample *sample __maybe_unused)
1001 {
1002         struct map *map = maps__find(machine__kernel_maps(machine), event->text_poke.addr);
1003         u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
1004         struct dso *dso = map ? map__dso(map) : NULL;
1005
1006         if (dump_trace)
1007                 perf_event__fprintf_text_poke(event, machine, stdout);
1008
1009         if (!event->text_poke.new_len)
1010                 return 0;
1011
1012         if (cpumode != PERF_RECORD_MISC_KERNEL) {
1013                 pr_debug("%s: unsupported cpumode - ignoring\n", __func__);
1014                 return 0;
1015         }
1016
1017         if (dso) {
1018                 u8 *new_bytes = event->text_poke.bytes + event->text_poke.old_len;
1019                 int ret;
1020
1021                 /*
1022                  * Kernel maps might be changed when loading symbols so loading
1023                  * must be done prior to using kernel maps.
1024                  */
1025                 map__load(map);
1026                 ret = dso__data_write_cache_addr(dso, map, machine,
1027                                                  event->text_poke.addr,
1028                                                  new_bytes,
1029                                                  event->text_poke.new_len);
1030                 if (ret != event->text_poke.new_len)
1031                         pr_debug("Failed to write kernel text poke at %#" PRI_lx64 "\n",
1032                                  event->text_poke.addr);
1033         } else {
1034                 pr_debug("Failed to find kernel text poke address map for %#" PRI_lx64 "\n",
1035                          event->text_poke.addr);
1036         }
1037
1038         return 0;
1039 }
1040
1041 static struct map *machine__addnew_module_map(struct machine *machine, u64 start,
1042                                               const char *filename)
1043 {
1044         struct map *map = NULL;
1045         struct kmod_path m;
1046         struct dso *dso;
1047         int err;
1048
1049         if (kmod_path__parse_name(&m, filename))
1050                 return NULL;
1051
1052         dso = machine__findnew_module_dso(machine, &m, filename);
1053         if (dso == NULL)
1054                 goto out;
1055
1056         map = map__new2(start, dso);
1057         if (map == NULL)
1058                 goto out;
1059
1060         err = maps__insert(machine__kernel_maps(machine), map);
1061         /* If maps__insert failed, return NULL. */
1062         if (err) {
1063                 map__put(map);
1064                 map = NULL;
1065         }
1066 out:
1067         /* put the dso here, corresponding to  machine__findnew_module_dso */
1068         dso__put(dso);
1069         zfree(&m.name);
1070         return map;
1071 }
1072
1073 size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
1074 {
1075         struct rb_node *nd;
1076         size_t ret = __dsos__fprintf(&machines->host.dsos.head, fp);
1077
1078         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
1079                 struct machine *pos = rb_entry(nd, struct machine, rb_node);
1080                 ret += __dsos__fprintf(&pos->dsos.head, fp);
1081         }
1082
1083         return ret;
1084 }
1085
1086 size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
1087                                      bool (skip)(struct dso *dso, int parm), int parm)
1088 {
1089         return __dsos__fprintf_buildid(&m->dsos.head, fp, skip, parm);
1090 }
1091
1092 size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
1093                                      bool (skip)(struct dso *dso, int parm), int parm)
1094 {
1095         struct rb_node *nd;
1096         size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
1097
1098         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
1099                 struct machine *pos = rb_entry(nd, struct machine, rb_node);
1100                 ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
1101         }
1102         return ret;
1103 }
1104
1105 size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
1106 {
1107         int i;
1108         size_t printed = 0;
1109         struct dso *kdso = machine__kernel_dso(machine);
1110
1111         if (kdso->has_build_id) {
1112                 char filename[PATH_MAX];
1113                 if (dso__build_id_filename(kdso, filename, sizeof(filename),
1114                                            false))
1115                         printed += fprintf(fp, "[0] %s\n", filename);
1116         }
1117
1118         for (i = 0; i < vmlinux_path__nr_entries; ++i)
1119                 printed += fprintf(fp, "[%d] %s\n",
1120                                    i + kdso->has_build_id, vmlinux_path[i]);
1121
1122         return printed;
1123 }
1124
1125 size_t machine__fprintf(struct machine *machine, FILE *fp)
1126 {
1127         struct rb_node *nd;
1128         size_t ret;
1129         int i;
1130
1131         for (i = 0; i < THREADS__TABLE_SIZE; i++) {
1132                 struct threads *threads = &machine->threads[i];
1133
1134                 down_read(&threads->lock);
1135
1136                 ret = fprintf(fp, "Threads: %u\n", threads->nr);
1137
1138                 for (nd = rb_first_cached(&threads->entries); nd;
1139                      nd = rb_next(nd)) {
1140                         struct thread *pos = rb_entry(nd, struct thread_rb_node, rb_node)->thread;
1141
1142                         ret += thread__fprintf(pos, fp);
1143                 }
1144
1145                 up_read(&threads->lock);
1146         }
1147         return ret;
1148 }
1149
1150 static struct dso *machine__get_kernel(struct machine *machine)
1151 {
1152         const char *vmlinux_name = machine->mmap_name;
1153         struct dso *kernel;
1154
1155         if (machine__is_host(machine)) {
1156                 if (symbol_conf.vmlinux_name)
1157                         vmlinux_name = symbol_conf.vmlinux_name;
1158
1159                 kernel = machine__findnew_kernel(machine, vmlinux_name,
1160                                                  "[kernel]", DSO_SPACE__KERNEL);
1161         } else {
1162                 if (symbol_conf.default_guest_vmlinux_name)
1163                         vmlinux_name = symbol_conf.default_guest_vmlinux_name;
1164
1165                 kernel = machine__findnew_kernel(machine, vmlinux_name,
1166                                                  "[guest.kernel]",
1167                                                  DSO_SPACE__KERNEL_GUEST);
1168         }
1169
1170         if (kernel != NULL && (!kernel->has_build_id))
1171                 dso__read_running_kernel_build_id(kernel, machine);
1172
1173         return kernel;
1174 }
1175
1176 void machine__get_kallsyms_filename(struct machine *machine, char *buf,
1177                                     size_t bufsz)
1178 {
1179         if (machine__is_default_guest(machine))
1180                 scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
1181         else
1182                 scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
1183 }
1184
1185 const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
1186
1187 /* Figure out the start address of kernel map from /proc/kallsyms.
1188  * Returns the name of the start symbol in *symbol_name. Pass in NULL as
1189  * symbol_name if it's not that important.
1190  */
1191 static int machine__get_running_kernel_start(struct machine *machine,
1192                                              const char **symbol_name,
1193                                              u64 *start, u64 *end)
1194 {
1195         char filename[PATH_MAX];
1196         int i, err = -1;
1197         const char *name;
1198         u64 addr = 0;
1199
1200         machine__get_kallsyms_filename(machine, filename, PATH_MAX);
1201
1202         if (symbol__restricted_filename(filename, "/proc/kallsyms"))
1203                 return 0;
1204
1205         for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
1206                 err = kallsyms__get_function_start(filename, name, &addr);
1207                 if (!err)
1208                         break;
1209         }
1210
1211         if (err)
1212                 return -1;
1213
1214         if (symbol_name)
1215                 *symbol_name = name;
1216
1217         *start = addr;
1218
1219         err = kallsyms__get_function_start(filename, "_etext", &addr);
1220         if (!err)
1221                 *end = addr;
1222
1223         return 0;
1224 }
1225
1226 int machine__create_extra_kernel_map(struct machine *machine,
1227                                      struct dso *kernel,
1228                                      struct extra_kernel_map *xm)
1229 {
1230         struct kmap *kmap;
1231         struct map *map;
1232         int err;
1233
1234         map = map__new2(xm->start, kernel);
1235         if (!map)
1236                 return -ENOMEM;
1237
1238         map__set_end(map, xm->end);
1239         map__set_pgoff(map, xm->pgoff);
1240
1241         kmap = map__kmap(map);
1242
1243         strlcpy(kmap->name, xm->name, KMAP_NAME_LEN);
1244
1245         err = maps__insert(machine__kernel_maps(machine), map);
1246
1247         if (!err) {
1248                 pr_debug2("Added extra kernel map %s %" PRIx64 "-%" PRIx64 "\n",
1249                         kmap->name, map__start(map), map__end(map));
1250         }
1251
1252         map__put(map);
1253
1254         return err;
1255 }
1256
1257 static u64 find_entry_trampoline(struct dso *dso)
1258 {
1259         /* Duplicates are removed so lookup all aliases */
1260         const char *syms[] = {
1261                 "_entry_trampoline",
1262                 "__entry_trampoline_start",
1263                 "entry_SYSCALL_64_trampoline",
1264         };
1265         struct symbol *sym = dso__first_symbol(dso);
1266         unsigned int i;
1267
1268         for (; sym; sym = dso__next_symbol(sym)) {
1269                 if (sym->binding != STB_GLOBAL)
1270                         continue;
1271                 for (i = 0; i < ARRAY_SIZE(syms); i++) {
1272                         if (!strcmp(sym->name, syms[i]))
1273                                 return sym->start;
1274                 }
1275         }
1276
1277         return 0;
1278 }
1279
1280 /*
1281  * These values can be used for kernels that do not have symbols for the entry
1282  * trampolines in kallsyms.
1283  */
1284 #define X86_64_CPU_ENTRY_AREA_PER_CPU   0xfffffe0000000000ULL
1285 #define X86_64_CPU_ENTRY_AREA_SIZE      0x2c000
1286 #define X86_64_ENTRY_TRAMPOLINE         0x6000
1287
1288 /* Map x86_64 PTI entry trampolines */
1289 int machine__map_x86_64_entry_trampolines(struct machine *machine,
1290                                           struct dso *kernel)
1291 {
1292         struct maps *kmaps = machine__kernel_maps(machine);
1293         int nr_cpus_avail, cpu;
1294         bool found = false;
1295         struct map_rb_node *rb_node;
1296         u64 pgoff;
1297
1298         /*
1299          * In the vmlinux case, pgoff is a virtual address which must now be
1300          * mapped to a vmlinux offset.
