machine.c

   1// SPDX-License-Identifier: GPL-2.0
   2#include <dirent.h>
   3#include <errno.h>
   4#include <inttypes.h>
   5#include <regex.h>
   6#include <stdlib.h>
   7#include "callchain.h"
   8#include "debug.h"
   9#include "dso.h"
  10#include "env.h"
  11#include "event.h"
  12#include "evsel.h"
  13#include "hist.h"
  14#include "machine.h"
  15#include "map.h"
  16#include "map_symbol.h"
  17#include "branch.h"
  18#include "mem-events.h"
  19#include "mem-info.h"
  20#include "path.h"
  21#include "srcline.h"
  22#include "symbol.h"
  23#include "synthetic-events.h"
  24#include "sort.h"
  25#include "strlist.h"
  26#include "target.h"
  27#include "thread.h"
  28#include "util.h"
  29#include "vdso.h"
  30#include <stdbool.h>
  31#include <sys/types.h>
  32#include <sys/stat.h>
  33#include <unistd.h>
  34#include "unwind.h"
  35#include "linux/hash.h"
  36#include "asm/bug.h"
  37#include "bpf-event.h"
  38#include <internal/lib.h> // page_size
  39#include "cgroup.h"
  40#include "arm64-frame-pointer-unwind-support.h"
  41#include <api/io_dir.h>
  42
  43#include <linux/ctype.h>
  44#include <symbol/kallsyms.h>
  45#include <linux/mman.h>
  46#include <linux/string.h>
  47#include <linux/zalloc.h>
  48
  49static struct dso *machine__kernel_dso(struct machine *machine)
  50{
  51	return map__dso(machine->vmlinux_map);
  52}
  53
  54static int machine__set_mmap_name(struct machine *machine)
  55{
  56	if (machine__is_host(machine))
  57		machine->mmap_name = strdup("[kernel.kallsyms]");
  58	else if (machine__is_default_guest(machine))
  59		machine->mmap_name = strdup("[guest.kernel.kallsyms]");
  60	else if (asprintf(&machine->mmap_name, "[guest.kernel.kallsyms.%d]",
  61			  machine->pid) < 0)
  62		machine->mmap_name = NULL;
  63
  64	return machine->mmap_name ? 0 : -ENOMEM;
  65}
  66
  67static void thread__set_guest_comm(struct thread *thread, pid_t pid)
  68{
  69	char comm[64];
  70
  71	snprintf(comm, sizeof(comm), "[guest/%d]", pid);
  72	thread__set_comm(thread, comm, 0);
  73}
  74
  75int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
  76{
  77	int err = -ENOMEM;
  78
  79	memset(machine, 0, sizeof(*machine));
  80	machine->kmaps = maps__new(machine);
  81	if (machine->kmaps == NULL)
  82		return -ENOMEM;
  83
  84	RB_CLEAR_NODE(&machine->rb_node);
  85	dsos__init(&machine->dsos);
  86
  87	threads__init(&machine->threads);
  88
  89	machine->vdso_info = NULL;
  90	machine->env = NULL;
  91
  92	machine->pid = pid;
  93
  94	machine->id_hdr_size = 0;
  95	machine->kptr_restrict_warned = false;
  96	machine->comm_exec = false;
  97	machine->kernel_start = 0;
  98	machine->vmlinux_map = NULL;
  99	/* There is no initial context switch in, so we start at 1. */
 100	machine->parallelism = 1;
 101
 102	machine->root_dir = strdup(root_dir);
 103	if (machine->root_dir == NULL)
 104		goto out;
 105
 106	if (machine__set_mmap_name(machine))
 107		goto out;
 108
 109	if (pid != HOST_KERNEL_ID) {
 110		struct thread *thread = machine__findnew_thread(machine, -1,
 111								pid);
 112
 113		if (thread == NULL)
 114			goto out;
 115
 116		thread__set_guest_comm(thread, pid);
 117		thread__put(thread);
 118	}
 119
 120	machine->current_tid = NULL;
 121	err = 0;
 122
 123out:
 124	if (err) {
 125		zfree(&machine->kmaps);
 126		zfree(&machine->root_dir);
 127		zfree(&machine->mmap_name);
 128	}
 129	return 0;
 130}
 131
 132static struct machine *__machine__new_host(struct perf_env *host_env, bool kernel_maps)
 133{
 134	struct machine *machine = malloc(sizeof(*machine));
 135
 136	if (!machine)
 137		return NULL;
 138
 139	machine__init(machine, "", HOST_KERNEL_ID);
 140
 141	if (kernel_maps && machine__create_kernel_maps(machine) < 0) {
 142		free(machine);
 143		return NULL;
 144	}
 145	machine->env = host_env;
 146	return machine;
 147}
 148
 149struct machine *machine__new_host(struct perf_env *host_env)
 150{
 151	return __machine__new_host(host_env, /*kernel_maps=*/true);
 152}
 153
 154static int mmap_handler(const struct perf_tool *tool __maybe_unused,
 155			union perf_event *event,
 156			struct perf_sample *sample,
 157			struct machine *machine)
 158{
 159	return machine__process_mmap2_event(machine, event, sample);
 160}
 161
 162static int machine__init_live(struct machine *machine, pid_t pid)
 163{
 164	union perf_event event;
 165
 166	memset(&event, 0, sizeof(event));
 167	return perf_event__synthesize_mmap_events(NULL, &event, pid, pid,
 168						  mmap_handler, machine, true);
 169}
 170
 171struct machine *machine__new_live(struct perf_env *host_env, bool kernel_maps, pid_t pid)
 172{
 173	struct machine *machine = __machine__new_host(host_env, kernel_maps);
 174
 175	if (!machine)
 176		return NULL;
 177
 178	if (machine__init_live(machine, pid)) {
 179		machine__delete(machine);
 180		return NULL;
 181	}
 182	return machine;
 183}
 184
 185struct machine *machine__new_kallsyms(struct perf_env *host_env)
 186{
 187	struct machine *machine = machine__new_host(host_env);
 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
 202void machine__delete_threads(struct machine *machine)
 203{
 204	threads__remove_all_threads(&machine->threads);
 205}
 206
 207void machine__exit(struct machine *machine)
 208{
 209	if (machine == NULL)
 210		return;
 211
 212	machine__destroy_kernel_maps(machine);
 213	maps__zput(machine->kmaps);
 214	dsos__exit(&machine->dsos);
 215	machine__exit_vdso(machine);
 216	zfree(&machine->root_dir);
 217	zfree(&machine->mmap_name);
 218	zfree(&machine->current_tid);
 219	zfree(&machine->kallsyms_filename);
 220
 221	threads__exit(&machine->threads);
 222}
 223
 224void machine__delete(struct machine *machine)
 225{
 226	if (machine) {
 227		machine__exit(machine);
 228		free(machine);
 229	}
 230}
 231
 232void machines__init(struct machines *machines)
 233{
 234	machine__init(&machines->host, "", HOST_KERNEL_ID);
 235	machines->guests = RB_ROOT_CACHED;
 236}
 237
 238void machines__exit(struct machines *machines)
 239{
 240	machine__exit(&machines->host);
 241	/* XXX exit guest */
 242}
 243
 244struct machine *machines__add(struct machines *machines, pid_t pid,
 245			      const char *root_dir)
 246{
 247	struct rb_node **p = &machines->guests.rb_root.rb_node;
 248	struct rb_node *parent = NULL;
 249	struct machine *pos, *machine = malloc(sizeof(*machine));
 250	bool leftmost = true;
 251
 252	if (machine == NULL)
 253		return NULL;
 254
 255	if (machine__init(machine, root_dir, pid) != 0) {
 256		free(machine);
 257		return NULL;
 258	}
 259
 260	while (*p != NULL) {
 261		parent = *p;
 262		pos = rb_entry(parent, struct machine, rb_node);
 263		if (pid < pos->pid)
 264			p = &(*p)->rb_left;
 265		else {
 266			p = &(*p)->rb_right;
 267			leftmost = false;
 268		}
 269	}
 270
 271	rb_link_node(&machine->rb_node, parent, p);
 272	rb_insert_color_cached(&machine->rb_node, &machines->guests, leftmost);
 273
 274	machine->machines = machines;
 275
 276	return machine;
 277}
 278
 279void machines__set_comm_exec(struct machines *machines, bool comm_exec)
 280{
 281	struct rb_node *nd;
 282
 283	machines->host.comm_exec = comm_exec;
 284
 285	for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
 286		struct machine *machine = rb_entry(nd, struct machine, rb_node);
 287
 288		machine->comm_exec = comm_exec;
 289	}
 290}
 291
 292struct machine *machines__find(struct machines *machines, pid_t pid)
 293{
 294	struct rb_node **p = &machines->guests.rb_root.rb_node;
 295	struct rb_node *parent = NULL;
 296	struct machine *machine;
 297	struct machine *default_machine = NULL;
 298
 299	if (pid == HOST_KERNEL_ID)
 300		return &machines->host;
 301
 302	while (*p != NULL) {
 303		parent = *p;
 304		machine = rb_entry(parent, struct machine, rb_node);
 305		if (pid < machine->pid)
 306			p = &(*p)->rb_left;
 307		else if (pid > machine->pid)
 308			p = &(*p)->rb_right;
 309		else
 310			return machine;
 311		if (!machine->pid)
 312			default_machine = machine;
 313	}
 314
 315	return default_machine;
 316}
 317
 318struct machine *machines__findnew(struct machines *machines, pid_t pid)
 319{
 320	char path[PATH_MAX];
 321	const char *root_dir = "";
 322	struct machine *machine = machines__find(machines, pid);
 323
 324	if (machine && (machine->pid == pid))
 325		goto out;
 326
 327	if ((pid != HOST_KERNEL_ID) &&
 328	    (pid != DEFAULT_GUEST_KERNEL_ID) &&
 329	    (symbol_conf.guestmount)) {
 330		sprintf(path, "%s/%d", symbol_conf.guestmount, pid);
 331		if (access(path, R_OK)) {
 332			static struct strlist *seen;
 333
 334			if (!seen)
 335				seen = strlist__new(NULL, NULL);
 336
 337			if (!strlist__has_entry(seen, path)) {
 338				pr_err("Can't access file %s\n", path);
 339				strlist__add(seen, path);
 340			}
 341			machine = NULL;
 342			goto out;
 343		}
 344		root_dir = path;
 345	}
 346
 347	machine = machines__add(machines, pid, root_dir);
 348out:
 349	return machine;
 350}
 351
 352struct machine *machines__find_guest(struct machines *machines, pid_t pid)
 353{
 354	struct machine *machine = machines__find(machines, pid);
 355
 356	if (!machine)
 357		machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
 358	return machine;
 359}
 360
 361/*
 362 * A common case for KVM test programs is that the test program acts as the
 363 * hypervisor, creating, running and destroying the virtual machine, and
 364 * providing the guest object code from its own object code. In this case,
 365 * the VM is not running an OS, but only the functions loaded into it by the
 366 * hypervisor test program, and conveniently, loaded at the same virtual
 367 * addresses.
 368 *
 369 * Normally to resolve addresses, MMAP events are needed to map addresses
 370 * back to the object code and debug symbols for that object code.
 371 *
 372 * Currently, there is no way to get such mapping information from guests
 373 * but, in the scenario described above, the guest has the same mappings
 374 * as the hypervisor, so support for that scenario can be achieved.
 375 *
 376 * To support that, copy the host thread's maps to the guest thread's maps.
 377 * Note, we do not discover the guest until we encounter a guest event,
 378 * which works well because it is not until then that we know that the host
 379 * thread's maps have been set up.
 380 *
 381 * This function returns the guest thread. Apart from keeping the data
 382 * structures sane, using a thread belonging to the guest machine, instead
 383 * of the host thread, allows it to have its own comm (refer
 384 * thread__set_guest_comm()).