1301          */
1302         maps__for_each_entry(kmaps, rb_node) {
1303                 struct map *dest_map, *map = rb_node->map;
1304                 struct kmap *kmap = __map__kmap(map);
1305
1306                 if (!kmap || !is_entry_trampoline(kmap->name))
1307                         continue;
1308
1309                 dest_map = maps__find(kmaps, map__pgoff(map));
1310                 if (dest_map != map)
1311                         map__set_pgoff(map, map__map_ip(dest_map, map__pgoff(map)));
1312                 found = true;
1313         }
1314         if (found || machine->trampolines_mapped)
1315                 return 0;
1316
1317         pgoff = find_entry_trampoline(kernel);
1318         if (!pgoff)
1319                 return 0;
1320
1321         nr_cpus_avail = machine__nr_cpus_avail(machine);
1322
1323         /* Add a 1 page map for each CPU's entry trampoline */
1324         for (cpu = 0; cpu < nr_cpus_avail; cpu++) {
1325                 u64 va = X86_64_CPU_ENTRY_AREA_PER_CPU +
1326                          cpu * X86_64_CPU_ENTRY_AREA_SIZE +
1327                          X86_64_ENTRY_TRAMPOLINE;
1328                 struct extra_kernel_map xm = {
1329                         .start = va,
1330                         .end   = va + page_size,
1331                         .pgoff = pgoff,
1332                 };
1333
1334                 strlcpy(xm.name, ENTRY_TRAMPOLINE_NAME, KMAP_NAME_LEN);
1335
1336                 if (machine__create_extra_kernel_map(machine, kernel, &xm) < 0)
1337                         return -1;
1338         }
1339
1340         machine->trampolines_mapped = nr_cpus_avail;
1341
1342         return 0;
1343 }
1344
1345 int __weak machine__create_extra_kernel_maps(struct machine *machine __maybe_unused,
1346                                              struct dso *kernel __maybe_unused)
1347 {
1348         return 0;
1349 }
1350
1351 static int
1352 __machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
1353 {
1354         /* In case of renewal the kernel map, destroy previous one */
1355         machine__destroy_kernel_maps(machine);
1356
1357         map__put(machine->vmlinux_map);
1358         machine->vmlinux_map = map__new2(0, kernel);
1359         if (machine->vmlinux_map == NULL)
1360                 return -ENOMEM;
1361
1362         map__set_map_ip(machine->vmlinux_map, identity__map_ip);
1363         map__set_unmap_ip(machine->vmlinux_map, identity__map_ip);
1364         return maps__insert(machine__kernel_maps(machine), machine->vmlinux_map);
1365 }
1366
1367 void machine__destroy_kernel_maps(struct machine *machine)
1368 {
1369         struct kmap *kmap;
1370         struct map *map = machine__kernel_map(machine);
1371
1372         if (map == NULL)
1373                 return;
1374
1375         kmap = map__kmap(map);
1376         maps__remove(machine__kernel_maps(machine), map);
1377         if (kmap && kmap->ref_reloc_sym) {
1378                 zfree((char **)&kmap->ref_reloc_sym->name);
1379                 zfree(&kmap->ref_reloc_sym);
1380         }
1381
1382         map__zput(machine->vmlinux_map);
1383 }
1384
1385 int machines__create_guest_kernel_maps(struct machines *machines)
1386 {
1387         int ret = 0;
1388         struct dirent **namelist = NULL;
1389         int i, items = 0;
1390         char path[PATH_MAX];
1391         pid_t pid;
1392         char *endp;
1393
1394         if (symbol_conf.default_guest_vmlinux_name ||
1395             symbol_conf.default_guest_modules ||
1396             symbol_conf.default_guest_kallsyms) {
1397                 machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
1398         }
1399
1400         if (symbol_conf.guestmount) {
1401                 items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
1402                 if (items <= 0)
1403                         return -ENOENT;
1404                 for (i = 0; i < items; i++) {
1405                         if (!isdigit(namelist[i]->d_name[0])) {
1406                                 /* Filter out . and .. */
1407                                 continue;
1408                         }
1409                         pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
1410                         if ((*endp != '\0') ||
1411                             (endp == namelist[i]->d_name) ||
1412                             (errno == ERANGE)) {
1413                                 pr_debug("invalid directory (%s). Skipping.\n",
1414                                          namelist[i]->d_name);
1415                                 continue;
1416                         }
1417                         sprintf(path, "%s/%s/proc/kallsyms",
1418                                 symbol_conf.guestmount,
1419                                 namelist[i]->d_name);
1420                         ret = access(path, R_OK);
1421                         if (ret) {
1422                                 pr_debug("Can't access file %s\n", path);
1423                                 goto failure;
1424                         }
1425                         machines__create_kernel_maps(machines, pid);
1426                 }
1427 failure:
1428                 free(namelist);
1429         }
1430
1431         return ret;
1432 }
1433
1434 void machines__destroy_kernel_maps(struct machines *machines)
1435 {
1436         struct rb_node *next = rb_first_cached(&machines->guests);
1437
1438         machine__destroy_kernel_maps(&machines->host);
1439
1440         while (next) {
1441                 struct machine *pos = rb_entry(next, struct machine, rb_node);
1442
1443                 next = rb_next(&pos->rb_node);
1444                 rb_erase_cached(&pos->rb_node, &machines->guests);
1445                 machine__delete(pos);
1446         }
1447 }
1448
1449 int machines__create_kernel_maps(struct machines *machines, pid_t pid)
1450 {
1451         struct machine *machine = machines__findnew(machines, pid);
1452
1453         if (machine == NULL)
1454                 return -1;
1455
1456         return machine__create_kernel_maps(machine);
1457 }
1458
1459 int machine__load_kallsyms(struct machine *machine, const char *filename)
1460 {
1461         struct map *map = machine__kernel_map(machine);
1462         struct dso *dso = map__dso(map);
1463         int ret = __dso__load_kallsyms(dso, filename, map, true);
1464
1465         if (ret > 0) {
1466                 dso__set_loaded(dso);
1467                 /*
1468                  * Since /proc/kallsyms will have multiple sessions for the
1469                  * kernel, with modules between them, fixup the end of all
1470                  * sections.
1471                  */
1472                 maps__fixup_end(machine__kernel_maps(machine));
1473         }
1474
1475         return ret;
1476 }
1477
1478 int machine__load_vmlinux_path(struct machine *machine)
1479 {
1480         struct map *map = machine__kernel_map(machine);
1481         struct dso *dso = map__dso(map);
1482         int ret = dso__load_vmlinux_path(dso, map);
1483
1484         if (ret > 0)
1485                 dso__set_loaded(dso);
1486
1487         return ret;
1488 }
1489
1490 static char *get_kernel_version(const char *root_dir)
1491 {
1492         char version[PATH_MAX];
1493         FILE *file;
1494         char *name, *tmp;
1495         const char *prefix = "Linux version ";
1496
1497         sprintf(version, "%s/proc/version", root_dir);
1498         file = fopen(version, "r");
1499         if (!file)
1500                 return NULL;
1501
1502         tmp = fgets(version, sizeof(version), file);
1503         fclose(file);
1504         if (!tmp)
1505                 return NULL;
1506
1507         name = strstr(version, prefix);
1508         if (!name)
1509                 return NULL;
1510         name += strlen(prefix);
1511         tmp = strchr(name, ' ');
1512         if (tmp)
1513                 *tmp = '\0';
1514
1515         return strdup(name);
1516 }
1517
1518 static bool is_kmod_dso(struct dso *dso)
1519 {
1520         return dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
1521                dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE;
1522 }
1523
1524 static int maps__set_module_path(struct maps *maps, const char *path, struct kmod_path *m)
1525 {
1526         char *long_name;
1527         struct dso *dso;
1528         struct map *map = maps__find_by_name(maps, m->name);
1529
1530         if (map == NULL)
1531                 return 0;
1532
1533         long_name = strdup(path);
1534         if (long_name == NULL)
1535                 return -ENOMEM;
1536
1537         dso = map__dso(map);
1538         dso__set_long_name(dso, long_name, true);
1539         dso__kernel_module_get_build_id(dso, "");
1540
1541         /*
1542          * Full name could reveal us kmod compression, so
1543          * we need to update the symtab_type if needed.
1544          */
1545         if (m->comp && is_kmod_dso(dso)) {
1546                 dso->symtab_type++;
1547                 dso->comp = m->comp;
1548         }
1549
1550         return 0;
1551 }
1552
1553 static int maps__set_modules_path_dir(struct maps *maps, const char *dir_name, int depth)
1554 {
1555         struct dirent *dent;
1556         DIR *dir = opendir(dir_name);
1557         int ret = 0;
1558
1559         if (!dir) {
1560                 pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
1561                 return -1;
1562         }
1563
1564         while ((dent = readdir(dir)) != NULL) {
1565                 char path[PATH_MAX];
1566                 struct stat st;
1567
1568                 /*sshfs might return bad dent->d_type, so we have to stat*/
1569                 path__join(path, sizeof(path), dir_name, dent->d_name);
1570                 if (stat(path, &st))
1571                         continue;
1572
1573                 if (S_ISDIR(st.st_mode)) {
1574                         if (!strcmp(dent->d_name, ".") ||
1575                             !strcmp(dent->d_name, ".."))
1576                                 continue;
1577
1578                         /* Do not follow top-level source and build symlinks */
1579                         if (depth == 0) {
1580                                 if (!strcmp(dent->d_name, "source") ||
1581                                     !strcmp(dent->d_name, "build"))
1582                                         continue;
1583                         }
1584
1585                         ret = maps__set_modules_path_dir(maps, path, depth + 1);
1586                         if (ret < 0)
1587                                 goto out;
1588                 } else {
1589                         struct kmod_path m;
1590
1591                         ret = kmod_path__parse_name(&m, dent->d_name);
1592                         if (ret)
1593                                 goto out;
1594
1595                         if (m.kmod)
1596                                 ret = maps__set_module_path(maps, path, &m);
1597
1598                         zfree(&m.name);
1599
1600                         if (ret)
1601                                 goto out;
1602                 }
1603         }
1604
1605 out:
1606         closedir(dir);
1607         return ret;
1608 }
1609
1610 static int machine__set_modules_path(struct machine *machine)
1611 {
1612         char *version;
1613         char modules_path[PATH_MAX];
1614
1615         version = get_kernel_version(machine->root_dir);
1616         if (!version)
1617                 return -1;
1618
1619         snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
1620                  machine->root_dir, version);
1621         free(version);
1622
1623         return maps__set_modules_path_dir(machine__kernel_maps(machine), modules_path, 0);
1624 }
1625 int __weak arch__fix_module_text_start(u64 *start __maybe_unused,
1626                                 u64 *size __maybe_unused,
1627                                 const char *name __maybe_unused)
1628 {
1629         return 0;
1630 }
1631
1632 static int machine__create_module(void *arg, const char *name, u64 start,
1633                                   u64 size)
1634 {
1635         struct machine *machine = arg;
1636         struct map *map;
1637
1638         if (arch__fix_module_text_start(&start, &size, name) < 0)
1639                 return -1;
1640
1641         map = machine__addnew_module_map(machine, start, name);
1642         if (map == NULL)
1643                 return -1;
1644         map__set_end(map, start + size);
1645
1646         dso__kernel_module_get_build_id(map__dso(map), machine->root_dir);
1647         map__put(map);
1648         return 0;
1649 }
1650
1651 static int machine__create_modules(struct machine *machine)
1652 {
1653         const char *modules;
1654         char path[PATH_MAX];
1655
1656         if (machine__is_default_guest(machine)) {
1657                 modules = symbol_conf.default_guest_modules;
1658         } else {
1659                 snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
1660                 modules = path;
1661         }
1662
1663         if (symbol__restricted_filename(modules, "/proc/modules"))
1664                 return -1;
1665
1666         if (modules__parse(modules, machine, machine__create_module))
1667                 return -1;
1668
1669         if (!machine__set_modules_path(machine))
1670                 return 0;
1671
1672         pr_debug("Problems setting modules path maps, continuing anyway...\n");
1673
1674         return 0;
1675 }
1676
1677 static void machine__set_kernel_mmap(struct machine *machine,
1678                                      u64 start, u64 end)
1679 {
1680         map__set_start(machine->vmlinux_map, start);
1681         map__set_end(machine->vmlinux_map, end);
1682         /*
1683          * Be a bit paranoid here, some perf.data file came with
1684          * a zero sized synthesized MMAP event for the kernel.
1685          */
1686         if (start == 0 && end == 0)
1687                 map__set_end(machine->vmlinux_map, ~0ULL);
1688 }
1689
1690 static int machine__update_kernel_mmap(struct machine *machine,
1691                                      u64 start, u64 end)
1692 {
1693         struct map *orig, *updated;
1694         int err;
1695
1696         orig = machine->vmlinux_map;
1697         updated = map__get(orig);
1698
1699         machine->vmlinux_map = updated;
1700         machine__set_kernel_mmap(machine, start, end);
1701         maps__remove(machine__kernel_maps(machine), orig);
1702         err = maps__insert(machine__kernel_maps(machine), updated);
1703         map__put(orig);
1704
1705         return err;
1706 }
1707
1708 int machine__create_kernel_maps(struct machine *machine)
1709 {
1710         struct dso *kernel = machine__get_kernel(machine);
1711         const char *name = NULL;
1712         u64 start = 0, end = ~0ULL;
1713         int ret;
1714
1715         if (kernel == NULL)
1716                 return -1;
1717
1718         ret = __machine__create_kernel_maps(machine, kernel);
1719         if (ret < 0)
1720                 goto out_put;
1721
1722         if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
1723                 if (machine__is_host(machine))
1724                         pr_debug("Problems creating module maps, "
1725                                  "continuing anyway...\n");
1726                 else
1727                         pr_debug("Problems creating module maps for guest %d, "
1728                                  "continuing anyway...\n", machine->pid);
1729         }
1730
1731         if (!machine__get_running_kernel_start(machine, &name, &start, &end)) {
1732                 if (name &&
1733                     map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map, name, start)) {
1734                         machine__destroy_kernel_maps(machine);
1735                         ret = -1;
1736                         goto out_put;
1737                 }
1738
1739                 /*
1740                  * we have a real start address now, so re-order the kmaps
1741                  * assume it's the last in the kmaps
1742                  */
1743                 ret = machine__update_kernel_mmap(machine, start, end);
1744                 if (ret < 0)
1745                         goto out_put;
1746         }
1747
1748         if (machine__create_extra_kernel_maps(machine, kernel))
1749                 pr_debug("Problems creating extra kernel maps, continuing anyway...\n");
1750
1751         if (end == ~0ULL) {
1752                 /* update end address of the kernel map using adjacent module address */
1753                 struct map_rb_node *rb_node = maps__find_node(machine__kernel_maps(machine),
1754                                                         machine__kernel_map(machine));
1755                 struct map_rb_node *next = map_rb_node__next(rb_node);
1756
1757                 if (next)
1758                         machine__set_kernel_mmap(machine, start, map__start(next->map));
1759         }
1760
1761 out_put:
1762         dso__put(kernel);
1763         return ret;
1764 }
1765
1766 static bool machine__uses_kcore(struct machine *machine)
1767 {
1768         struct dso *dso;
1769
1770         list_for_each_entry(dso, &machine->dsos.head, node) {
1771                 if (dso__is_kcore(dso))
1772                         return true;
1773         }
1774
1775         return false;
1776 }
1777
1778 static bool perf_event__is_extra_kernel_mmap(struct machine *machine,
1779                                              struct extra_kernel_map *xm)
1780 {
1781         return machine__is(machine, "x86_64") &&
1782                is_entry_trampoline(xm->name);
1783 }
1784
1785 static int machine__process_extra_kernel_map(struct machine *machine,
1786                                              struct extra_kernel_map *xm)
1787 {
1788         struct dso *kernel = machine__kernel_dso(machine);
1789
1790         if (kernel == NULL)
1791                 return -1;
1792
1793         return machine__create_extra_kernel_map(machine, kernel, xm);
1794 }
1795
1796 static int machine__process_kernel_mmap_event(struct machine *machine,
1797                                               struct extra_kernel_map *xm,
1798                                               struct build_id *bid)
1799 {
1800         enum dso_space_type dso_space;
1801         bool is_kernel_mmap;
1802         const char *mmap_name = machine->mmap_name;
1803
1804         /* If we have maps from kcore then we do not need or want any others */
1805         if (machine__uses_kcore(machine))
1806                 return 0;
1807
1808         if (machine__is_host(machine))
1809                 dso_space = DSO_SPACE__KERNEL;
1810         else
1811                 dso_space = DSO_SPACE__KERNEL_GUEST;
1812
1813         is_kernel_mmap = memcmp(xm->name, mmap_name, strlen(mmap_name) - 1) == 0;
1814         if (!is_kernel_mmap && !machine__is_host(machine)) {
1815                 /*
1816                  * If the event was recorded inside the guest and injected into
1817                  * the host perf.data file, then it will match a host mmap_name,
1818                  * so try that - see machine__set_mmap_name().
1819                  */
1820                 mmap_name = "[kernel.kallsyms]";
1821                 is_kernel_mmap = memcmp(xm->name, mmap_name, strlen(mmap_name) - 1) == 0;
1822         }
1823         if (xm->name[0] == '/' ||
1824             (!is_kernel_mmap && xm->name[0] == '[')) {
1825                 struct map *map = machine__addnew_module_map(machine, xm->start, xm->name);
1826
1827                 if (map == NULL)
1828                         goto out_problem;
1829
1830                 map__set_end(map, map__start(map) + xm->end - xm->start);
1831
1832                 if (build_id__is_defined(bid))
1833                         dso__set_build_id(map__dso(map), bid);
1834
1835                 map__put(map);
1836         } else if (is_kernel_mmap) {
1837                 const char *symbol_name = xm->name + strlen(mmap_name);
1838                 /*
1839                  * Should be there already, from the build-id table in
1840                  * the header.
1841                  */
1842                 struct dso *kernel = NULL;
1843                 struct dso *dso;
1844
1845                 down_read(&machine->dsos.lock);
1846
1847                 list_for_each_entry(dso, &machine->dsos.head, node) {
1848
1849                         /*
1850                          * The cpumode passed to is_kernel_module is not the
1851                          * cpumode of *this* event. If we insist on passing
1852                          * correct cpumode to is_kernel_module, we should
1853                          * record the cpumode when we adding this dso to the
1854                          * linked list.
1855                          *
1856                          * However we don't really need passing correct
1857                          * cpumode.  We know the correct cpumode must be kernel
1858                          * mode (if not, we should not link it onto kernel_dsos
1859                          * list).
1860                          *
1861                          * Therefore, we pass PERF_RECORD_MISC_CPUMODE_UNKNOWN.
1862                          * is_kernel_module() treats it as a kernel cpumode.
1863                          */
1864
1865                         if (!dso->kernel ||
1866                             is_kernel_module(dso->long_name,
1867                                              PERF_RECORD_MISC_CPUMODE_UNKNOWN))
1868                                 continue;
1869
1870
1871                         kernel = dso__get(dso);
1872                         break;
1873                 }
1874
1875                 up_read(&machine->dsos.lock);
1876
1877                 if (kernel == NULL)
1878                         kernel = machine__findnew_dso(machine, machine->mmap_name);
1879                 if (kernel == NULL)
1880                         goto out_problem;
1881
1882                 kernel->kernel = dso_space;
1883                 if (__machine__create_kernel_maps(machine, kernel) < 0) {
1884                         dso__put(kernel);
1885                         goto out_problem;
1886                 }
1887
1888                 if (strstr(kernel->long_name, "vmlinux"))
1889                         dso__set_short_name(kernel, "[kernel.vmlinux]", false);
1890
1891                 if (machine__update_kernel_mmap(machine, xm->start, xm->end) < 0) {
1892                         dso__put(kernel);
1893                         goto out_problem;
1894                 }
1895
1896                 if (build_id__is_defined(bid))
1897                         dso__set_build_id(kernel, bid);
1898
1899                 /*
1900                  * Avoid using a zero address (kptr_restrict) for the ref reloc
1901                  * symbol. Effectively having zero here means that at record
1902                  * time /proc/sys/kernel/kptr_restrict was non zero.