 385 */
 386static struct thread *findnew_guest_code(struct machine *machine,
 387					 struct machine *host_machine,
 388					 pid_t pid)
 389{
 390	struct thread *host_thread;
 391	struct thread *thread;
 392	int err;
 393
 394	if (!machine)
 395		return NULL;
 396
 397	thread = machine__findnew_thread(machine, -1, pid);
 398	if (!thread)
 399		return NULL;
 400
 401	/* Assume maps are set up if there are any */
 402	if (!maps__empty(thread__maps(thread)))
 403		return thread;
 404
 405	host_thread = machine__find_thread(host_machine, -1, pid);
 406	if (!host_thread)
 407		goto out_err;
 408
 409	thread__set_guest_comm(thread, pid);
 410
 411	/*
 412	 * Guest code can be found in hypervisor process at the same address
 413	 * so copy host maps.
 414	 */
 415	err = maps__copy_from(thread__maps(thread), thread__maps(host_thread));
 416	thread__put(host_thread);
 417	if (err)
 418		goto out_err;
 419
 420	return thread;
 421
 422out_err:
 423	thread__zput(thread);
 424	return NULL;
 425}
 426
 427struct thread *machines__findnew_guest_code(struct machines *machines, pid_t pid)
 428{
 429	struct machine *host_machine = machines__find(machines, HOST_KERNEL_ID);
 430	struct machine *machine = machines__findnew(machines, pid);
 431
 432	return findnew_guest_code(machine, host_machine, pid);
 433}
 434
 435struct thread *machine__findnew_guest_code(struct machine *machine, pid_t pid)
 436{
 437	struct machines *machines = machine->machines;
 438	struct machine *host_machine;
 439
 440	if (!machines)
 441		return NULL;
 442
 443	host_machine = machines__find(machines, HOST_KERNEL_ID);
 444
 445	return findnew_guest_code(machine, host_machine, pid);
 446}
 447
 448void machines__process_guests(struct machines *machines,
 449			      machine__process_t process, void *data)
 450{
 451	struct rb_node *nd;
 452
 453	for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
 454		struct machine *pos = rb_entry(nd, struct machine, rb_node);
 455		process(pos, data);
 456	}
 457}
 458
 459void machines__set_id_hdr_size(struct machines *machines, u16 id_hdr_size)
 460{
 461	struct rb_node *node;
 462	struct machine *machine;
 463
 464	machines->host.id_hdr_size = id_hdr_size;
 465
 466	for (node = rb_first_cached(&machines->guests); node;
 467	     node = rb_next(node)) {
 468		machine = rb_entry(node, struct machine, rb_node);
 469		machine->id_hdr_size = id_hdr_size;
 470	}
 471
 472	return;
 473}
 474
 475static void machine__update_thread_pid(struct machine *machine,
 476				       struct thread *th, pid_t pid)
 477{
 478	struct thread *leader;
 479
 480	if (pid == thread__pid(th) || pid == -1 || thread__pid(th) != -1)
 481		return;
 482
 483	thread__set_pid(th, pid);
 484
 485	if (thread__pid(th) == thread__tid(th))
 486		return;
 487
 488	leader = machine__findnew_thread(machine, thread__pid(th), thread__pid(th));
 489	if (!leader)
 490		goto out_err;
 491
 492	if (!thread__maps(leader))
 493		thread__set_maps(leader, maps__new(machine));
 494
 495	if (!thread__maps(leader))
 496		goto out_err;
 497
 498	if (thread__maps(th) == thread__maps(leader))
 499		goto out_put;
 500
 501	if (thread__maps(th)) {
 502		/*
 503		 * Maps are created from MMAP events which provide the pid and
 504		 * tid.  Consequently there never should be any maps on a thread
 505		 * with an unknown pid.  Just print an error if there are.
 506		 */
 507		if (!maps__empty(thread__maps(th)))
 508			pr_err("Discarding thread maps for %d:%d\n",
 509				thread__pid(th), thread__tid(th));
 510		maps__put(thread__maps(th));
 511	}
 512
 513	thread__set_maps(th, maps__get(thread__maps(leader)));
 514out_put:
 515	thread__put(leader);
 516	return;
 517out_err:
 518	pr_err("Failed to join map groups for %d:%d\n", thread__pid(th), thread__tid(th));
 519	goto out_put;
 520}
 521
 522/*
 523 * Caller must eventually drop thread->refcnt returned with a successful
 524 * lookup/new thread inserted.
 525 */
 526static struct thread *__machine__findnew_thread(struct machine *machine,
 527						pid_t pid,
 528						pid_t tid,
 529						bool create)
 530{
 531	struct thread *th = threads__find(&machine->threads, tid);
 532	bool created;
 533
 534	if (th) {
 535		machine__update_thread_pid(machine, th, pid);
 536		return th;
 537	}
 538	if (!create)
 539		return NULL;
 540
 541	th = threads__findnew(&machine->threads, pid, tid, &created);
 542	if (created) {
 543		/*
 544		 * We have to initialize maps separately after rb tree is
 545		 * updated.
 546		 *
 547		 * The reason is that we call machine__findnew_thread within
 548		 * thread__init_maps to find the thread leader and that would
 549		 * screwed the rb tree.
 550		 */
 551		if (thread__init_maps(th, machine)) {
 552			pr_err("Thread init failed thread %d\n", pid);
 553			threads__remove(&machine->threads, th);
 554			thread__put(th);
 555			return NULL;
 556		}
 557	} else
 558		machine__update_thread_pid(machine, th, pid);
 559
 560	return th;
 561}
 562
 563struct thread *machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid)
 564{
 565	return __machine__findnew_thread(machine, pid, tid, /*create=*/true);
 566}
 567
 568struct thread *machine__find_thread(struct machine *machine, pid_t pid,
 569				    pid_t tid)
 570{
 571	return __machine__findnew_thread(machine, pid, tid, /*create=*/false);
 572}
 573
 574/*
 575 * Threads are identified by pid and tid, and the idle task has pid == tid == 0.
 576 * So here a single thread is created for that, but actually there is a separate
 577 * idle task per cpu, so there should be one 'struct thread' per cpu, but there
 578 * is only 1. That causes problems for some tools, requiring workarounds. For
 579 * example get_idle_thread() in builtin-sched.c, or thread_stack__per_cpu().
 580 */
 581struct thread *machine__idle_thread(struct machine *machine)
 582{
 583	struct thread *thread = machine__findnew_thread(machine, 0, 0);
 584
 585	if (!thread || thread__set_comm(thread, "swapper", 0) ||
 586	    thread__set_namespaces(thread, 0, NULL))
 587		pr_err("problem inserting idle task for machine pid %d\n", machine->pid);
 588
 589	return thread;
 590}
 591
 592struct comm *machine__thread_exec_comm(struct machine *machine,
 593				       struct thread *thread)
 594{
 595	if (machine->comm_exec)
 596		return thread__exec_comm(thread);
 597	else
 598		return thread__comm(thread);
 599}
 600
 601int machine__process_comm_event(struct machine *machine, union perf_event *event,
 602				struct perf_sample *sample)
 603{
 604	struct thread *thread = machine__findnew_thread(machine,
 605							event->comm.pid,
 606							event->comm.tid);
 607	bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
 608	int err = 0;
 609
 610	if (exec)
 611		machine->comm_exec = true;
 612
 613	if (dump_trace)
 614		perf_event__fprintf_comm(event, stdout);
 615
 616	if (thread == NULL ||
 617	    __thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
 618		dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
 619		err = -1;
 620	}
 621
 622	thread__put(thread);
 623
 624	return err;
 625}
 626
 627int machine__process_namespaces_event(struct machine *machine __maybe_unused,
 628				      union perf_event *event,
 629				      struct perf_sample *sample __maybe_unused)
 630{
 631	struct thread *thread = machine__findnew_thread(machine,
 632							event->namespaces.pid,
 633							event->namespaces.tid);
 634	int err = 0;
 635
 636	WARN_ONCE(event->namespaces.nr_namespaces > NR_NAMESPACES,
 637		  "\nWARNING: kernel seems to support more namespaces than perf"
 638		  " tool.\nTry updating the perf tool..\n\n");
 639
 640	WARN_ONCE(event->namespaces.nr_namespaces < NR_NAMESPACES,
 641		  "\nWARNING: perf tool seems to support more namespaces than"
 642		  " the kernel.\nTry updating the kernel..\n\n");
 643
 644	if (dump_trace)
 645		perf_event__fprintf_namespaces(event, stdout);
 646
 647	if (thread == NULL ||
 648	    thread__set_namespaces(thread, sample->time, &event->namespaces)) {
 649		dump_printf("problem processing PERF_RECORD_NAMESPACES, skipping event.\n");
 650		err = -1;
 651	}
 652
 653	thread__put(thread);
 654
 655	return err;
 656}
 657
 658int machine__process_cgroup_event(struct machine *machine,
 659				  union perf_event *event,
 660				  struct perf_sample *sample __maybe_unused)
 661{
 662	struct cgroup *cgrp;
 663
 664	if (dump_trace)
 665		perf_event__fprintf_cgroup(event, stdout);
 666
 667	cgrp = cgroup__findnew(machine->env, event->cgroup.id, event->cgroup.path);
 668	if (cgrp == NULL)
 669		return -ENOMEM;
 670
 671	return 0;
 672}
 673
 674int machine__process_lost_event(struct machine *machine __maybe_unused,
 675				union perf_event *event, struct perf_sample *sample __maybe_unused)
 676{
 677	dump_printf(": id:%" PRI_lu64 ": lost:%" PRI_lu64 "\n",
 678		    event->lost.id, event->lost.lost);
 679	return 0;
 680}
 681
 682int machine__process_lost_samples_event(struct machine *machine __maybe_unused,
 683					union perf_event *event, struct perf_sample *sample)
 684{
 685	dump_printf(": id:%" PRIu64 ": lost samples :%" PRI_lu64 "%s\n",
 686		    sample->id, event->lost_samples.lost,
 687		    event->header.misc & PERF_RECORD_MISC_LOST_SAMPLES_BPF ? " (BPF)" : "");
 688	return 0;
 689}
 690
 691int machine__process_aux_event(struct machine *machine __maybe_unused,
 692			       union perf_event *event)
 693{
 694	if (dump_trace)
 695		perf_event__fprintf_aux(event, stdout);
 696	return 0;
 697}
 698
 699int machine__process_itrace_start_event(struct machine *machine __maybe_unused,
 700					union perf_event *event)
 701{
 702	if (dump_trace)
 703		perf_event__fprintf_itrace_start(event, stdout);
 704	return 0;
 705}
 706
 707int machine__process_aux_output_hw_id_event(struct machine *machine __maybe_unused,
 708					    union perf_event *event)
 709{
 710	if (dump_trace)
 711		perf_event__fprintf_aux_output_hw_id(event, stdout);
 712	return 0;
 713}
 714
 715int machine__process_switch_event(struct machine *machine __maybe_unused,
 716				  union perf_event *event)
 717{
 718	bool out = event->header.misc & PERF_RECORD_MISC_SWITCH_OUT;
 719
 720	if (dump_trace)
 721		perf_event__fprintf_switch(event, stdout);
 722	machine->parallelism += out ? -1 : 1;
 723	return 0;
 724}
 725
 726static int machine__process_ksymbol_register(struct machine *machine,
 727					     union perf_event *event,
 728					     struct perf_sample *sample __maybe_unused)
 729{