1903                  */
1904                 if (xm->pgoff != 0) {
1905                         map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map,
1906                                                         symbol_name,
1907                                                         xm->pgoff);
1908                 }
1909
1910                 if (machine__is_default_guest(machine)) {
1911                         /*
1912                          * preload dso of guest kernel and modules
1913                          */
1914                         dso__load(kernel, machine__kernel_map(machine));
1915                 }
1916                 dso__put(kernel);
1917         } else if (perf_event__is_extra_kernel_mmap(machine, xm)) {
1918                 return machine__process_extra_kernel_map(machine, xm);
1919         }
1920         return 0;
1921 out_problem:
1922         return -1;
1923 }
1924
1925 int machine__process_mmap2_event(struct machine *machine,
1926                                  union perf_event *event,
1927                                  struct perf_sample *sample)
1928 {
1929         struct thread *thread;
1930         struct map *map;
1931         struct dso_id dso_id = {
1932                 .maj = event->mmap2.maj,
1933                 .min = event->mmap2.min,
1934                 .ino = event->mmap2.ino,
1935                 .ino_generation = event->mmap2.ino_generation,
1936         };
1937         struct build_id __bid, *bid = NULL;
1938         int ret = 0;
1939
1940         if (dump_trace)
1941                 perf_event__fprintf_mmap2(event, stdout);
1942
1943         if (event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID) {
1944                 bid = &__bid;
1945                 build_id__init(bid, event->mmap2.build_id, event->mmap2.build_id_size);
1946         }
1947
1948         if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1949             sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1950                 struct extra_kernel_map xm = {
1951                         .start = event->mmap2.start,
1952                         .end   = event->mmap2.start + event->mmap2.len,
1953                         .pgoff = event->mmap2.pgoff,
1954                 };
1955
1956                 strlcpy(xm.name, event->mmap2.filename, KMAP_NAME_LEN);
1957                 ret = machine__process_kernel_mmap_event(machine, &xm, bid);
1958                 if (ret < 0)
1959                         goto out_problem;
1960                 return 0;
1961         }
1962
1963         thread = machine__findnew_thread(machine, event->mmap2.pid,
1964                                         event->mmap2.tid);
1965         if (thread == NULL)
1966                 goto out_problem;
1967
1968         map = map__new(machine, event->mmap2.start,
1969                         event->mmap2.len, event->mmap2.pgoff,
1970                         &dso_id, event->mmap2.prot,
1971                         event->mmap2.flags, bid,
1972                         event->mmap2.filename, thread);
1973
1974         if (map == NULL)
1975                 goto out_problem_map;
1976
1977         ret = thread__insert_map(thread, map);
1978         if (ret)
1979                 goto out_problem_insert;
1980
1981         thread__put(thread);
1982         map__put(map);
1983         return 0;
1984
1985 out_problem_insert:
1986         map__put(map);
1987 out_problem_map:
1988         thread__put(thread);
1989 out_problem:
1990         dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
1991         return 0;
1992 }
1993
1994 int machine__process_mmap_event(struct machine *machine, union perf_event *event,
1995                                 struct perf_sample *sample)
1996 {
1997         struct thread *thread;
1998         struct map *map;
1999         u32 prot = 0;
2000         int ret = 0;
2001
2002         if (dump_trace)
2003                 perf_event__fprintf_mmap(event, stdout);
2004
2005         if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
2006             sample->cpumode == PERF_RECORD_MISC_KERNEL) {
2007                 struct extra_kernel_map xm = {
2008                         .start = event->mmap.start,
2009                         .end   = event->mmap.start + event->mmap.len,
2010                         .pgoff = event->mmap.pgoff,
2011                 };
2012
2013                 strlcpy(xm.name, event->mmap.filename, KMAP_NAME_LEN);
2014                 ret = machine__process_kernel_mmap_event(machine, &xm, NULL);
2015                 if (ret < 0)
2016                         goto out_problem;
2017                 return 0;
2018         }
2019
2020         thread = machine__findnew_thread(machine, event->mmap.pid,
2021                                          event->mmap.tid);
2022         if (thread == NULL)
2023                 goto out_problem;
2024
2025         if (!(event->header.misc & PERF_RECORD_MISC_MMAP_DATA))
2026                 prot = PROT_EXEC;
2027
2028         map = map__new(machine, event->mmap.start,
2029                         event->mmap.len, event->mmap.pgoff,
2030                         NULL, prot, 0, NULL, event->mmap.filename, thread);
2031
2032         if (map == NULL)
2033                 goto out_problem_map;
2034
2035         ret = thread__insert_map(thread, map);
2036         if (ret)
2037                 goto out_problem_insert;
2038
2039         thread__put(thread);
2040         map__put(map);
2041         return 0;
2042
2043 out_problem_insert:
2044         map__put(map);
2045 out_problem_map:
2046         thread__put(thread);
2047 out_problem:
2048         dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
2049         return 0;
2050 }
2051
2052 static void __machine__remove_thread(struct machine *machine, struct thread_rb_node *nd,
2053                                      struct thread *th, bool lock)
2054 {
2055         struct threads *threads = machine__threads(machine, thread__tid(th));
2056
2057         if (!nd)
2058                 nd = thread_rb_node__find(th, &threads->entries.rb_root);
2059
2060         if (threads->last_match && RC_CHK_ACCESS(threads->last_match) == RC_CHK_ACCESS(th))
2061                 threads__set_last_match(threads, NULL);
2062
2063         if (lock)
2064                 down_write(&threads->lock);
2065
2066         BUG_ON(refcount_read(thread__refcnt(th)) == 0);
2067
2068         thread__put(nd->thread);
2069         rb_erase_cached(&nd->rb_node, &threads->entries);
2070         RB_CLEAR_NODE(&nd->rb_node);
2071         --threads->nr;
2072
2073         free(nd);
2074
2075         if (lock)
2076                 up_write(&threads->lock);
2077 }
2078
2079 void machine__remove_thread(struct machine *machine, struct thread *th)
2080 {
2081         return __machine__remove_thread(machine, NULL, th, true);
2082 }
2083
2084 int machine__process_fork_event(struct machine *machine, union perf_event *event,
2085                                 struct perf_sample *sample)
2086 {
2087         struct thread *thread = machine__find_thread(machine,
2088                                                      event->fork.pid,
2089                                                      event->fork.tid);
2090         struct thread *parent = machine__findnew_thread(machine,
2091                                                         event->fork.ppid,
2092                                                         event->fork.ptid);
2093         bool do_maps_clone = true;
2094         int err = 0;
2095
2096         if (dump_trace)
2097                 perf_event__fprintf_task(event, stdout);
2098
2099         /*
2100          * There may be an existing thread that is not actually the parent,
2101          * either because we are processing events out of order, or because the
2102          * (fork) event that would have removed the thread was lost. Assume the
2103          * latter case and continue on as best we can.
2104          */
2105         if (thread__pid(parent) != (pid_t)event->fork.ppid) {
2106                 dump_printf("removing erroneous parent thread %d/%d\n",
2107                             thread__pid(parent), thread__tid(parent));
2108                 machine__remove_thread(machine, parent);
2109                 thread__put(parent);
2110                 parent = machine__findnew_thread(machine, event->fork.ppid,
2111                                                  event->fork.ptid);
2112         }
2113
2114         /* if a thread currently exists for the thread id remove it */
2115         if (thread != NULL) {
2116                 machine__remove_thread(machine, thread);
2117                 thread__put(thread);
2118         }
2119
2120         thread = machine__findnew_thread(machine, event->fork.pid,
2121                                          event->fork.tid);
2122         /*
2123          * When synthesizing FORK events, we are trying to create thread
2124          * objects for the already running tasks on the machine.
2125          *
2126          * Normally, for a kernel FORK event, we want to clone the parent's
2127          * maps because that is what the kernel just did.
2128          *
2129          * But when synthesizing, this should not be done.  If we do, we end up
2130          * with overlapping maps as we process the synthesized MMAP2 events that
2131          * get delivered shortly thereafter.
2132          *
2133          * Use the FORK event misc flags in an internal way to signal this
2134          * situation, so we can elide the map clone when appropriate.
2135          */
2136         if (event->fork.header.misc & PERF_RECORD_MISC_FORK_EXEC)
2137                 do_maps_clone = false;
2138
2139         if (thread == NULL || parent == NULL ||
2140             thread__fork(thread, parent, sample->time, do_maps_clone) < 0) {
2141                 dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
2142                 err = -1;
2143         }
2144         thread__put(thread);
2145         thread__put(parent);
2146
2147         return err;
2148 }
2149
2150 int machine__process_exit_event(struct machine *machine, union perf_event *event,
2151                                 struct perf_sample *sample __maybe_unused)
2152 {
2153         struct thread *thread = machine__find_thread(machine,
2154                                                      event->fork.pid,
2155                                                      event->fork.tid);
2156
2157         if (dump_trace)
2158                 perf_event__fprintf_task(event, stdout);
2159
2160         if (thread != NULL)
2161                 thread__put(thread);
2162
2163         return 0;
2164 }
2165
2166 int machine__process_event(struct machine *machine, union perf_event *event,
2167                            struct perf_sample *sample)
2168 {
2169         int ret;
2170
2171         switch (event->header.type) {
2172         case PERF_RECORD_COMM:
2173                 ret = machine__process_comm_event(machine, event, sample); break;
2174         case PERF_RECORD_MMAP:
2175                 ret = machine__process_mmap_event(machine, event, sample); break;
2176         case PERF_RECORD_NAMESPACES:
2177                 ret = machine__process_namespaces_event(machine, event, sample); break;
2178         case PERF_RECORD_CGROUP:
2179                 ret = machine__process_cgroup_event(machine, event, sample); break;
2180         case PERF_RECORD_MMAP2:
2181                 ret = machine__process_mmap2_event(machine, event, sample); break;
2182         case PERF_RECORD_FORK:
2183                 ret = machine__process_fork_event(machine, event, sample); break;
2184         case PERF_RECORD_EXIT:
2185                 ret = machine__process_exit_event(machine, event, sample); break;
2186         case PERF_RECORD_LOST:
2187                 ret = machine__process_lost_event(machine, event, sample); break;
2188         case PERF_RECORD_AUX:
2189                 ret = machine__process_aux_event(machine, event); break;
2190         case PERF_RECORD_ITRACE_START:
2191                 ret = machine__process_itrace_start_event(machine, event); break;
2192         case PERF_RECORD_LOST_SAMPLES:
2193                 ret = machine__process_lost_samples_event(machine, event, sample); break;
2194         case PERF_RECORD_SWITCH:
2195         case PERF_RECORD_SWITCH_CPU_WIDE:
2196                 ret = machine__process_switch_event(machine, event); break;
2197         case PERF_RECORD_KSYMBOL:
2198                 ret = machine__process_ksymbol(machine, event, sample); break;
2199         case PERF_RECORD_BPF_EVENT:
2200                 ret = machine__process_bpf(machine, event, sample); break;
2201         case PERF_RECORD_TEXT_POKE:
2202                 ret = machine__process_text_poke(machine, event, sample); break;
2203         case PERF_RECORD_AUX_OUTPUT_HW_ID:
2204                 ret = machine__process_aux_output_hw_id_event(machine, event); break;
2205         default:
2206                 ret = -1;
2207                 break;
2208         }
2209
2210         return ret;
2211 }
2212
2213 static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
2214 {
2215         if (!regexec(regex, sym->name, 0, NULL, 0))
2216                 return true;
2217         return false;
2218 }
2219
2220 static void ip__resolve_ams(struct thread *thread,
2221                             struct addr_map_symbol *ams,
2222                             u64 ip)
2223 {
2224         struct addr_location al;
2225
2226         addr_location__init(&al);
2227         /*
2228          * We cannot use the header.misc hint to determine whether a
2229          * branch stack address is user, kernel, guest, hypervisor.
2230          * Branches may straddle the kernel/user/hypervisor boundaries.