 730	struct symbol *sym;
 731	struct dso *dso = NULL;
 732	struct map *map = maps__find(machine__kernel_maps(machine), event->ksymbol.addr);
 733	int err = 0;
 734
 735	if (!map) {
 736		dso = dso__new(event->ksymbol.name);
 737
 738		if (!dso) {
 739			err = -ENOMEM;
 740			goto out;
 741		}
 742		dso__set_kernel(dso, DSO_SPACE__KERNEL);
 743		map = map__new2(0, dso);
 744		if (!map) {
 745			err = -ENOMEM;
 746			goto out;
 747		}
 748		if (event->ksymbol.ksym_type == PERF_RECORD_KSYMBOL_TYPE_OOL) {
 749			dso__set_binary_type(dso, DSO_BINARY_TYPE__OOL);
 750			dso__data(dso)->file_size = event->ksymbol.len;
 751			dso__set_loaded(dso);
 752		}
 753
 754		map__set_start(map, event->ksymbol.addr);
 755		map__set_end(map, map__start(map) + event->ksymbol.len);
 756		err = maps__fixup_overlap_and_insert(machine__kernel_maps(machine), map);
 757		if (err) {
 758			err = -ENOMEM;
 759			goto out;
 760		}
 761
 762		dso__set_loaded(dso);
 763
 764		if (is_bpf_image(event->ksymbol.name)) {
 765			dso__set_binary_type(dso, DSO_BINARY_TYPE__BPF_IMAGE);
 766			dso__set_long_name(dso, "", false);
 767		}
 768	} else {
 769		dso = dso__get(map__dso(map));
 770	}
 771
 772	sym = symbol__new(map__map_ip(map, map__start(map)),
 773			  event->ksymbol.len,
 774			  0, 0, event->ksymbol.name);
 775	if (!sym) {
 776		err = -ENOMEM;
 777		goto out;
 778	}
 779	dso__insert_symbol(dso, sym);
 780out:
 781	map__put(map);
 782	dso__put(dso);
 783	return err;
 784}
 785
 786static int machine__process_ksymbol_unregister(struct machine *machine,
 787					       union perf_event *event,
 788					       struct perf_sample *sample __maybe_unused)
 789{
 790	struct symbol *sym;
 791	struct map *map;
 792
 793	map = maps__find(machine__kernel_maps(machine), event->ksymbol.addr);
 794	if (!map)
 795		return 0;
 796
 797	if (!RC_CHK_EQUAL(map, machine->vmlinux_map))
 798		maps__remove(machine__kernel_maps(machine), map);
 799	else {
 800		struct dso *dso = map__dso(map);
 801
 802		sym = dso__find_symbol(dso, map__map_ip(map, map__start(map)));
 803		if (sym)
 804			dso__delete_symbol(dso, sym);
 805	}
 806	map__put(map);
 807	return 0;
 808}
 809
 810int machine__process_ksymbol(struct machine *machine __maybe_unused,
 811			     union perf_event *event,
 812			     struct perf_sample *sample)
 813{
 814	if (dump_trace)
 815		perf_event__fprintf_ksymbol(event, stdout);
 816
 817	/* no need to process non-JIT BPF as it cannot get samples */
 818	if (event->ksymbol.len == 0)
 819		return 0;
 820
 821	if (event->ksymbol.flags & PERF_RECORD_KSYMBOL_FLAGS_UNREGISTER)
 822		return machine__process_ksymbol_unregister(machine, event,
 823							   sample);
 824	return machine__process_ksymbol_register(machine, event, sample);
 825}
 826
 827int machine__process_text_poke(struct machine *machine, union perf_event *event,
 828			       struct perf_sample *sample __maybe_unused)
 829{
 830	struct map *map = maps__find(machine__kernel_maps(machine), event->text_poke.addr);
 831	u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
 832	struct dso *dso = map ? map__dso(map) : NULL;
 833
 834	if (dump_trace)
 835		perf_event__fprintf_text_poke(event, machine, stdout);
 836
 837	if (!event->text_poke.new_len)
 838		goto out;
 839
 840	if (cpumode != PERF_RECORD_MISC_KERNEL) {
 841		pr_debug("%s: unsupported cpumode - ignoring\n", __func__);
 842		goto out;
 843	}
 844
 845	if (dso) {
 846		u8 *new_bytes = event->text_poke.bytes + event->text_poke.old_len;
 847		int ret;
 848
 849		/*
 850		 * Kernel maps might be changed when loading symbols so loading
 851		 * must be done prior to using kernel maps.
 852		 */
 853		map__load(map);
 854		ret = dso__data_write_cache_addr(dso, map, machine,
 855						 event->text_poke.addr,
 856						 new_bytes,
 857						 event->text_poke.new_len);
 858		if (ret != event->text_poke.new_len)
 859			pr_debug("Failed to write kernel text poke at %#" PRI_lx64 "\n",
 860				 event->text_poke.addr);
 861	} else {
 862		pr_debug("Failed to find kernel text poke address map for %#" PRI_lx64 "\n",
 863			 event->text_poke.addr);
 864	}
 865out:
 866	map__put(map);
 867	return 0;
 868}
 869
 870static struct map *machine__addnew_module_map(struct machine *machine, u64 start,
 871					      const char *filename)
 872{
 873	struct map *map = NULL;
 874	struct kmod_path m;
 875	struct dso *dso;
 876	int err;
 877
 878	if (kmod_path__parse_name(&m, filename))
 879		return NULL;
 880
 881	dso = dsos__findnew_module_dso(&machine->dsos, machine, &m, filename);
 882	if (dso == NULL)
 883		goto out;
 884
 885	map = map__new2(start, dso);
 886	if (map == NULL)
 887		goto out;
 888
 889	err = maps__insert(machine__kernel_maps(machine), map);
 890	/* If maps__insert failed, return NULL. */
 891	if (err) {
 892		map__put(map);
 893		map = NULL;
 894	}
 895out:
 896	/* put the dso here, corresponding to  machine__findnew_module_dso */
 897	dso__put(dso);
 898	zfree(&m.name);
 899	return map;
 900}
 901
 902size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
 903{
 904	struct rb_node *nd;
 905	size_t ret = dsos__fprintf(&machines->host.dsos, fp);
 906
 907	for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
 908		struct machine *pos = rb_entry(nd, struct machine, rb_node);
 909		ret += dsos__fprintf(&pos->dsos, fp);
 910	}
 911
 912	return ret;
 913}
 914
 915size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
 916				     bool (skip)(struct dso *dso, int parm), int parm)
 917{
 918	return dsos__fprintf_buildid(&m->dsos, fp, skip, parm);
 919}
 920
 921size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
 922				     bool (skip)(struct dso *dso, int parm), int parm)
 923{
 924	struct rb_node *nd;
 925	size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
 926
 927	for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
 928		struct machine *pos = rb_entry(nd, struct machine, rb_node);
 929		ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
 930	}
 931	return ret;
 932}
 933
 934struct machine_fprintf_cb_args {
 935	FILE *fp;
 936	size_t printed;
 937};
 938
 939static int machine_fprintf_cb(struct thread *thread, void *data)
 940{
 941	struct machine_fprintf_cb_args *args = data;
 942
 943	/* TODO: handle fprintf errors. */
 944	args->printed += thread__fprintf(thread, args->fp);
 945	return 0;
 946}
 947
 948size_t machine__fprintf(struct machine *machine, FILE *fp)
 949{
 950	struct machine_fprintf_cb_args args = {
 951		.fp = fp,
 952		.printed = 0,
 953	};
 954	size_t ret = fprintf(fp, "Threads: %zu\n", threads__nr(&machine->threads));
 955
 956	machine__for_each_thread(machine, machine_fprintf_cb, &args);
 957	return ret + args.printed;
 958}
 959
 960static struct dso *machine__get_kernel(struct machine *machine)
 961{
 962	const char *vmlinux_name = machine->mmap_name;
 963	struct dso *kernel;
 964
 965	if (machine__is_host(machine)) {
 966		if (symbol_conf.vmlinux_name)
 967			vmlinux_name = symbol_conf.vmlinux_name;
 968
 969		kernel = machine__findnew_kernel(machine, vmlinux_name,
 970						 "[kernel]", DSO_SPACE__KERNEL);
 971	} else {
 972		if (symbol_conf.default_guest_vmlinux_name)
 973			vmlinux_name = symbol_conf.default_guest_vmlinux_name;
 974
 975		kernel = machine__findnew_kernel(machine, vmlinux_name,
 976						 "[guest.kernel]",
 977						 DSO_SPACE__KERNEL_GUEST);
 978	}
 979
 980	if (kernel != NULL && (!dso__has_build_id(kernel)))
 981		dso__read_running_kernel_build_id(kernel, machine);
 982
 983	return kernel;
 984}
 985
 986void machine__get_kallsyms_filename(struct machine *machine, char *buf,
 987				    size_t bufsz)
 988{
 989	if (machine__is_default_guest(machine))
 990		scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
 991	else
 992		scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
 993}
 994
 995const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
 996
 997/* Figure out the start address of kernel map from /proc/kallsyms.
 998 * Returns the name of the start symbol in *symbol_name. Pass in NULL as
 999 * symbol_name if it's not that important.
1000 */
1001static int machine__get_running_kernel_start(struct machine *machine,
1002					     const char **symbol_name,
1003					     u64 *start, u64 *end)
1004{
1005	char filename[PATH_MAX];
1006	int i, err = -1;
1007	const char *name;
1008	u64 addr = 0;
1009
1010	machine__get_kallsyms_filename(machine, filename, PATH_MAX);
1011
1012	if (symbol__restricted_filename(filename, "/proc/kallsyms"))
1013		return 0;
1014
1015	for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
1016		err = kallsyms__get_function_start(filename, name, &addr);
1017		if (!err)
1018			break;
1019	}
1020
1021	if (err)
1022		return -1;
1023
1024	if (symbol_name)
1025		*symbol_name = name;
1026
1027	*start = addr;
1028
1029	err = kallsyms__get_symbol_start(filename, "_edata", &addr);
1030	if (err)
1031		err = kallsyms__get_symbol_start(filename, "_etext", &addr);
1032	if (!err)
1033		*end = addr;
1034
1035	return 0;
1036}
1037
1038int machine__create_extra_kernel_map(struct machine *machine,
1039				     struct dso *kernel,
1040				     struct extra_kernel_map *xm)
1041{
1042	struct kmap *kmap;
1043	struct map *map;
1044	int err;
1045
1046	map = map__new2(xm->start, kernel);
1047	if (!map)
1048		return -ENOMEM;
1049
1050	map__set_end(map, xm->end);
1051	map__set_pgoff(map, xm->pgoff);
1052
1053	kmap = map__kmap(map);
1054
1055	strlcpy(kmap->name, xm->name, KMAP_NAME_LEN);
1056
1057	err = maps__insert(machine__kernel_maps(machine), map);
1058
1059	if (!err) {
1060		pr_debug2("Added extra kernel map %s %" PRIx64 "-%" PRIx64 "\n",
1061			kmap->name, map__start(map), map__end(map));
1062	}
1063
1064	map__put(map);
1065
1066	return err;
1067}
1068
1069static u64 find_entry_trampoline(struct dso *dso)
1070{
1071	/* Duplicates are removed so lookup all aliases */
1072	const char *syms[] = {
1073		"_entry_trampoline",
1074		"__entry_trampoline_start",
1075		"entry_SYSCALL_64_trampoline",
1076	};
1077	struct symbol *sym = dso__first_symbol(dso);
1078	unsigned int i;
1079
1080	for (; sym; sym = dso__next_symbol(sym)) {
1081		if (sym->binding != STB_GLOBAL)
1082			continue;
1083		for (i = 0; i < ARRAY_SIZE(syms); i++) {
1084			if (!strcmp(sym->name, syms[i]))
1085				return sym->start;
1086		}
1087	}
1088
1089	return 0;
1090}
1091
1092/*
1093 * These values can be used for kernels that do not have symbols for the entry
1094 * trampolines in kallsyms.
1095 */
1096#define X86_64_CPU_ENTRY_AREA_PER_CPU	0xfffffe0000000000ULL
1097#define X86_64_CPU_ENTRY_AREA_SIZE	0x2c000
1098#define X86_64_ENTRY_TRAMPOLINE		0x6000
1099
1100struct machine__map_x86_64_entry_trampolines_args {
1101	struct maps *kmaps;
1102	bool found;
1103};
1104
1105static int machine__map_x86_64_entry_trampolines_cb(struct map *map, void *data)
1106{
1107	struct machine__map_x86_64_entry_trampolines_args *args = data;
1108	struct map *dest_map;
1109	struct kmap *kmap = __map__kmap(map);
1110
1111	if (!kmap || !is_entry_trampoline(kmap->name))
1112		return 0;
1113
1114	dest_map = maps__find(args->kmaps, map__pgoff(map));
1115	if (RC_CHK_ACCESS(dest_map) != RC_CHK_ACCESS(map))
1116		map__set_pgoff(map, map__map_ip(dest_map, map__pgoff(map)));
1117
1118	map__put(dest_map);
1119	args->found = true;
1120	return 0;
1121}
1122
1123/* Map x86_64 PTI entry trampolines */
1124int machine__map_x86_64_entry_trampolines(struct machine *machine,
1125					  struct dso *kernel)
1126{
1127	struct machine__map_x86_64_entry_trampolines_args args = {
1128		.kmaps = machine__kernel_maps(machine),
1129		.found = false,
1130	};
1131	int nr_cpus_avail, cpu;
1132	u64 pgoff;
1133
1134	/*
1135	 * In the vmlinux case, pgoff is a virtual address which must now be
1136	 * mapped to a vmlinux offset.
1137	 */
1138	maps__for_each_map(args.kmaps, machine__map_x86_64_entry_trampolines_cb, &args);
1139
1140	if (args.found || machine->trampolines_mapped)
1141		return 0;
1142
1143	pgoff = find_entry_trampoline(kernel);
1144	if (!pgoff)
1145		return 0;
1146
1147	nr_cpus_avail = machine__nr_cpus_avail(machine);
1148
1149	/* Add a 1 page map for each CPU's entry trampoline */
1150	for (cpu = 0; cpu < nr_cpus_avail; cpu++) {
1151		u64 va = X86_64_CPU_ENTRY_AREA_PER_CPU +
1152			 cpu * X86_64_CPU_ENTRY_AREA_SIZE +
1153			 X86_64_ENTRY_TRAMPOLINE;
1154		struct extra_kernel_map xm = {
1155			.start = va,
1156			.end   = va + page_size,
1157			.pgoff = pgoff,
1158		};
1159
1160		strlcpy(xm.name, ENTRY_TRAMPOLINE_NAME, KMAP_NAME_LEN);
1161
1162		if (machine__create_extra_kernel_map(machine, kernel, &xm) < 0)
1163			return -1;
1164	}
1165
1166	machine->trampolines_mapped = nr_cpus_avail;
1167
1168	return 0;
1169}
1170
1171int __weak machine__create_extra_kernel_maps(struct machine *machine __maybe_unused,
1172					     struct dso *kernel __maybe_unused)
1173{
1174	return 0;
1175}
1176
1177static int
1178__machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
1179{
1180	/* In case of renewal the kernel map, destroy previous one */
1181	machine__destroy_kernel_maps(machine);
1182
1183	map__put(machine->vmlinux_map);
1184	machine->vmlinux_map = map__new2(0, kernel);
1185	if (machine->vmlinux_map == NULL)
1186		return -ENOMEM;
1187
1188	map__set_mapping_type(machine->vmlinux_map, MAPPING_TYPE__IDENTITY);
1189	return maps__insert(machine__kernel_maps(machine), machine->vmlinux_map);
1190}
1191
1192void machine__destroy_kernel_maps(struct machine *machine)
1193{
1194	struct kmap *kmap;
1195	struct map *map = machine__kernel_map(machine);
1196
1197	if (map == NULL)
1198		return;
1199
1200	kmap = map__kmap(map);
1201	maps__remove(machine__kernel_maps(machine), map);
1202	if (kmap && kmap->ref_reloc_sym) {
1203		zfree((char **)&kmap->ref_reloc_sym->name);
1204		zfree(&kmap->ref_reloc_sym);
1205	}
1206
1207	map__zput(machine->vmlinux_map);
1208}
1209
1210int machines__create_guest_kernel_maps(struct machines *machines)
1211{
1212	int ret = 0;
1213	struct dirent **namelist = NULL;
1214	int i, items = 0;
1215	char path[PATH_MAX];
1216	pid_t pid;
1217	char *endp;
1218
1219	if (symbol_conf.default_guest_vmlinux_name ||
1220	    symbol_conf.default_guest_modules ||
1221	    symbol_conf.default_guest_kallsyms) {
1222		machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
1223	}
1224
1225	if (symbol_conf.guestmount) {
1226		items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
1227		if (items <= 0)
1228			return -ENOENT;
1229		for (i = 0; i < items; i++) {
1230			if (!isdigit(namelist[i]->d_name[0])) {
1231				/* Filter out . and .. */
1232				continue;
1233			}
1234			pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
1235			if ((*endp != '\0') ||
1236			    (endp == namelist[i]->d_name) ||
1237			    (errno == ERANGE)) {
1238				pr_debug("invalid directory (%s). Skipping.\n",
1239					 namelist[i]->d_name);
1240				continue;
1241			}
1242			sprintf(path, "%s/%s/proc/kallsyms",
1243				symbol_conf.guestmount,
1244				namelist[i]->d_name);
1245			ret = access(path, R_OK);
1246			if (ret) {
1247				pr_debug("Can't access file %s\n", path);
1248				goto failure;
1249			}
1250			machines__create_kernel_maps(machines, pid);
1251		}
1252failure:
1253		free(namelist);
1254	}
1255
1256	return ret;
1257}
1258
1259void machines__destroy_kernel_maps(struct machines *machines)
1260{
1261	struct rb_node *next = rb_first_cached(&machines->guests);
1262
1263	machine__destroy_kernel_maps(&machines->host);
1264
1265	while (next) {
1266		struct machine *pos = rb_entry(next, struct machine, rb_node);
1267
1268		next = rb_next(&pos->rb_node);
1269		rb_erase_cached(&pos->rb_node, &machines->guests);
1270		machine__delete(pos);
1271	}
1272}
1273
1274int machines__create_kernel_maps(struct machines *machines, pid_t pid)
1275{
1276	struct machine *machine = machines__findnew(machines, pid);
1277
1278	if (machine == NULL)
1279		return -1;
1280
1281	return machine__create_kernel_maps(machine);
1282}
1283
1284int machine__load_kallsyms(struct machine *machine, const char *filename)
1285{
1286	struct map *map = machine__kernel_map(machine);
1287	struct dso *dso = map__dso(map);
1288	int ret = __dso__load_kallsyms(dso, filename, map, true);
1289
1290	if (ret > 0) {
1291		dso__set_loaded(dso);
1292		/*
1293		 * Since /proc/kallsyms will have multiple sessions for the
1294		 * kernel, with modules between them, fixup the end of all
1295		 * sections.
1296		 */
1297		maps__fixup_end(machine__kernel_maps(machine));
1298	}
1299
1300	return ret;
1301}
1302
1303int machine__load_vmlinux_path(struct machine *machine)
1304{
1305	struct map *map = machine__kernel_map(machine);
1306	struct dso *dso = map__dso(map);
1307	int ret = dso__load_vmlinux_path(dso, map);
1308
1309	if (ret > 0)
1310		dso__set_loaded(dso);
1311
1312	return ret;
1313}
1314
1315static char *get_kernel_version(const char *root_dir)
1316{
1317	char version[PATH_MAX];
1318	FILE *file;
1319	char *name, *tmp;
1320	const char *prefix = "Linux version ";
1321
1322	sprintf(version, "%s/proc/version", root_dir);
1323	file = fopen(version, "r");
1324	if (!file)
1325		return NULL;
1326
1327	tmp = fgets(version, sizeof(version), file);
1328	fclose(file);
1329	if (!tmp)
1330		return NULL;
1331
1332	name = strstr(version, prefix);
1333	if (!name)
1334		return NULL;
1335	name += strlen(prefix);
1336	tmp = strchr(name, ' ');
1337	if (tmp)
1338		*tmp = '\0';
1339
1340	return strdup(name);
1341}
1342
1343static bool is_kmod_dso(struct dso *dso)
1344{
1345	return dso__symtab_type(dso) == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
1346	       dso__symtab_type(dso) == DSO_BINARY_TYPE__GUEST_KMODULE;
1347}
1348
1349static int maps__set_module_path(struct maps *maps, const char *path, struct kmod_path *m)
1350{
1351	char *long_name;
1352	struct dso *dso;
1353	struct map *map = maps__find_by_name(maps, m->name);
1354
1355	if (map == NULL)
1356		return 0;
1357
1358	long_name = strdup(path);
1359	if (long_name == NULL) {
1360		map__put(map);
1361		return -ENOMEM;
1362	}
1363
1364	dso = map__dso(map);
1365	dso__set_long_name(dso, long_name, true);
1366	dso__kernel_module_get_build_id(dso, "");
1367
1368	/*
1369	 * Full name could reveal us kmod compression, so
1370	 * we need to update the symtab_type if needed.
1371	 */
1372	if (m->comp && is_kmod_dso(dso)) {
1373		dso__set_symtab_type(dso, dso__symtab_type(dso)+1);
1374		dso__set_comp(dso, m->comp);
1375	}
1376	map__put(map);
1377	return 0;
1378}
1379
1380static int maps__set_modules_path_dir(struct maps *maps, char *path, size_t path_size, int depth)
1381{
1382	struct io_dirent64 *dent;
1383	struct io_dir iod;
1384	size_t root_len = strlen(path);
1385	int ret = 0;
1386
1387	io_dir__init(&iod, open(path, O_CLOEXEC | O_DIRECTORY | O_RDONLY));
1388	if (iod.dirfd < 0) {
1389		pr_debug("%s: cannot open %s dir\n", __func__, path);
1390		return -1;
1391	}
1392	/* Bounds check, should never happen. */
1393	if (root_len >= path_size)
1394		return -1;
1395	path[root_len++] = '/';
1396	while ((dent = io_dir__readdir(&iod)) != NULL) {
1397		if (io_dir__is_dir(&iod, dent)) {
1398			if (!strcmp(dent->d_name, ".") ||
1399			    !strcmp(dent->d_name, ".."))