2231          * Thus, we have to try consecutively until we find a match
2232          * or else, the symbol is unknown
2233          */
2234         thread__find_cpumode_addr_location(thread, ip, &al);
2235
2236         ams->addr = ip;
2237         ams->al_addr = al.addr;
2238         ams->al_level = al.level;
2239         ams->ms.maps = maps__get(al.maps);
2240         ams->ms.sym = al.sym;
2241         ams->ms.map = map__get(al.map);
2242         ams->phys_addr = 0;
2243         ams->data_page_size = 0;
2244         addr_location__exit(&al);
2245 }
2246
2247 static void ip__resolve_data(struct thread *thread,
2248                              u8 m, struct addr_map_symbol *ams,
2249                              u64 addr, u64 phys_addr, u64 daddr_page_size)
2250 {
2251         struct addr_location al;
2252
2253         addr_location__init(&al);
2254
2255         thread__find_symbol(thread, m, addr, &al);
2256
2257         ams->addr = addr;
2258         ams->al_addr = al.addr;
2259         ams->al_level = al.level;
2260         ams->ms.maps = maps__get(al.maps);
2261         ams->ms.sym = al.sym;
2262         ams->ms.map = map__get(al.map);
2263         ams->phys_addr = phys_addr;
2264         ams->data_page_size = daddr_page_size;
2265         addr_location__exit(&al);
2266 }
2267
2268 struct mem_info *sample__resolve_mem(struct perf_sample *sample,
2269                                      struct addr_location *al)
2270 {
2271         struct mem_info *mi = mem_info__new();
2272
2273         if (!mi)
2274                 return NULL;
2275
2276         ip__resolve_ams(al->thread, &mi->iaddr, sample->ip);
2277         ip__resolve_data(al->thread, al->cpumode, &mi->daddr,
2278                          sample->addr, sample->phys_addr,
2279                          sample->data_page_size);
2280         mi->data_src.val = sample->data_src;
2281
2282         return mi;
2283 }
2284
2285 static char *callchain_srcline(struct map_symbol *ms, u64 ip)
2286 {
2287         struct map *map = ms->map;
2288         char *srcline = NULL;
2289         struct dso *dso;
2290
2291         if (!map || callchain_param.key == CCKEY_FUNCTION)
2292                 return srcline;
2293
2294         dso = map__dso(map);
2295         srcline = srcline__tree_find(&dso->srclines, ip);
2296         if (!srcline) {
2297                 bool show_sym = false;
2298                 bool show_addr = callchain_param.key == CCKEY_ADDRESS;
2299
2300                 srcline = get_srcline(dso, map__rip_2objdump(map, ip),
2301                                       ms->sym, show_sym, show_addr, ip);
2302                 srcline__tree_insert(&dso->srclines, ip, srcline);
2303         }
2304
2305         return srcline;
2306 }
2307
2308 struct iterations {
2309         int nr_loop_iter;
2310         u64 cycles;
2311 };
2312
2313 static int add_callchain_ip(struct thread *thread,
2314                             struct callchain_cursor *cursor,
2315                             struct symbol **parent,
2316                             struct addr_location *root_al,
2317                             u8 *cpumode,
2318                             u64 ip,
2319                             bool branch,
2320                             struct branch_flags *flags,
2321                             struct iterations *iter,
2322                             u64 branch_from)
2323 {
2324         struct map_symbol ms = {};
2325         struct addr_location al;
2326         int nr_loop_iter = 0, err = 0;
2327         u64 iter_cycles = 0;
2328         const char *srcline = NULL;
2329
2330         addr_location__init(&al);
2331         al.filtered = 0;
2332         al.sym = NULL;
2333         al.srcline = NULL;
2334         if (!cpumode) {
2335                 thread__find_cpumode_addr_location(thread, ip, &al);
2336         } else {
2337                 if (ip >= PERF_CONTEXT_MAX) {
2338                         switch (ip) {
2339                         case PERF_CONTEXT_HV:
2340                                 *cpumode = PERF_RECORD_MISC_HYPERVISOR;
2341                                 break;
2342                         case PERF_CONTEXT_KERNEL:
2343                                 *cpumode = PERF_RECORD_MISC_KERNEL;
2344                                 break;
2345                         case PERF_CONTEXT_USER:
2346                                 *cpumode = PERF_RECORD_MISC_USER;
2347                                 break;
2348                         default:
2349                                 pr_debug("invalid callchain context: "
2350                                          "%"PRId64"\n", (s64) ip);
2351                                 /*
2352                                  * It seems the callchain is corrupted.
2353                                  * Discard all.
2354                                  */
2355                                 callchain_cursor_reset(cursor);
2356                                 err = 1;
2357                                 goto out;
2358                         }
2359                         goto out;
2360                 }
2361                 thread__find_symbol(thread, *cpumode, ip, &al);
2362         }
2363
2364         if (al.sym != NULL) {
2365                 if (perf_hpp_list.parent && !*parent &&
2366                     symbol__match_regex(al.sym, &parent_regex))
2367                         *parent = al.sym;
2368                 else if (have_ignore_callees && root_al &&
2369                   symbol__match_regex(al.sym, &ignore_callees_regex)) {
2370                         /* Treat this symbol as the root,
2371                            forgetting its callees. */
2372                         addr_location__copy(root_al, &al);
2373                         callchain_cursor_reset(cursor);
2374                 }
2375         }
2376
2377         if (symbol_conf.hide_unresolved && al.sym == NULL)
2378                 goto out;
2379
2380         if (iter) {
2381                 nr_loop_iter = iter->nr_loop_iter;
2382                 iter_cycles = iter->cycles;
2383         }
2384
2385         ms.maps = maps__get(al.maps);
2386         ms.map = map__get(al.map);
2387         ms.sym = al.sym;
2388
2389         if (!branch && append_inlines(cursor, &ms, ip) == 0)
2390                 goto out;
2391
2392         srcline = callchain_srcline(&ms, al.addr);
2393         err = callchain_cursor_append(cursor, ip, &ms,
2394                                       branch, flags, nr_loop_iter,
2395                                       iter_cycles, branch_from, srcline);
2396 out:
2397         addr_location__exit(&al);
2398         maps__put(ms.maps);
2399         map__put(ms.map);
2400         return err;
2401 }
2402
2403 struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
2404                                            struct addr_location *al)
2405 {
2406         unsigned int i;
2407         const struct branch_stack *bs = sample->branch_stack;
2408         struct branch_entry *entries = perf_sample__branch_entries(sample);
2409         struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
2410
2411         if (!bi)
2412                 return NULL;
2413
2414         for (i = 0; i < bs->nr; i++) {
2415                 ip__resolve_ams(al->thread, &bi[i].to, entries[i].to);
2416                 ip__resolve_ams(al->thread, &bi[i].from, entries[i].from);
2417                 bi[i].flags = entries[i].flags;
2418         }
2419         return bi;
2420 }
2421
2422 static void save_iterations(struct iterations *iter,
2423                             struct branch_entry *be, int nr)
2424 {
2425         int i;
2426
2427         iter->nr_loop_iter++;
2428         iter->cycles = 0;
2429
2430         for (i = 0; i < nr; i++)
2431                 iter->cycles += be[i].flags.cycles;
2432 }
2433
2434 #define CHASHSZ 127
2435 #define CHASHBITS 7
2436 #define NO_ENTRY 0xff
2437
2438 #define PERF_MAX_BRANCH_DEPTH 127
2439
2440 /* Remove loops. */
2441 static int remove_loops(struct branch_entry *l, int nr,
2442                         struct iterations *iter)
2443 {
2444         int i, j, off;
2445         unsigned char chash[CHASHSZ];
2446
2447         memset(chash, NO_ENTRY, sizeof(chash));
2448
2449         BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
2450
2451         for (i = 0; i < nr; i++) {
2452                 int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
2453
2454                 /* no collision handling for now */
2455                 if (chash[h] == NO_ENTRY) {
2456                         chash[h] = i;
2457                 } else if (l[chash[h]].from == l[i].from) {
2458                         bool is_loop = true;
2459                         /* check if it is a real loop */
2460                         off = 0;
2461                         for (j = chash[h]; j < i && i + off < nr; j++, off++)
2462                                 if (l[j].from != l[i + off].from) {
2463                                         is_loop = false;
2464                                         break;
2465                                 }
2466                         if (is_loop) {
2467                                 j = nr - (i + off);
2468                                 if (j > 0) {
2469                                         save_iterations(iter + i + off,
2470                                                 l + i, off);
2471
2472                                         memmove(iter + i, iter + i + off,
2473                                                 j * sizeof(*iter));
2474
2475                                         memmove(l + i, l + i + off,
2476                                                 j * sizeof(*l));
2477                                 }
2478
2479                                 nr -= off;
2480                         }
2481                 }
2482         }
2483         return nr;
2484 }
2485
2486 static int lbr_callchain_add_kernel_ip(struct thread *thread,
2487                                        struct callchain_cursor *cursor,
2488                                        struct perf_sample *sample,
2489                                        struct symbol **parent,
2490                                        struct addr_location *root_al,
2491                                        u64 branch_from,
2492                                        bool callee, int end)
2493 {
2494         struct ip_callchain *chain = sample->callchain;
2495         u8 cpumode = PERF_RECORD_MISC_USER;
2496         int err, i;
2497
2498         if (callee) {
2499                 for (i = 0; i < end + 1; i++) {
2500                         err = add_callchain_ip(thread, cursor, parent,
2501                                                root_al, &cpumode, chain->ips[i],
2502                                                false, NULL, NULL, branch_from);
2503                         if (err)
2504                                 return err;
2505                 }
2506                 return 0;
2507         }
2508
2509         for (i = end; i >= 0; i--) {
2510                 err = add_callchain_ip(thread, cursor, parent,
2511                                        root_al, &cpumode, chain->ips[i],
2512                                        false, NULL, NULL, branch_from);
2513                 if (err)
2514                         return err;
2515         }
2516
2517         return 0;
2518 }
2519
2520 static void save_lbr_cursor_node(struct thread *thread,
2521                                  struct callchain_cursor *cursor,
2522                                  int idx)
2523 {
2524         struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2525
2526         if (!lbr_stitch)
2527                 return;
2528
2529         if (cursor->pos == cursor->nr) {
2530                 lbr_stitch->prev_lbr_cursor[idx].valid = false;
2531                 return;
2532         }
2533
2534         if (!cursor->curr)
2535                 cursor->curr = cursor->first;
2536         else
2537                 cursor->curr = cursor->curr->next;
2538         memcpy(&lbr_stitch->prev_lbr_cursor[idx], cursor->curr,
2539                sizeof(struct callchain_cursor_node));
2540
2541         lbr_stitch->prev_lbr_cursor[idx].valid = true;
2542         cursor->pos++;
2543 }
2544
2545 static int lbr_callchain_add_lbr_ip(struct thread *thread,
2546                                     struct callchain_cursor *cursor,
2547                                     struct perf_sample *sample,
2548                                     struct symbol **parent,
2549                                     struct addr_location *root_al,
2550                                     u64 *branch_from,
2551                                     bool callee)
2552 {
2553         struct branch_stack *lbr_stack = sample->branch_stack;
2554         struct branch_entry *entries = perf_sample__branch_entries(sample);
2555         u8 cpumode = PERF_RECORD_MISC_USER;
2556         int lbr_nr = lbr_stack->nr;
2557         struct branch_flags *flags;
2558         int err, i;
2559         u64 ip;
2560
2561         /*
2562          * The curr and pos are not used in writing session. They are cleared
2563          * in callchain_cursor_commit() when the writing session is closed.