1400				continue;
1401
1402			/* Do not follow top-level source and build symlinks */
1403			if (depth == 0) {
1404				if (!strcmp(dent->d_name, "source") ||
1405				    !strcmp(dent->d_name, "build"))
1406					continue;
1407			}
1408
1409			/* Bounds check, should never happen. */
1410			if (root_len + strlen(dent->d_name) >= path_size)
1411				continue;
1412
1413			strcpy(path + root_len, dent->d_name);
1414			ret = maps__set_modules_path_dir(maps, path, path_size, depth + 1);
1415			if (ret < 0)
1416				goto out;
1417		} else {
1418			struct kmod_path m;
1419
1420			ret = kmod_path__parse_name(&m, dent->d_name);
1421			if (ret)
1422				goto out;
1423
1424			if (m.kmod) {
1425				/* Bounds check, should never happen. */
1426				if (root_len + strlen(dent->d_name) < path_size) {
1427					strcpy(path + root_len, dent->d_name);
1428					ret = maps__set_module_path(maps, path, &m);
1429
1430				}
1431			}
1432			zfree(&m.name);
1433
1434			if (ret)
1435				goto out;
1436		}
1437	}
1438
1439out:
1440	close(iod.dirfd);
1441	return ret;
1442}
1443
1444static int machine__set_modules_path(struct machine *machine)
1445{
1446	char *version;
1447	char modules_path[PATH_MAX];
1448
1449	version = get_kernel_version(machine->root_dir);
1450	if (!version)
1451		return -1;
1452
1453	snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
1454		 machine->root_dir, version);
1455	free(version);
1456
1457	return maps__set_modules_path_dir(machine__kernel_maps(machine),
1458					  modules_path, sizeof(modules_path), 0);
1459}
1460int __weak arch__fix_module_text_start(u64 *start __maybe_unused,
1461				u64 *size __maybe_unused,
1462				const char *name __maybe_unused)
1463{
1464	return 0;
1465}
1466
1467static int machine__create_module(void *arg, const char *name, u64 start,
1468				  u64 size)
1469{
1470	struct machine *machine = arg;
1471	struct map *map;
1472
1473	if (arch__fix_module_text_start(&start, &size, name) < 0)
1474		return -1;
1475
1476	map = machine__addnew_module_map(machine, start, name);
1477	if (map == NULL)
1478		return -1;
1479	map__set_end(map, start + size);
1480
1481	dso__kernel_module_get_build_id(map__dso(map), machine->root_dir);
1482	map__put(map);
1483	return 0;
1484}
1485
1486static int machine__create_modules(struct machine *machine)
1487{
1488	const char *modules;
1489	char path[PATH_MAX];
1490
1491	if (machine__is_default_guest(machine)) {
1492		modules = symbol_conf.default_guest_modules;
1493	} else {
1494		snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
1495		modules = path;
1496	}
1497
1498	if (symbol__restricted_filename(modules, "/proc/modules"))
1499		return -1;
1500
1501	if (modules__parse(modules, machine, machine__create_module))
1502		return -1;
1503
1504	if (!machine__set_modules_path(machine))
1505		return 0;
1506
1507	pr_debug("Problems setting modules path maps, continuing anyway...\n");
1508
1509	return 0;
1510}
1511
1512static void machine__set_kernel_mmap(struct machine *machine,
1513				     u64 start, u64 end)
1514{
1515	map__set_start(machine->vmlinux_map, start);
1516	map__set_end(machine->vmlinux_map, end);
1517	/*
1518	 * Be a bit paranoid here, some perf.data file came with
1519	 * a zero sized synthesized MMAP event for the kernel.
1520	 */
1521	if (start == 0 && end == 0)
1522		map__set_end(machine->vmlinux_map, ~0ULL);
1523}
1524
1525static int machine__update_kernel_mmap(struct machine *machine,
1526				     u64 start, u64 end)
1527{
1528	struct map *orig, *updated;
1529	int err;
1530
1531	orig = machine->vmlinux_map;
1532	updated = map__get(orig);
1533
1534	machine->vmlinux_map = updated;
1535	maps__remove(machine__kernel_maps(machine), orig);
1536	machine__set_kernel_mmap(machine, start, end);
1537	err = maps__insert(machine__kernel_maps(machine), updated);
1538	map__put(orig);
1539
1540	return err;
1541}
1542
1543int machine__create_kernel_maps(struct machine *machine)
1544{
1545	struct dso *kernel = machine__get_kernel(machine);
1546	const char *name = NULL;
1547	u64 start = 0, end = ~0ULL;
1548	int ret;
1549
1550	if (kernel == NULL)
1551		return -1;
1552
1553	ret = __machine__create_kernel_maps(machine, kernel);
1554	if (ret < 0)
1555		goto out_put;
1556
1557	if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
1558		if (machine__is_host(machine))
1559			pr_debug("Problems creating module maps, "
1560				 "continuing anyway...\n");
1561		else
1562			pr_debug("Problems creating module maps for guest %d, "
1563				 "continuing anyway...\n", machine->pid);
1564	}
1565
1566	if (!machine__get_running_kernel_start(machine, &name, &start, &end)) {
1567		if (name &&
1568		    map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map, name, start)) {
1569			machine__destroy_kernel_maps(machine);
1570			ret = -1;
1571			goto out_put;
1572		}
1573
1574		/*
1575		 * we have a real start address now, so re-order the kmaps
1576		 * assume it's the last in the kmaps
1577		 */
1578		ret = machine__update_kernel_mmap(machine, start, end);
1579		if (ret < 0)
1580			goto out_put;
1581	}
1582
1583	if (machine__create_extra_kernel_maps(machine, kernel))
1584		pr_debug("Problems creating extra kernel maps, continuing anyway...\n");
1585
1586	if (end == ~0ULL) {
1587		/* update end address of the kernel map using adjacent module address */
1588		struct map *next = maps__find_next_entry(machine__kernel_maps(machine),
1589							 machine__kernel_map(machine));
1590
1591		if (next) {
1592			machine__set_kernel_mmap(machine, start, map__start(next));
1593			map__put(next);
1594		}
1595	}
1596
1597	maps__fixup_end(machine__kernel_maps(machine));
1598
1599out_put:
1600	dso__put(kernel);
1601	return ret;
1602}
1603
1604static int machine__uses_kcore_cb(struct dso *dso, void *data __maybe_unused)
1605{
1606	return dso__is_kcore(dso) ? 1 : 0;
1607}
1608
1609static bool machine__uses_kcore(struct machine *machine)
1610{
1611	return dsos__for_each_dso(&machine->dsos, machine__uses_kcore_cb, NULL) != 0 ? true : false;
1612}
1613
1614static bool perf_event__is_extra_kernel_mmap(struct machine *machine,
1615					     struct extra_kernel_map *xm)
1616{
1617	return machine__is(machine, "x86_64") &&
1618	       is_entry_trampoline(xm->name);
1619}
1620
1621static int machine__process_extra_kernel_map(struct machine *machine,
1622					     struct extra_kernel_map *xm)
1623{
1624	struct dso *kernel = machine__kernel_dso(machine);
1625
1626	if (kernel == NULL)
1627		return -1;
1628
1629	return machine__create_extra_kernel_map(machine, kernel, xm);
1630}
1631
1632static int machine__process_kernel_mmap_event(struct machine *machine,
1633					      struct extra_kernel_map *xm,
1634					      struct build_id *bid)
1635{
1636	enum dso_space_type dso_space;
1637	bool is_kernel_mmap;
1638	const char *mmap_name = machine->mmap_name;
1639
1640	/* If we have maps from kcore then we do not need or want any others */
1641	if (machine__uses_kcore(machine))
1642		return 0;
1643
1644	if (machine__is_host(machine))
1645		dso_space = DSO_SPACE__KERNEL;
1646	else
1647		dso_space = DSO_SPACE__KERNEL_GUEST;
1648
1649	is_kernel_mmap = memcmp(xm->name, mmap_name, strlen(mmap_name) - 1) == 0;
1650	if (!is_kernel_mmap && !machine__is_host(machine)) {
1651		/*
1652		 * If the event was recorded inside the guest and injected into
1653		 * the host perf.data file, then it will match a host mmap_name,
1654		 * so try that - see machine__set_mmap_name().
1655		 */
1656		mmap_name = "[kernel.kallsyms]";
1657		is_kernel_mmap = memcmp(xm->name, mmap_name, strlen(mmap_name) - 1) == 0;
1658	}
1659	if (xm->name[0] == '/' ||
1660	    (!is_kernel_mmap && xm->name[0] == '[')) {
1661		struct map *map = machine__addnew_module_map(machine, xm->start, xm->name);
1662
1663		if (map == NULL)
1664			goto out_problem;
1665
1666		map__set_end(map, map__start(map) + xm->end - xm->start);
1667
1668		if (build_id__is_defined(bid))
1669			dso__set_build_id(map__dso(map), bid);
1670
1671		map__put(map);
1672	} else if (is_kernel_mmap) {
1673		const char *symbol_name = xm->name + strlen(mmap_name);
1674		/*
1675		 * Should be there already, from the build-id table in
1676		 * the header.
1677		 */
1678		struct dso *kernel = dsos__find_kernel_dso(&machine->dsos);
1679
1680		if (kernel == NULL)
1681			kernel = machine__findnew_dso(machine, machine->mmap_name);
1682		if (kernel == NULL)
1683			goto out_problem;
1684
1685		dso__set_kernel(kernel, dso_space);
1686		if (__machine__create_kernel_maps(machine, kernel) < 0) {
1687			dso__put(kernel);
1688			goto out_problem;
1689		}
1690
1691		if (strstr(dso__long_name(kernel), "vmlinux"))
1692			dso__set_short_name(kernel, "[kernel.vmlinux]", false);
1693
1694		if (machine__update_kernel_mmap(machine, xm->start, xm->end) < 0) {
1695			dso__put(kernel);
1696			goto out_problem;
1697		}
1698
1699		if (build_id__is_defined(bid))
1700			dso__set_build_id(kernel, bid);
1701
1702		/*
1703		 * Avoid using a zero address (kptr_restrict) for the ref reloc
1704		 * symbol. Effectively having zero here means that at record
1705		 * time /proc/sys/kernel/kptr_restrict was non zero.
1706		 */
1707		if (xm->pgoff != 0) {
1708			map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map,
1709							symbol_name,
1710							xm->pgoff);
1711		}
1712
1713		if (machine__is_default_guest(machine)) {
1714			/*
1715			 * preload dso of guest kernel and modules
1716			 */
1717			dso__load(kernel, machine__kernel_map(machine));
1718		}
1719		dso__put(kernel);
1720	} else if (perf_event__is_extra_kernel_mmap(machine, xm)) {
1721		return machine__process_extra_kernel_map(machine, xm);
1722	}
1723	return 0;
1724out_problem:
1725	return -1;
1726}
1727
1728int machine__process_mmap2_event(struct machine *machine,
1729				 union perf_event *event,
1730				 struct perf_sample *sample)
1731{
1732	struct thread *thread;
1733	struct map *map;
1734	struct dso_id dso_id = dso_id_empty;
1735	int ret = 0;
1736
1737	if (dump_trace)
1738		perf_event__fprintf_mmap2(event, stdout);
1739
1740	if (event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID) {
1741		build_id__init(&dso_id.build_id, event->mmap2.build_id, event->mmap2.build_id_size);
1742	} else {
1743		dso_id.maj = event->mmap2.maj;
1744		dso_id.min = event->mmap2.min;
1745		dso_id.ino = event->mmap2.ino;
1746		dso_id.ino_generation = event->mmap2.ino_generation;
1747		dso_id.mmap2_valid = true;
1748		dso_id.mmap2_ino_generation_valid = true;
1749	}
1750
1751	if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1752	    sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1753		struct extra_kernel_map xm = {
1754			.start = event->mmap2.start,
1755			.end   = event->mmap2.start + event->mmap2.len,
1756			.pgoff = event->mmap2.pgoff,
1757		};
1758
1759		strlcpy(xm.name, event->mmap2.filename, KMAP_NAME_LEN);
1760		ret = machine__process_kernel_mmap_event(machine, &xm, &dso_id.build_id);
1761		if (ret < 0)
1762			goto out_problem;
1763		return 0;
1764	}
1765
1766	thread = machine__findnew_thread(machine, event->mmap2.pid,
1767					event->mmap2.tid);
1768	if (thread == NULL)
1769		goto out_problem;
1770
1771	map = map__new(machine, event->mmap2.start,
1772			event->mmap2.len, event->mmap2.pgoff,
1773			&dso_id, event->mmap2.prot,
1774			event->mmap2.flags,
1775			event->mmap2.filename, thread);
1776
1777	if (map == NULL)
1778		goto out_problem_map;
1779
1780	ret = thread__insert_map(thread, map);
1781	if (ret)
1782		goto out_problem_insert;
1783
1784	thread__put(thread);
1785	map__put(map);
1786	return 0;
1787
1788out_problem_insert:
1789	map__put(map);
1790out_problem_map:
1791	thread__put(thread);
1792out_problem:
1793	dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
1794	return 0;
1795}
1796
1797int machine__process_mmap_event(struct machine *machine, union perf_event *event,
1798				struct perf_sample *sample)
1799{
1800	struct thread *thread;
1801	struct map *map;
1802	u32 prot = 0;
1803	int ret = 0;
1804
1805	if (dump_trace)
1806		perf_event__fprintf_mmap(event, stdout);
1807
1808	if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1809	    sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1810		struct extra_kernel_map xm = {
1811			.start = event->mmap.start,
1812			.end   = event->mmap.start + event->mmap.len,
1813			.pgoff = event->mmap.pgoff,
1814		};
1815
1816		strlcpy(xm.name, event->mmap.filename, KMAP_NAME_LEN);
1817		ret = machine__process_kernel_mmap_event(machine, &xm, NULL);
1818		if (ret < 0)
1819			goto out_problem;
1820		return 0;
1821	}
1822
1823	thread = machine__findnew_thread(machine, event->mmap.pid,
1824					 event->mmap.tid);
1825	if (thread == NULL)
1826		goto out_problem;
1827
1828	if (!(event->header.misc & PERF_RECORD_MISC_MMAP_DATA))
1829		prot = PROT_EXEC;
1830
1831	map = map__new(machine, event->mmap.start,
1832		       event->mmap.len, event->mmap.pgoff,
1833		       &dso_id_empty, prot, /*flags=*/0, event->mmap.filename, thread);
1834
1835	if (map == NULL)
1836		goto out_problem_map;
1837
1838	ret = thread__insert_map(thread, map);
1839	if (ret)
1840		goto out_problem_insert;
1841
1842	thread__put(thread);
1843	map__put(map);
1844	return 0;
1845
1846out_problem_insert:
1847	map__put(map);
1848out_problem_map:
1849	thread__put(thread);
1850out_problem:
1851	dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
1852	return 0;
1853}
1854
1855void machine__remove_thread(struct machine *machine, struct thread *th)
1856{
1857	return threads__remove(&machine->threads, th);
1858}
1859
1860int machine__process_fork_event(struct machine *machine, union perf_event *event,
1861				struct perf_sample *sample)
1862{
1863	struct thread *thread = machine__find_thread(machine,
1864						     event->fork.pid,
1865						     event->fork.tid);
1866	struct thread *parent = machine__findnew_thread(machine,
1867							event->fork.ppid,
1868							event->fork.ptid);
1869	bool do_maps_clone = true;
1870	int err = 0;
1871
1872	if (dump_trace)
1873		perf_event__fprintf_task(event, stdout);
1874
1875	/*
1876	 * There may be an existing thread that is not actually the parent,
1877	 * either because we are processing events out of order, or because the
1878	 * (fork) event that would have removed the thread was lost. Assume the
1879	 * latter case and continue on as best we can.
1880	 */
1881	if (thread__pid(parent) != (pid_t)event->fork.ppid) {
1882		dump_printf("removing erroneous parent thread %d/%d\n",
1883			    thread__pid(parent), thread__tid(parent));
1884		machine__remove_thread(machine, parent);
1885		thread__put(parent);
1886		parent = machine__findnew_thread(machine, event->fork.ppid,
1887						 event->fork.ptid);
1888	}
1889
1890	/* if a thread currently exists for the thread id remove it */
1891	if (thread != NULL) {
1892		machine__remove_thread(machine, thread);
1893		thread__put(thread);
1894	}
1895
1896	thread = machine__findnew_thread(machine, event->fork.pid,
1897					 event->fork.tid);
1898	/*
1899	 * When synthesizing FORK events, we are trying to create thread
1900	 * objects for the already running tasks on the machine.
1901	 *
1902	 * Normally, for a kernel FORK event, we want to clone the parent's
1903	 * maps because that is what the kernel just did.
1904	 *
1905	 * But when synthesizing, this should not be done.  If we do, we end up
1906	 * with overlapping maps as we process the synthesized MMAP2 events that
1907	 * get delivered shortly thereafter.
1908	 *
1909	 * Use the FORK event misc flags in an internal way to signal this
1910	 * situation, so we can elide the map clone when appropriate.
1911	 */
1912	if (event->fork.header.misc & PERF_RECORD_MISC_FORK_EXEC)
1913		do_maps_clone = false;
1914
1915	if (thread == NULL || parent == NULL ||
1916	    thread__fork(thread, parent, sample->time, do_maps_clone) < 0) {
1917		dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
1918		err = -1;
1919	}
1920	thread__put(thread);
1921	thread__put(parent);
1922
1923	return err;
1924}
1925
1926int machine__process_exit_event(struct machine *machine, union perf_event *event,
1927				struct perf_sample *sample __maybe_unused)
1928{
1929	struct thread *thread = machine__find_thread(machine,
1930						     event->fork.pid,
1931						     event->fork.tid);
1932
1933	if (dump_trace)
1934		perf_event__fprintf_task(event, stdout);
1935
1936	/* There is no context switch out before exit, so we decrement here. */
1937	machine->parallelism--;
1938	if (thread != NULL) {
1939		if (symbol_conf.keep_exited_threads)
1940			thread__set_exited(thread, /*exited=*/true);
1941		else
1942			machine__remove_thread(machine, thread);
1943	}
1944	thread__put(thread);
1945	return 0;
1946}
1947
1948int machine__process_event(struct machine *machine, union perf_event *event,
1949			   struct perf_sample *sample)
1950{
1951	int ret;
1952
1953	switch (event->header.type) {
1954	case PERF_RECORD_COMM:
1955		ret = machine__process_comm_event(machine, event, sample); break;
1956	case PERF_RECORD_MMAP:
1957		ret = machine__process_mmap_event(machine, event, sample); break;
1958	case PERF_RECORD_NAMESPACES:
1959		ret = machine__process_namespaces_event(machine, event, sample); break;
1960	case PERF_RECORD_CGROUP:
1961		ret = machine__process_cgroup_event(machine, event, sample); break;
1962	case PERF_RECORD_MMAP2:
1963		ret = machine__process_mmap2_event(machine, event, sample); break;
1964	case PERF_RECORD_FORK:
1965		ret = machine__process_fork_event(machine, event, sample); break;
1966	case PERF_RECORD_EXIT:
1967		ret = machine__process_exit_event(machine, event, sample); break;
1968	case PERF_RECORD_LOST:
1969		ret = machine__process_lost_event(machine, event, sample); break;
1970	case PERF_RECORD_AUX:
1971		ret = machine__process_aux_event(machine, event); break;
1972	case PERF_RECORD_ITRACE_START:
1973		ret = machine__process_itrace_start_event(machine, event); break;
1974	case PERF_RECORD_LOST_SAMPLES:
1975		ret = machine__process_lost_samples_event(machine, event, sample); break;
1976	case PERF_RECORD_SWITCH:
1977	case PERF_RECORD_SWITCH_CPU_WIDE:
1978		ret = machine__process_switch_event(machine, event); break;
1979	case PERF_RECORD_KSYMBOL:
1980		ret = machine__process_ksymbol(machine, event, sample); break;
1981	case PERF_RECORD_BPF_EVENT:
1982		ret = machine__process_bpf(machine, event, sample); break;
1983	case PERF_RECORD_TEXT_POKE:
1984		ret = machine__process_text_poke(machine, event, sample); break;
1985	case PERF_RECORD_AUX_OUTPUT_HW_ID:
1986		ret = machine__process_aux_output_hw_id_event(machine, event); break;
1987	default:
1988		ret = -1;
1989		break;
1990	}
1991
1992	return ret;
1993}
1994
1995static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
1996{
1997	return regexec(regex, sym->name, 0, NULL, 0) == 0;
1998}
1999
2000static void ip__resolve_ams(struct thread *thread,
2001			    struct addr_map_symbol *ams,
2002			    u64 ip)
2003{
2004	struct addr_location al;
2005
2006	addr_location__init(&al);
2007	/*
2008	 * We cannot use the header.misc hint to determine whether a
2009	 * branch stack address is user, kernel, guest, hypervisor.
2010	 * Branches may straddle the kernel/user/hypervisor boundaries.
2011	 * Thus, we have to try consecutively until we find a match
2012	 * or else, the symbol is unknown
2013	 */
2014	thread__find_cpumode_addr_location(thread, ip, /*symbols=*/true, &al);
2015
2016	ams->addr = ip;
2017	ams->al_addr = al.addr;
2018	ams->al_level = al.level;
2019	ams->ms.maps = maps__get(al.maps);
2020	ams->ms.sym = al.sym;
2021	ams->ms.map = map__get(al.map);
2022	ams->phys_addr = 0;
2023	ams->data_page_size = 0;
2024	addr_location__exit(&al);
2025}
2026
2027static void ip__resolve_data(struct thread *thread,
2028			     u8 m, struct addr_map_symbol *ams,
2029			     u64 addr, u64 phys_addr, u64 daddr_page_size)
2030{
2031	struct addr_location al;
2032
2033	addr_location__init(&al);
2034
2035	thread__find_symbol(thread, m, addr, &al);
2036
2037	ams->addr = addr;
2038	ams->al_addr = al.addr;
2039	ams->al_level = al.level;
2040	ams->ms.maps = maps__get(al.maps);
2041	ams->ms.sym = al.sym;
2042	ams->ms.map = map__get(al.map);
2043	ams->phys_addr = phys_addr;
2044	ams->data_page_size = daddr_page_size;
2045	addr_location__exit(&al);
2046}
2047
2048struct mem_info *sample__resolve_mem(struct perf_sample *sample,
2049				     struct addr_location *al)
2050{
2051	struct mem_info *mi = mem_info__new();
2052
2053	if (!mi)
2054		return NULL;
2055
2056	ip__resolve_ams(al->thread, mem_info__iaddr(mi), sample->ip);
2057	ip__resolve_data(al->thread, al->cpumode, mem_info__daddr(mi),
2058			 sample->addr, sample->phys_addr,
2059			 sample->data_page_size);
2060	mem_info__data_src(mi)->val = sample->data_src;
2061
2062	return mi;
2063}
2064
2065static char *callchain_srcline(struct map_symbol *ms, u64 ip)
2066{
2067	struct map *map = ms->map;
2068	char *srcline = NULL;
2069	struct dso *dso;
2070
2071	if (!map || callchain_param.key == CCKEY_FUNCTION)
2072		return srcline;
2073
2074	dso = map__dso(map);
2075	srcline = srcline__tree_find(dso__srclines(dso), ip);
2076	if (!srcline) {
2077		bool show_sym = false;
2078		bool show_addr = callchain_param.key == CCKEY_ADDRESS;
2079
2080		srcline = get_srcline(dso, map__rip_2objdump(map, ip),
2081				      ms->sym, show_sym, show_addr, ip);
2082		srcline__tree_insert(dso__srclines(dso), ip, srcline);
2083	}
2084
2085	return srcline;
2086}
2087
2088struct iterations {
2089	int nr_loop_iter;
2090	u64 cycles;
2091};
2092
2093static int add_callchain_ip(struct thread *thread,
2094			    struct callchain_cursor *cursor,
2095			    struct symbol **parent,
2096			    struct addr_location *root_al,
2097			    u8 *cpumode,
2098			    u64 ip,
2099			    bool branch,
2100			    struct branch_flags *flags,
2101			    struct iterations *iter,
2102			    u64 branch_from,
2103			    bool symbols)
2104{
2105	struct map_symbol ms = {};
2106	struct addr_location al;
2107	int nr_loop_iter = 0, err = 0;
2108	u64 iter_cycles = 0;
2109	const char *srcline = NULL;
2110
2111	addr_location__init(&al);
2112	al.filtered = 0;
2113	al.sym = NULL;
2114	al.srcline = NULL;
2115	if (!cpumode) {
2116		thread__find_cpumode_addr_location(thread, ip, symbols, &al);
2117	} else {
2118		if (ip >= PERF_CONTEXT_MAX) {
2119			switch (ip) {
2120			case PERF_CONTEXT_HV:
2121				*cpumode = PERF_RECORD_MISC_HYPERVISOR;
2122				break;
2123			case PERF_CONTEXT_KERNEL:
2124				*cpumode = PERF_RECORD_MISC_KERNEL;
2125				break;
2126			case PERF_CONTEXT_USER:
2127			case PERF_CONTEXT_USER_DEFERRED:
2128				*cpumode = PERF_RECORD_MISC_USER;
2129				break;
2130			default:
2131				pr_debug("invalid callchain context: "
2132					 "%"PRId64"\n", (s64) ip);
2133				/*
2134				 * It seems the callchain is corrupted.