2564          * Using curr and pos to track the current cursor node.
2565          */
2566         if (thread__lbr_stitch(thread)) {
2567                 cursor->curr = NULL;
2568                 cursor->pos = cursor->nr;
2569                 if (cursor->nr) {
2570                         cursor->curr = cursor->first;
2571                         for (i = 0; i < (int)(cursor->nr - 1); i++)
2572                                 cursor->curr = cursor->curr->next;
2573                 }
2574         }
2575
2576         if (callee) {
2577                 /* Add LBR ip from first entries.to */
2578                 ip = entries[0].to;
2579                 flags = &entries[0].flags;
2580                 *branch_from = entries[0].from;
2581                 err = add_callchain_ip(thread, cursor, parent,
2582                                        root_al, &cpumode, ip,
2583                                        true, flags, NULL,
2584                                        *branch_from);
2585                 if (err)
2586                         return err;
2587
2588                 /*
2589                  * The number of cursor node increases.
2590                  * Move the current cursor node.
2591                  * But does not need to save current cursor node for entry 0.
2592                  * It's impossible to stitch the whole LBRs of previous sample.
2593                  */
2594                 if (thread__lbr_stitch(thread) && (cursor->pos != cursor->nr)) {
2595                         if (!cursor->curr)
2596                                 cursor->curr = cursor->first;
2597                         else
2598                                 cursor->curr = cursor->curr->next;
2599                         cursor->pos++;
2600                 }
2601
2602                 /* Add LBR ip from entries.from one by one. */
2603                 for (i = 0; i < lbr_nr; i++) {
2604                         ip = entries[i].from;
2605                         flags = &entries[i].flags;
2606                         err = add_callchain_ip(thread, cursor, parent,
2607                                                root_al, &cpumode, ip,
2608                                                true, flags, NULL,
2609                                                *branch_from);
2610                         if (err)
2611                                 return err;
2612                         save_lbr_cursor_node(thread, cursor, i);
2613                 }
2614                 return 0;
2615         }
2616
2617         /* Add LBR ip from entries.from one by one. */
2618         for (i = lbr_nr - 1; i >= 0; i--) {
2619                 ip = entries[i].from;
2620                 flags = &entries[i].flags;
2621                 err = add_callchain_ip(thread, cursor, parent,
2622                                        root_al, &cpumode, ip,
2623                                        true, flags, NULL,
2624                                        *branch_from);
2625                 if (err)
2626                         return err;
2627                 save_lbr_cursor_node(thread, cursor, i);
2628         }
2629
2630         /* Add LBR ip from first entries.to */
2631         ip = entries[0].to;
2632         flags = &entries[0].flags;
2633         *branch_from = entries[0].from;
2634         err = add_callchain_ip(thread, cursor, parent,
2635                                root_al, &cpumode, ip,
2636                                true, flags, NULL,
2637                                *branch_from);
2638         if (err)
2639                 return err;
2640
2641         return 0;
2642 }
2643
2644 static int lbr_callchain_add_stitched_lbr_ip(struct thread *thread,
2645                                              struct callchain_cursor *cursor)
2646 {
2647         struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2648         struct callchain_cursor_node *cnode;
2649         struct stitch_list *stitch_node;
2650         int err;
2651
2652         list_for_each_entry(stitch_node, &lbr_stitch->lists, node) {
2653                 cnode = &stitch_node->cursor;
2654
2655                 err = callchain_cursor_append(cursor, cnode->ip,
2656                                               &cnode->ms,
2657                                               cnode->branch,
2658                                               &cnode->branch_flags,
2659                                               cnode->nr_loop_iter,
2660                                               cnode->iter_cycles,
2661                                               cnode->branch_from,
2662                                               cnode->srcline);
2663                 if (err)
2664                         return err;
2665         }
2666         return 0;
2667 }
2668
2669 static struct stitch_list *get_stitch_node(struct thread *thread)
2670 {
2671         struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2672         struct stitch_list *stitch_node;
2673
2674         if (!list_empty(&lbr_stitch->free_lists)) {
2675                 stitch_node = list_first_entry(&lbr_stitch->free_lists,
2676                                                struct stitch_list, node);
2677                 list_del(&stitch_node->node);
2678
2679                 return stitch_node;
2680         }
2681
2682         return malloc(sizeof(struct stitch_list));
2683 }
2684
2685 static bool has_stitched_lbr(struct thread *thread,
2686                              struct perf_sample *cur,
2687                              struct perf_sample *prev,
2688                              unsigned int max_lbr,
2689                              bool callee)
2690 {
2691         struct branch_stack *cur_stack = cur->branch_stack;
2692         struct branch_entry *cur_entries = perf_sample__branch_entries(cur);
2693         struct branch_stack *prev_stack = prev->branch_stack;
2694         struct branch_entry *prev_entries = perf_sample__branch_entries(prev);
2695         struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2696         int i, j, nr_identical_branches = 0;
2697         struct stitch_list *stitch_node;
2698         u64 cur_base, distance;
2699
2700         if (!cur_stack || !prev_stack)
2701                 return false;
2702
2703         /* Find the physical index of the base-of-stack for current sample. */
2704         cur_base = max_lbr - cur_stack->nr + cur_stack->hw_idx + 1;
2705
2706         distance = (prev_stack->hw_idx > cur_base) ? (prev_stack->hw_idx - cur_base) :
2707                                                      (max_lbr + prev_stack->hw_idx - cur_base);
2708         /* Previous sample has shorter stack. Nothing can be stitched. */
2709         if (distance + 1 > prev_stack->nr)
2710                 return false;
2711
2712         /*
2713          * Check if there are identical LBRs between two samples.
2714          * Identical LBRs must have same from, to and flags values. Also,
2715          * they have to be saved in the same LBR registers (same physical
2716          * index).
2717          *
2718          * Starts from the base-of-stack of current sample.
2719          */
2720         for (i = distance, j = cur_stack->nr - 1; (i >= 0) && (j >= 0); i--, j--) {
2721                 if ((prev_entries[i].from != cur_entries[j].from) ||
2722                     (prev_entries[i].to != cur_entries[j].to) ||
2723                     (prev_entries[i].flags.value != cur_entries[j].flags.value))
2724                         break;
2725                 nr_identical_branches++;
2726         }
2727
2728         if (!nr_identical_branches)
2729                 return false;
2730
2731         /*
2732          * Save the LBRs between the base-of-stack of previous sample
2733          * and the base-of-stack of current sample into lbr_stitch->lists.
2734          * These LBRs will be stitched later.
2735          */
2736         for (i = prev_stack->nr - 1; i > (int)distance; i--) {
2737
2738                 if (!lbr_stitch->prev_lbr_cursor[i].valid)
2739                         continue;
2740
2741                 stitch_node = get_stitch_node(thread);
2742                 if (!stitch_node)
2743                         return false;
2744
2745                 memcpy(&stitch_node->cursor, &lbr_stitch->prev_lbr_cursor[i],
2746                        sizeof(struct callchain_cursor_node));
2747
2748                 if (callee)
2749                         list_add(&stitch_node->node, &lbr_stitch->lists);
2750                 else
2751                         list_add_tail(&stitch_node->node, &lbr_stitch->lists);
2752         }
2753
2754         return true;
2755 }
2756
2757 static bool alloc_lbr_stitch(struct thread *thread, unsigned int max_lbr)
2758 {
2759         if (thread__lbr_stitch(thread))
2760                 return true;
2761
2762         thread__set_lbr_stitch(thread, zalloc(sizeof(struct lbr_stitch)));
2763         if (!thread__lbr_stitch(thread))
2764                 goto err;
2765
2766         thread__lbr_stitch(thread)->prev_lbr_cursor =
2767                 calloc(max_lbr + 1, sizeof(struct callchain_cursor_node));
2768         if (!thread__lbr_stitch(thread)->prev_lbr_cursor)
2769                 goto free_lbr_stitch;
2770
2771         INIT_LIST_HEAD(&thread__lbr_stitch(thread)->lists);
2772         INIT_LIST_HEAD(&thread__lbr_stitch(thread)->free_lists);
2773
2774         return true;
2775
2776 free_lbr_stitch:
2777         free(thread__lbr_stitch(thread));
2778         thread__set_lbr_stitch(thread, NULL);
2779 err:
2780         pr_warning("Failed to allocate space for stitched LBRs. Disable LBR stitch\n");
2781         thread__set_lbr_stitch_enable(thread, false);
2782         return false;
2783 }
2784
2785 /*
2786  * Resolve LBR callstack chain sample
2787  * Return:
2788  * 1 on success get LBR callchain information
2789  * 0 no available LBR callchain information, should try fp
2790  * negative error code on other errors.