2135				 * Discard all.
2136				 */
2137				callchain_cursor_reset(cursor);
2138				err = 1;
2139				goto out;
2140			}
2141			goto out;
2142		}
2143		if (symbols)
2144			thread__find_symbol(thread, *cpumode, ip, &al);
2145		else
2146			thread__find_map(thread, *cpumode, ip, &al);
2147	}
2148
2149	if (al.sym != NULL) {
2150		if (perf_hpp_list.parent && !*parent &&
2151		    symbol__match_regex(al.sym, &parent_regex))
2152			*parent = al.sym;
2153		else if (have_ignore_callees && root_al &&
2154		  symbol__match_regex(al.sym, &ignore_callees_regex)) {
2155			/* Treat this symbol as the root,
2156			   forgetting its callees. */
2157			addr_location__copy(root_al, &al);
2158			callchain_cursor_reset(cursor);
2159		}
2160	}
2161
2162	if (symbol_conf.hide_unresolved && al.sym == NULL)
2163		goto out;
2164
2165	if (iter) {
2166		nr_loop_iter = iter->nr_loop_iter;
2167		iter_cycles = iter->cycles;
2168	}
2169
2170	ms.maps = maps__get(al.maps);
2171	ms.map = map__get(al.map);
2172	ms.sym = al.sym;
2173	srcline = callchain_srcline(&ms, al.addr);
2174	err = callchain_cursor_append(cursor, ip, &ms,
2175				      branch, flags, nr_loop_iter,
2176				      iter_cycles, branch_from, srcline);
2177out:
2178	addr_location__exit(&al);
2179	map_symbol__exit(&ms);
2180	return err;
2181}
2182
2183struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
2184					   struct addr_location *al)
2185{
2186	unsigned int i;
2187	const struct branch_stack *bs = sample->branch_stack;
2188	struct branch_entry *entries = perf_sample__branch_entries(sample);
2189	u64 *branch_stack_cntr = sample->branch_stack_cntr;
2190	struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
2191
2192	if (!bi)
2193		return NULL;
2194
2195	for (i = 0; i < bs->nr; i++) {
2196		ip__resolve_ams(al->thread, &bi[i].to, entries[i].to);
2197		ip__resolve_ams(al->thread, &bi[i].from, entries[i].from);
2198		bi[i].flags = entries[i].flags;
2199		if (branch_stack_cntr)
2200			bi[i].branch_stack_cntr  = branch_stack_cntr[i];
2201	}
2202	return bi;
2203}
2204
2205static void save_iterations(struct iterations *iter,
2206			    struct branch_entry *be, int nr)
2207{
2208	int i;
2209
2210	iter->nr_loop_iter++;
2211	iter->cycles = 0;
2212
2213	for (i = 0; i < nr; i++)
2214		iter->cycles += be[i].flags.cycles;
2215}
2216
2217#define CHASHSZ 127
2218#define CHASHBITS 7
2219#define NO_ENTRY 0xff
2220
2221#define PERF_MAX_BRANCH_DEPTH 127
2222
2223/* Remove loops. */
2224static int remove_loops(struct branch_entry *l, int nr,
2225			struct iterations *iter)
2226{
2227	int i, j, off;
2228	unsigned char chash[CHASHSZ];
2229
2230	memset(chash, NO_ENTRY, sizeof(chash));
2231
2232	BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
2233
2234	for (i = 0; i < nr; i++) {
2235		int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
2236
2237		/* no collision handling for now */
2238		if (chash[h] == NO_ENTRY) {
2239			chash[h] = i;
2240		} else if (l[chash[h]].from == l[i].from) {
2241			bool is_loop = true;
2242			/* check if it is a real loop */
2243			off = 0;
2244			for (j = chash[h]; j < i && i + off < nr; j++, off++)
2245				if (l[j].from != l[i + off].from) {
2246					is_loop = false;
2247					break;
2248				}
2249			if (is_loop) {
2250				j = nr - (i + off);
2251				if (j > 0) {
2252					save_iterations(iter + i + off,
2253						l + i, off);
2254
2255					memmove(iter + i, iter + i + off,
2256						j * sizeof(*iter));
2257
2258					memmove(l + i, l + i + off,
2259						j * sizeof(*l));
2260				}
2261
2262				nr -= off;
2263			}
2264		}
2265	}
2266	return nr;
2267}
2268
2269static int lbr_callchain_add_kernel_ip(struct thread *thread,
2270				       struct callchain_cursor *cursor,
2271				       struct perf_sample *sample,
2272				       struct symbol **parent,
2273				       struct addr_location *root_al,
2274				       u64 branch_from,
2275				       bool callee, int end,
2276				       bool symbols)
2277{
2278	struct ip_callchain *chain = sample->callchain;
2279	u8 cpumode = PERF_RECORD_MISC_USER;
2280	int err, i;
2281
2282	if (callee) {
2283		for (i = 0; i < end + 1; i++) {
2284			err = add_callchain_ip(thread, cursor, parent,
2285					       root_al, &cpumode, chain->ips[i],
2286					       false, NULL, NULL, branch_from,
2287					       symbols);
2288			if (err)
2289				return err;
2290		}
2291		return 0;
2292	}
2293
2294	for (i = end; i >= 0; i--) {
2295		err = add_callchain_ip(thread, cursor, parent,
2296				       root_al, &cpumode, chain->ips[i],
2297				       false, NULL, NULL, branch_from,
2298				       symbols);
2299		if (err)
2300			return err;
2301	}
2302
2303	return 0;
2304}
2305
2306static void save_lbr_cursor_node(struct thread *thread,
2307				 struct callchain_cursor *cursor,
2308				 int idx)
2309{
2310	struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2311
2312	if (!lbr_stitch)
2313		return;
2314
2315	if (cursor->pos == cursor->nr) {
2316		lbr_stitch->prev_lbr_cursor[idx].valid = false;
2317		return;
2318	}
2319
2320	if (!cursor->curr)
2321		cursor->curr = cursor->first;
2322	else
2323		cursor->curr = cursor->curr->next;
2324
2325	map_symbol__exit(&lbr_stitch->prev_lbr_cursor[idx].ms);
2326	memcpy(&lbr_stitch->prev_lbr_cursor[idx], cursor->curr,
2327	       sizeof(struct callchain_cursor_node));
2328	lbr_stitch->prev_lbr_cursor[idx].ms.maps = maps__get(cursor->curr->ms.maps);
2329	lbr_stitch->prev_lbr_cursor[idx].ms.map = map__get(cursor->curr->ms.map);
2330
2331	lbr_stitch->prev_lbr_cursor[idx].valid = true;
2332	cursor->pos++;
2333}
2334
2335static int lbr_callchain_add_lbr_ip(struct thread *thread,
2336				    struct callchain_cursor *cursor,
2337				    struct perf_sample *sample,
2338				    struct symbol **parent,
2339				    struct addr_location *root_al,
2340				    u64 *branch_from,
2341				    bool callee,
2342				    bool symbols)
2343{
2344	struct branch_stack *lbr_stack = sample->branch_stack;
2345	struct branch_entry *entries = perf_sample__branch_entries(sample);
2346	u8 cpumode = PERF_RECORD_MISC_USER;
2347	int lbr_nr = lbr_stack->nr;
2348	struct branch_flags *flags;
2349	int err, i;
2350	u64 ip;
2351
2352	/*
2353	 * The curr and pos are not used in writing session. They are cleared
2354	 * in callchain_cursor_commit() when the writing session is closed.
2355	 * Using curr and pos to track the current cursor node.
2356	 */
2357	if (thread__lbr_stitch(thread)) {
2358		cursor->curr = NULL;
2359		cursor->pos = cursor->nr;
2360		if (cursor->nr) {
2361			cursor->curr = cursor->first;
2362			for (i = 0; i < (int)(cursor->nr - 1); i++)
2363				cursor->curr = cursor->curr->next;
2364		}
2365	}
2366
2367	if (callee) {
2368		/* Add LBR ip from first entries.to */
2369		ip = entries[0].to;
2370		flags = &entries[0].flags;
2371		*branch_from = entries[0].from;
2372		err = add_callchain_ip(thread, cursor, parent,
2373				       root_al, &cpumode, ip,
2374				       true, flags, NULL,
2375				       *branch_from, symbols);
2376		if (err)
2377			return err;
2378
2379		/*
2380		 * The number of cursor node increases.
2381		 * Move the current cursor node.
2382		 * But does not need to save current cursor node for entry 0.
2383		 * It's impossible to stitch the whole LBRs of previous sample.
2384		 */
2385		if (thread__lbr_stitch(thread) && (cursor->pos != cursor->nr)) {
2386			if (!cursor->curr)
2387				cursor->curr = cursor->first;
2388			else
2389				cursor->curr = cursor->curr->next;
2390			cursor->pos++;
2391		}
2392
2393		/* Add LBR ip from entries.from one by one. */
2394		for (i = 0; i < lbr_nr; i++) {
2395			ip = entries[i].from;
2396			flags = &entries[i].flags;
2397			err = add_callchain_ip(thread, cursor, parent,
2398					       root_al, &cpumode, ip,
2399					       true, flags, NULL,
2400					       *branch_from, symbols);
2401			if (err)
2402				return err;
2403			save_lbr_cursor_node(thread, cursor, i);
2404		}
2405		return 0;
2406	}
2407
2408	/* Add LBR ip from entries.from one by one. */
2409	for (i = lbr_nr - 1; i >= 0; i--) {
2410		ip = entries[i].from;
2411		flags = &entries[i].flags;
2412		err = add_callchain_ip(thread, cursor, parent,
2413				       root_al, &cpumode, ip,
2414				       true, flags, NULL,
2415				       *branch_from, symbols);
2416		if (err)
2417			return err;
2418		save_lbr_cursor_node(thread, cursor, i);
2419	}
2420
2421	if (lbr_nr > 0) {
2422		/* Add LBR ip from first entries.to */
2423		ip = entries[0].to;
2424		flags = &entries[0].flags;
2425		*branch_from = entries[0].from;
2426		err = add_callchain_ip(thread, cursor, parent,
2427				root_al, &cpumode, ip,
2428				true, flags, NULL,
2429				*branch_from, symbols);
2430		if (err)
2431			return err;
2432	}
2433
2434	return 0;
2435}
2436
2437static int lbr_callchain_add_stitched_lbr_ip(struct thread *thread,
2438					     struct callchain_cursor *cursor)
2439{
2440	struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2441	struct callchain_cursor_node *cnode;
2442	struct stitch_list *stitch_node;
2443	int err;
2444
2445	list_for_each_entry(stitch_node, &lbr_stitch->lists, node) {
2446		cnode = &stitch_node->cursor;
2447
2448		err = callchain_cursor_append(cursor, cnode->ip,
2449					      &cnode->ms,
2450					      cnode->branch,
2451					      &cnode->branch_flags,
2452					      cnode->nr_loop_iter,
2453					      cnode->iter_cycles,
2454					      cnode->branch_from,
2455					      cnode->srcline);
2456		if (err)
2457			return err;
2458	}
2459	return 0;
2460}
2461
2462static struct stitch_list *get_stitch_node(struct thread *thread)
2463{
2464	struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2465	struct stitch_list *stitch_node;
2466
2467	if (!list_empty(&lbr_stitch->free_lists)) {
2468		stitch_node = list_first_entry(&lbr_stitch->free_lists,
2469					       struct stitch_list, node);
2470		list_del(&stitch_node->node);
2471
2472		return stitch_node;
2473	}
2474
2475	return malloc(sizeof(struct stitch_list));
2476}
2477
2478static bool has_stitched_lbr(struct thread *thread,
2479			     struct perf_sample *cur,
2480			     struct perf_sample *prev,
2481			     unsigned int max_lbr,
2482			     bool callee)
2483{
2484	struct branch_stack *cur_stack = cur->branch_stack;
2485	struct branch_entry *cur_entries = perf_sample__branch_entries(cur);
2486	struct branch_stack *prev_stack = prev->branch_stack;
2487	struct branch_entry *prev_entries = perf_sample__branch_entries(prev);
2488	struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2489	int i, j, nr_identical_branches = 0;
2490	struct stitch_list *stitch_node;
2491	u64 cur_base, distance;
2492
2493	if (!cur_stack || !prev_stack)
2494		return false;
2495
2496	/* Find the physical index of the base-of-stack for current sample. */
2497	cur_base = max_lbr - cur_stack->nr + cur_stack->hw_idx + 1;
2498
2499	distance = (prev_stack->hw_idx > cur_base) ? (prev_stack->hw_idx - cur_base) :
2500						     (max_lbr + prev_stack->hw_idx - cur_base);
2501	/* Previous sample has shorter stack. Nothing can be stitched. */
2502	if (distance + 1 > prev_stack->nr)
2503		return false;
2504
2505	/*
2506	 * Check if there are identical LBRs between two samples.