2791  */
2792 static int resolve_lbr_callchain_sample(struct thread *thread,
2793                                         struct callchain_cursor *cursor,
2794                                         struct perf_sample *sample,
2795                                         struct symbol **parent,
2796                                         struct addr_location *root_al,
2797                                         int max_stack,
2798                                         unsigned int max_lbr)
2799 {
2800         bool callee = (callchain_param.order == ORDER_CALLEE);
2801         struct ip_callchain *chain = sample->callchain;
2802         int chain_nr = min(max_stack, (int)chain->nr), i;
2803         struct lbr_stitch *lbr_stitch;
2804         bool stitched_lbr = false;
2805         u64 branch_from = 0;
2806         int err;
2807
2808         for (i = 0; i < chain_nr; i++) {
2809                 if (chain->ips[i] == PERF_CONTEXT_USER)
2810                         break;
2811         }
2812
2813         /* LBR only affects the user callchain */
2814         if (i == chain_nr)
2815                 return 0;
2816
2817         if (thread__lbr_stitch_enable(thread) && !sample->no_hw_idx &&
2818             (max_lbr > 0) && alloc_lbr_stitch(thread, max_lbr)) {
2819                 lbr_stitch = thread__lbr_stitch(thread);
2820
2821                 stitched_lbr = has_stitched_lbr(thread, sample,
2822                                                 &lbr_stitch->prev_sample,
2823                                                 max_lbr, callee);
2824
2825                 if (!stitched_lbr && !list_empty(&lbr_stitch->lists)) {
2826                         list_replace_init(&lbr_stitch->lists,
2827                                           &lbr_stitch->free_lists);
2828                 }
2829                 memcpy(&lbr_stitch->prev_sample, sample, sizeof(*sample));
2830         }
2831
2832         if (callee) {
2833                 /* Add kernel ip */
2834                 err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2835                                                   parent, root_al, branch_from,
2836                                                   true, i);
2837                 if (err)
2838                         goto error;
2839
2840                 err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2841                                                root_al, &branch_from, true);
2842                 if (err)
2843                         goto error;
2844
2845                 if (stitched_lbr) {
2846                         err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2847                         if (err)
2848                                 goto error;
2849                 }
2850
2851         } else {
2852                 if (stitched_lbr) {
2853                         err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2854                         if (err)
2855                                 goto error;
2856                 }
2857                 err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2858                                                root_al, &branch_from, false);
2859                 if (err)
2860                         goto error;
2861
2862                 /* Add kernel ip */
2863                 err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2864                                                   parent, root_al, branch_from,
2865                                                   false, i);
2866                 if (err)
2867                         goto error;
2868         }
2869         return 1;
2870
2871 error:
2872         return (err < 0) ? err : 0;
2873 }
2874
2875 static int find_prev_cpumode(struct ip_callchain *chain, struct thread *thread,
2876                              struct callchain_cursor *cursor,
2877                              struct symbol **parent,
2878                              struct addr_location *root_al,
2879                              u8 *cpumode, int ent)
2880 {
2881         int err = 0;
2882
2883         while (--ent >= 0) {
2884                 u64 ip = chain->ips[ent];
2885
2886                 if (ip >= PERF_CONTEXT_MAX) {
2887                         err = add_callchain_ip(thread, cursor, parent,
2888                                                root_al, cpumode, ip,
2889                                                false, NULL, NULL, 0);
2890                         break;
2891                 }
2892         }
2893         return err;
2894 }
2895
2896 static u64 get_leaf_frame_caller(struct perf_sample *sample,
2897                 struct thread *thread, int usr_idx)
2898 {
2899         if (machine__normalized_is(maps__machine(thread__maps(thread)), "arm64"))
2900                 return get_leaf_frame_caller_aarch64(sample, thread, usr_idx);
2901         else
2902                 return 0;
2903 }
2904
2905 static int thread__resolve_callchain_sample(struct thread *thread,
2906                                             struct callchain_cursor *cursor,
2907                                             struct evsel *evsel,
2908                                             struct perf_sample *sample,
2909                                             struct symbol **parent,
2910                                             struct addr_location *root_al,
2911                                             int max_stack)
2912 {
2913         struct branch_stack *branch = sample->branch_stack;
2914         struct branch_entry *entries = perf_sample__branch_entries(sample);
2915         struct ip_callchain *chain = sample->callchain;
2916         int chain_nr = 0;
2917         u8 cpumode = PERF_RECORD_MISC_USER;
2918         int i, j, err, nr_entries, usr_idx;
2919         int skip_idx = -1;
2920         int first_call = 0;
2921         u64 leaf_frame_caller;
2922
2923         if (chain)
2924                 chain_nr = chain->nr;
2925
2926         if (evsel__has_branch_callstack(evsel)) {
2927                 struct perf_env *env = evsel__env(evsel);
2928
2929                 err = resolve_lbr_callchain_sample(thread, cursor, sample, parent,
2930                                                    root_al, max_stack,
2931                                                    !env ? 0 : env->max_branches);
2932                 if (err)
2933                         return (err < 0) ? err : 0;
2934         }
2935
2936         /*
2937          * Based on DWARF debug information, some architectures skip
2938          * a callchain entry saved by the kernel.
2939          */
2940         skip_idx = arch_skip_callchain_idx(thread, chain);
2941
2942         /*
2943          * Add branches to call stack for easier browsing. This gives
2944          * more context for a sample than just the callers.
2945          *
2946          * This uses individual histograms of paths compared to the
2947          * aggregated histograms the normal LBR mode uses.
2948          *
2949          * Limitations for now:
2950          * - No extra filters
2951          * - No annotations (should annotate somehow)
2952          */
2953
2954         if (branch && callchain_param.branch_callstack) {
2955                 int nr = min(max_stack, (int)branch->nr);
2956                 struct branch_entry be[nr];
2957                 struct iterations iter[nr];
2958
2959                 if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
2960                         pr_warning("corrupted branch chain. skipping...\n");
2961                         goto check_calls;
2962                 }
2963
2964                 for (i = 0; i < nr; i++) {
2965                         if (callchain_param.order == ORDER_CALLEE) {
2966                                 be[i] = entries[i];
2967
2968                                 if (chain == NULL)
2969                                         continue;
2970
2971                                 /*
2972                                  * Check for overlap into the callchain.
2973                                  * The return address is one off compared to
2974                                  * the branch entry. To adjust for this
2975                                  * assume the calling instruction is not longer
2976                                  * than 8 bytes.
2977                                  */
2978                                 if (i == skip_idx ||
2979                                     chain->ips[first_call] >= PERF_CONTEXT_MAX)
2980                                         first_call++;
2981                                 else if (be[i].from < chain->ips[first_call] &&
2982                                     be[i].from >= chain->ips[first_call] - 8)
2983                                         first_call++;
2984                         } else
2985                                 be[i] = entries[branch->nr - i - 1];
2986                 }
2987
2988                 memset(iter, 0, sizeof(struct iterations) * nr);
2989                 nr = remove_loops(be, nr, iter);
2990
2991                 for (i = 0; i < nr; i++) {
2992                         err = add_callchain_ip(thread, cursor, parent,
2993                                                root_al,
2994                                                NULL, be[i].to,
2995                                                true, &be[i].flags,
2996                                                NULL, be[i].from);
2997
2998                         if (!err)
2999                                 err = add_callchain_ip(thread, cursor, parent, root_al,
3000                                                        NULL, be[i].from,
3001                                                        true, &be[i].flags,
3002                                                        &iter[i], 0);
3003                         if (err == -EINVAL)
3004                                 break;
3005                         if (err)
3006                                 return err;
3007                 }
3008
3009                 if (chain_nr == 0)
3010                         return 0;
3011
3012                 chain_nr -= nr;
3013         }
3014
3015 check_calls:
3016         if (chain && callchain_param.order != ORDER_CALLEE) {
3017                 err = find_prev_cpumode(chain, thread, cursor, parent, root_al,
3018                                         &cpumode, chain->nr - first_call);
3019                 if (err)
3020                         return (err < 0) ? err : 0;
3021         }
3022         for (i = first_call, nr_entries = 0;
3023              i < chain_nr && nr_entries < max_stack; i++) {
3024                 u64 ip;
3025
3026                 if (callchain_param.order == ORDER_CALLEE)
3027                         j = i;
3028                 else
3029                         j = chain->nr - i - 1;
3030
3031 #ifdef HAVE_SKIP_CALLCHAIN_IDX
3032                 if (j == skip_idx)
3033                         continue;
3034 #endif
3035                 ip = chain->ips[j];
3036                 if (ip < PERF_CONTEXT_MAX)
3037                        ++nr_entries;
3038                 else if (callchain_param.order != ORDER_CALLEE) {
3039                         err = find_prev_cpumode(chain, thread, cursor, parent,
3040                                                 root_al, &cpumode, j);
3041                         if (err)
3042                                 return (err < 0) ? err : 0;
3043                         continue;
3044                 }
3045
3046                 /*
3047                  * PERF_CONTEXT_USER allows us to locate where the user stack ends.
3048                  * Depending on callchain_param.order and the position of PERF_CONTEXT_USER,
3049                  * the index will be different in order to add the missing frame
3050                  * at the right place.
3051                  */
3052
3053                 usr_idx = callchain_param.order == ORDER_CALLEE ? j-2 : j-1;
3054
3055                 if (usr_idx >= 0 && chain->ips[usr_idx] == PERF_CONTEXT_USER) {
3056
3057                         leaf_frame_caller = get_leaf_frame_caller(sample, thread, usr_idx);
3058
3059                         /*
3060                          * check if leaf_frame_Caller != ip to not add the same
3061                          * value twice.
3062                          */
3063
3064                         if (leaf_frame_caller && leaf_frame_caller != ip) {
3065
3066                                 err = add_callchain_ip(thread, cursor, parent,
3067                                                root_al, &cpumode, leaf_frame_caller,
3068                                                false, NULL, NULL, 0);
3069                                 if (err)
3070                                         return (err < 0) ? err : 0;
3071                         }
3072                 }
3073
3074                 err = add_callchain_ip(thread, cursor, parent,
3075                                        root_al, &cpumode, ip,
3076                                        false, NULL, NULL, 0);
3077
3078                 if (err)
3079                         return (err < 0) ? err : 0;
3080         }
3081
3082         return 0;
3083 }
3084
3085 static int append_inlines(struct callchain_cursor *cursor, struct map_symbol *ms, u64 ip)
3086 {
3087         struct symbol *sym = ms->sym;
3088         struct map *map = ms->map;
3089         struct inline_node *inline_node;
3090         struct inline_list *ilist;
3091         struct dso *dso;
3092         u64 addr;
3093         int ret = 1;
3094         struct map_symbol ilist_ms;
3095
3096         if (!symbol_conf.inline_name || !map || !sym)
3097                 return ret;
3098
3099         addr = map__dso_map_ip(map, ip);
3100         addr = map__rip_2objdump(map, addr);
3101         dso = map__dso(map);
3102
3103         inline_node = inlines__tree_find(&dso->inlined_nodes, addr);
3104         if (!inline_node) {
3105                 inline_node = dso__parse_addr_inlines(dso, addr, sym);
3106                 if (!inline_node)
3107                         return ret;
3108                 inlines__tree_insert(&dso->inlined_nodes, inline_node);
3109         }
3110
3111         ilist_ms = (struct map_symbol) {
3112                 .maps = maps__get(ms->maps),
3113                 .map = map__get(map),
3114         };
3115         list_for_each_entry(ilist, &inline_node->val, list) {
3116                 ilist_ms.sym = ilist->symbol;
3117                 ret = callchain_cursor_append(cursor, ip, &ilist_ms, false,
3118                                               NULL, 0, 0, 0, ilist->srcline);
3119
3120                 if (ret != 0)
3121                         return ret;
3122         }
3123         map__put(ilist_ms.map);
3124         maps__put(ilist_ms.maps);
3125
3126         return ret;
3127 }
3128
3129 static int unwind_entry(struct unwind_entry *entry, void *arg)
3130 {
3131         struct callchain_cursor *cursor = arg;
3132         const char *srcline = NULL;
3133         u64 addr = entry->ip;
3134
3135         if (symbol_conf.hide_unresolved && entry->ms.sym == NULL)
3136                 return 0;
3137
3138         if (append_inlines(cursor, &entry->ms, entry->ip) == 0)
3139                 return 0;
3140
3141         /*
3142          * Convert entry->ip from a virtual address to an offset in
3143          * its corresponding binary.
3144          */
3145         if (entry->ms.map)
3146                 addr = map__dso_map_ip(entry->ms.map, entry->ip);
3147
3148         srcline = callchain_srcline(&entry->ms, addr);
3149         return callchain_cursor_append(cursor, entry->ip, &entry->ms,
3150                                        false, NULL, 0, 0, 0, srcline);
3151 }
3152
3153 static int thread__resolve_callchain_unwind(struct thread *thread,
3154                                             struct callchain_cursor *cursor,
3155                                             struct evsel *evsel,
3156                                             struct perf_sample *sample,
3157                                             int max_stack)
3158 {
3159         /* Can we do dwarf post unwind? */
3160         if (!((evsel->core.attr.sample_type & PERF_SAMPLE_REGS_USER) &&
3161               (evsel->core.attr.sample_type & PERF_SAMPLE_STACK_USER)))
3162                 return 0;
3163
3164         /* Bail out if nothing was captured. */
3165         if ((!sample->user_regs.regs) ||
3166             (!sample->user_stack.size))
3167                 return 0;
3168
3169         return unwind__get_entries(unwind_entry, cursor,
3170                                    thread, sample, max_stack, false);
3171 }
3172
3173 int thread__resolve_callchain(struct thread *thread,
3174                               struct callchain_cursor *cursor,
3175                               struct evsel *evsel,
3176                               struct perf_sample *sample,
3177                               struct symbol **parent,
3178                               struct addr_location *root_al,
3179                               int max_stack)
3180 {
3181         int ret = 0;
3182
3183         if (cursor == NULL)
3184                 return -ENOMEM;
3185
3186         callchain_cursor_reset(cursor);
3187
3188         if (callchain_param.order == ORDER_CALLEE) {
3189                 ret = thread__resolve_callchain_sample(thread, cursor,
3190                                                        evsel, sample,
3191                                                        parent, root_al,
3192                                                        max_stack);
3193                 if (ret)
3194                         return ret;
3195                 ret = thread__resolve_callchain_unwind(thread, cursor,
3196                                                        evsel, sample,
3197                                                        max_stack);
3198         } else {
3199                 ret = thread__resolve_callchain_unwind(thread, cursor,
3200                                                        evsel, sample,
3201                                                        max_stack);
3202                 if (ret)
3203                         return ret;
3204                 ret = thread__resolve_callchain_sample(thread, cursor,
3205                                                        evsel, sample,
3206                                                        parent, root_al,
3207                                                        max_stack);
3208         }
3209
3210         return ret;
3211 }
3212
3213 int machine__for_each_thread(struct machine *machine,
3214                              int (*fn)(struct thread *thread, void *p),
3215                              void *priv)
3216 {
3217         struct threads *threads;
3218         struct rb_node *nd;
3219         int rc = 0;
3220         int i;
3221
3222         for (i = 0; i < THREADS__TABLE_SIZE; i++) {
3223                 threads = &machine->threads[i];
3224                 for (nd = rb_first_cached(&threads->entries); nd;
3225                      nd = rb_next(nd)) {
3226                         struct thread_rb_node *trb = rb_entry(nd, struct thread_rb_node, rb_node);
3227
3228                         rc = fn(trb->thread, priv);
3229                         if (rc != 0)
3230                                 return rc;
3231                 }
3232         }
3233         return rc;
3234 }
3235
3236 int machines__for_each_thread(struct machines *machines,
3237                               int (*fn)(struct thread *thread, void *p),
3238                               void *priv)
3239 {
3240         struct rb_node *nd;
3241         int rc = 0;
3242
3243         rc = machine__for_each_thread(&machines->host, fn, priv);
3244         if (rc != 0)
3245                 return rc;
3246
3247         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
3248                 struct machine *machine = rb_entry(nd, struct machine, rb_node);
3249
3250                 rc = machine__for_each_thread(machine, fn, priv);
3251                 if (rc != 0)
3252                         return rc;
3253         }
3254         return rc;
3255 }
3256
3257 pid_t machine__get_current_tid(struct machine *machine, int cpu)
3258 {
3259         if (cpu < 0 || (size_t)cpu >= machine->current_tid_sz)
3260                 return -1;
3261
3262         return machine->current_tid[cpu];
3263 }
3264
3265 int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
3266                              pid_t tid)
3267 {
3268         struct thread *thread;
3269         const pid_t init_val = -1;
3270
3271         if (cpu < 0)
3272                 return -EINVAL;
3273
3274         if (realloc_array_as_needed(machine->current_tid,
3275                                     machine->current_tid_sz,
3276                                     (unsigned int)cpu,
3277                                     &init_val))
3278                 return -ENOMEM;
3279
3280         machine->current_tid[cpu] = tid;
3281
3282         thread = machine__findnew_thread(machine, pid, tid);
3283         if (!thread)
3284                 return -ENOMEM;
3285
3286         thread__set_cpu(thread, cpu);
3287         thread__put(thread);
3288
3289         return 0;
3290 }
3291
3292 /*
3293  * Compares the raw arch string. N.B. see instead perf_env__arch() or
3294  * machine__normalized_is() if a normalized arch is needed.
3295  */
3296 bool machine__is(struct machine *machine, const char *arch)
3297 {
3298         return machine && !strcmp(perf_env__raw_arch(machine->env), arch);
3299 }
3300
3301 bool machine__normalized_is(struct machine *machine, const char *arch)
3302 {
3303         return machine && !strcmp(perf_env__arch(machine->env), arch);
3304 }
3305
3306 int machine__nr_cpus_avail(struct machine *machine)
3307 {
3308         return machine ? perf_env__nr_cpus_avail(machine->env) : 0;
3309 }
3310
3311 int machine__get_kernel_start(struct machine *machine)
3312 {
3313         struct map *map = machine__kernel_map(machine);
3314         int err = 0;
3315
3316         /*
3317          * The only addresses above 2^63 are kernel addresses of a 64-bit
3318          * kernel.  Note that addresses are unsigned so that on a 32-bit system
3319          * all addresses including kernel addresses are less than 2^32.  In
3320          * that case (32-bit system), if the kernel mapping is unknown, all
3321          * addresses will be assumed to be in user space - see
3322          * machine__kernel_ip().
3323          */
3324         machine->kernel_start = 1ULL << 63;
3325         if (map) {
3326                 err = map__load(map);
3327                 /*
3328                  * On x86_64, PTI entry trampolines are less than the
3329                  * start of kernel text, but still above 2^63. So leave
3330                  * kernel_start = 1ULL << 63 for x86_64.
3331                  */
3332                 if (!err && !machine__is(machine, "x86_64"))
3333                         machine->kernel_start = map__start(map);
3334         }
3335         return err;
3336 }
3337
3338 u8 machine__addr_cpumode(struct machine *machine, u8 cpumode, u64 addr)
3339 {
3340         u8 addr_cpumode = cpumode;
3341         bool kernel_ip;
3342
3343         if (!machine->single_address_space)
3344                 goto out;
3345
3346         kernel_ip = machine__kernel_ip(machine, addr);
3347         switch (cpumode) {
3348         case PERF_RECORD_MISC_KERNEL:
3349         case PERF_RECORD_MISC_USER:
3350                 addr_cpumode = kernel_ip ? PERF_RECORD_MISC_KERNEL :
3351                                            PERF_RECORD_MISC_USER;
3352                 break;
3353         case PERF_RECORD_MISC_GUEST_KERNEL:
3354         case PERF_RECORD_MISC_GUEST_USER:
3355                 addr_cpumode = kernel_ip ? PERF_RECORD_MISC_GUEST_KERNEL :
3356                                            PERF_RECORD_MISC_GUEST_USER;
3357                 break;
3358         default:
3359                 break;
3360         }
3361 out:
3362         return addr_cpumode;
3363 }
3364
3365 struct dso *machine__findnew_dso_id(struct machine *machine, const char *filename, struct dso_id *id)
3366 {
3367         return dsos__findnew_id(&machine->dsos, filename, id);
3368 }
3369
3370 struct dso *machine__findnew_dso(struct machine *machine, const char *filename)
3371 {
3372         return machine__findnew_dso_id(machine, filename, NULL);
3373 }
3374
3375 char *machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
3376 {
3377         struct machine *machine = vmachine;
3378         struct map *map;
3379         struct symbol *sym = machine__find_kernel_symbol(machine, *addrp, &map);
3380
3381         if (sym == NULL)
3382                 return NULL;
3383
3384         *modp = __map__is_kmodule(map) ? (char *)map__dso(map)->short_name : NULL;
3385         *addrp = map__unmap_ip(map, sym->start);
3386         return sym->name;
3387 }
3388
3389 int machine__for_each_dso(struct machine *machine, machine__dso_t fn, void *priv)
3390 {
3391         struct dso *pos;
3392         int err = 0;
3393
3394         list_for_each_entry(pos, &machine->dsos.head, node) {
3395                 if (fn(pos, machine, priv))
3396                         err = -1;
3397         }
3398         return err;
3399 }
3400
3401 int machine__for_each_kernel_map(struct machine *machine, machine__map_t fn, void *priv)
3402 {
3403         struct maps *maps = machine__kernel_maps(machine);
3404         struct map_rb_node *pos;
3405         int err = 0;
3406
3407         maps__for_each_entry(maps, pos) {
3408                 err = fn(pos->map, priv);
3409                 if (err != 0) {
3410                         break;
3411                 }
3412         }
3413         return err;
3414 }
3415
3416 bool machine__is_lock_function(struct machine *machine, u64 addr)
3417 {
3418         if (!machine->sched.text_start) {
3419                 struct map *kmap;
3420                 struct symbol *sym = machine__find_kernel_symbol_by_name(machine, "__sched_text_start", &kmap);
3421
3422                 if (!sym) {
3423                         /* to avoid retry */
3424                         machine->sched.text_start = 1;
3425                         return false;
3426                 }
3427
3428                 machine->sched.text_start = map__unmap_ip(kmap, sym->start);
3429
3430                 /* should not fail from here */
3431                 sym = machine__find_kernel_symbol_by_name(machine, "__sched_text_end", &kmap);
3432                 machine->sched.text_end = map__unmap_ip(kmap, sym->start);
3433
3434                 sym = machine__find_kernel_symbol_by_name(machine, "__lock_text_start", &kmap);
3435                 machine->lock.text_start = map__unmap_ip(kmap, sym->start);
3436
3437                 sym = machine__find_kernel_symbol_by_name(machine, "__lock_text_end", &kmap);
3438                 machine->lock.text_end = map__unmap_ip(kmap, sym->start);
3439         }
3440
3441         /* failed to get kernel symbols */
3442         if (machine->sched.text_start == 1)
3443                 return false;
3444
3445         /* mutex and rwsem functions are in sched text section */
3446         if (machine->sched.text_start <= addr && addr < machine->sched.text_end)
3447                 return true;
3448
3449         /* spinlock functions are in lock text section */
3450         if (machine->lock.text_start <= addr && addr < machine->lock.text_end)
3451                 return true;
3452
3453         return false;
3454 }