2507	 * Identical LBRs must have same from, to and flags values. Also,
2508	 * they have to be saved in the same LBR registers (same physical
2509	 * index).
2510	 *
2511	 * Starts from the base-of-stack of current sample.
2512	 */
2513	for (i = distance, j = cur_stack->nr - 1; (i >= 0) && (j >= 0); i--, j--) {
2514		if ((prev_entries[i].from != cur_entries[j].from) ||
2515		    (prev_entries[i].to != cur_entries[j].to) ||
2516		    (prev_entries[i].flags.value != cur_entries[j].flags.value))
2517			break;
2518		nr_identical_branches++;
2519	}
2520
2521	if (!nr_identical_branches)
2522		return false;
2523
2524	/*
2525	 * Save the LBRs between the base-of-stack of previous sample
2526	 * and the base-of-stack of current sample into lbr_stitch->lists.
2527	 * These LBRs will be stitched later.
2528	 */
2529	for (i = prev_stack->nr - 1; i > (int)distance; i--) {
2530
2531		if (!lbr_stitch->prev_lbr_cursor[i].valid)
2532			continue;
2533
2534		stitch_node = get_stitch_node(thread);
2535		if (!stitch_node)
2536			return false;
2537
2538		memcpy(&stitch_node->cursor, &lbr_stitch->prev_lbr_cursor[i],
2539		       sizeof(struct callchain_cursor_node));
2540
2541		stitch_node->cursor.ms.maps = maps__get(lbr_stitch->prev_lbr_cursor[i].ms.maps);
2542		stitch_node->cursor.ms.map = map__get(lbr_stitch->prev_lbr_cursor[i].ms.map);
2543
2544		if (callee)
2545			list_add(&stitch_node->node, &lbr_stitch->lists);
2546		else
2547			list_add_tail(&stitch_node->node, &lbr_stitch->lists);
2548	}
2549
2550	return true;
2551}
2552
2553static bool alloc_lbr_stitch(struct thread *thread, unsigned int max_lbr)
2554{
2555	if (thread__lbr_stitch(thread))
2556		return true;
2557
2558	thread__set_lbr_stitch(thread, zalloc(sizeof(struct lbr_stitch)));
2559	if (!thread__lbr_stitch(thread))
2560		goto err;
2561
2562	thread__lbr_stitch(thread)->prev_lbr_cursor =
2563		calloc(max_lbr + 1, sizeof(struct callchain_cursor_node));
2564	if (!thread__lbr_stitch(thread)->prev_lbr_cursor)
2565		goto free_lbr_stitch;
2566
2567	thread__lbr_stitch(thread)->prev_lbr_cursor_size = max_lbr + 1;
2568
2569	INIT_LIST_HEAD(&thread__lbr_stitch(thread)->lists);
2570	INIT_LIST_HEAD(&thread__lbr_stitch(thread)->free_lists);
2571
2572	return true;
2573
2574free_lbr_stitch:
2575	free(thread__lbr_stitch(thread));
2576	thread__set_lbr_stitch(thread, NULL);
2577err:
2578	pr_warning("Failed to allocate space for stitched LBRs. Disable LBR stitch\n");
2579	thread__set_lbr_stitch_enable(thread, false);
2580	return false;
2581}
2582
2583/*
2584 * Resolve LBR callstack chain sample
2585 * Return:
2586 * 1 on success get LBR callchain information
2587 * 0 no available LBR callchain information, should try fp
2588 * negative error code on other errors.
2589 */
2590static int resolve_lbr_callchain_sample(struct thread *thread,
2591					struct callchain_cursor *cursor,
2592					struct perf_sample *sample,
2593					struct symbol **parent,
2594					struct addr_location *root_al,
2595					int max_stack,
2596					unsigned int max_lbr,
2597					bool symbols)
2598{
2599	bool callee = (callchain_param.order == ORDER_CALLEE);
2600	struct ip_callchain *chain = sample->callchain;
2601	int chain_nr = min(max_stack, (int)chain->nr), i;
2602	struct lbr_stitch *lbr_stitch;
2603	bool stitched_lbr = false;
2604	u64 branch_from = 0;
2605	int err;
2606
2607	for (i = 0; i < chain_nr; i++) {
2608		if (chain->ips[i] == PERF_CONTEXT_USER)
2609			break;
2610	}
2611
2612	/* LBR only affects the user callchain */
2613	if (i == chain_nr)
2614		return 0;
2615
2616	if (thread__lbr_stitch_enable(thread) && !sample->no_hw_idx &&
2617	    (max_lbr > 0) && alloc_lbr_stitch(thread, max_lbr)) {
2618		lbr_stitch = thread__lbr_stitch(thread);
2619
2620		stitched_lbr = has_stitched_lbr(thread, sample,
2621						&lbr_stitch->prev_sample,
2622						max_lbr, callee);
2623
2624		if (!stitched_lbr && !list_empty(&lbr_stitch->lists)) {
2625			struct stitch_list *stitch_node;
2626
2627			list_for_each_entry(stitch_node, &lbr_stitch->lists, node)
2628				map_symbol__exit(&stitch_node->cursor.ms);
2629
2630			list_splice_init(&lbr_stitch->lists, &lbr_stitch->free_lists);
2631		}
2632		memcpy(&lbr_stitch->prev_sample, sample, sizeof(*sample));
2633	}
2634
2635	if (callee) {
2636		/* Add kernel ip */
2637		err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2638						  parent, root_al, branch_from,
2639						  true, i, symbols);
2640		if (err)
2641			goto error;
2642
2643		err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2644					       root_al, &branch_from, true, symbols);
2645		if (err)
2646			goto error;
2647
2648		if (stitched_lbr) {
2649			err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2650			if (err)
2651				goto error;
2652		}
2653
2654	} else {
2655		if (stitched_lbr) {
2656			err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2657			if (err)
2658				goto error;
2659		}
2660		err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2661					       root_al, &branch_from, false, symbols);
2662		if (err)
2663			goto error;
2664
2665		/* Add kernel ip */
2666		err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2667						  parent, root_al, branch_from,
2668						  false, i, symbols);
2669		if (err)
2670			goto error;
2671	}
2672	return 1;
2673
2674error:
2675	return (err < 0) ? err : 0;
2676}
2677
2678static int find_prev_cpumode(struct ip_callchain *chain, struct thread *thread,
2679			     struct callchain_cursor *cursor,
2680			     struct symbol **parent,
2681			     struct addr_location *root_al,
2682			     u8 *cpumode, int ent, bool symbols)
2683{
2684	int err = 0;
2685
2686	while (--ent >= 0) {
2687		u64 ip = chain->ips[ent];
2688
2689		if (ip >= PERF_CONTEXT_MAX) {
2690			err = add_callchain_ip(thread, cursor, parent,
2691					       root_al, cpumode, ip,
2692					       false, NULL, NULL, 0, symbols);
2693			break;
2694		}
2695	}
2696	return err;
2697}
2698
2699static u64 get_leaf_frame_caller(struct perf_sample *sample,
2700		struct thread *thread, int usr_idx)
2701{
2702	if (machine__normalized_is(maps__machine(thread__maps(thread)), "arm64"))
2703		return get_leaf_frame_caller_aarch64(sample, thread, usr_idx);
2704	else
2705		return 0;
2706}
2707
2708static int thread__resolve_callchain_sample(struct thread *thread,
2709					    struct callchain_cursor *cursor,
2710					    struct evsel *evsel,
2711					    struct perf_sample *sample,
2712					    struct symbol **parent,
2713					    struct addr_location *root_al,
2714					    int max_stack,
2715					    bool symbols)
2716{
2717	struct branch_stack *branch = sample->branch_stack;
2718	struct branch_entry *entries = perf_sample__branch_entries(sample);
2719	struct ip_callchain *chain = sample->callchain;
2720	int chain_nr = 0;
2721	u8 cpumode = PERF_RECORD_MISC_USER;
2722	int i, j, err, nr_entries, usr_idx;
2723	int skip_idx = -1;
2724	int first_call = 0;
2725	u64 leaf_frame_caller;
2726
2727	if (chain)
2728		chain_nr = chain->nr;
2729
2730	if (evsel__has_branch_callstack(evsel)) {
2731		struct perf_env *env = evsel__env(evsel);
2732
2733		err = resolve_lbr_callchain_sample(thread, cursor, sample, parent,
2734						   root_al, max_stack,
2735						   !env ? 0 : env->max_branches,
2736						   symbols);
2737		if (err)
2738			return (err < 0) ? err : 0;
2739	}
2740
2741	/*
2742	 * Based on DWARF debug information, some architectures skip
2743	 * a callchain entry saved by the kernel.
2744	 */
2745	skip_idx = arch_skip_callchain_idx(thread, chain);
2746
2747	/*
2748	 * Add branches to call stack for easier browsing. This gives
2749	 * more context for a sample than just the callers.
2750	 *
2751	 * This uses individual histograms of paths compared to the
2752	 * aggregated histograms the normal LBR mode uses.
2753	 *
2754	 * Limitations for now:
2755	 * - No extra filters
2756	 * - No annotations (should annotate somehow)
2757	 */
2758
2759	if (branch && callchain_param.branch_callstack) {
2760		int nr = min(max_stack, (int)branch->nr);
2761		struct branch_entry be[nr];
2762		struct iterations iter[nr];
2763
2764		if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
2765			pr_warning("corrupted branch chain. skipping...\n");
2766			goto check_calls;
2767		}
2768
2769		for (i = 0; i < nr; i++) {
2770			if (callchain_param.order == ORDER_CALLEE) {
2771				be[i] = entries[i];
2772
2773				if (chain == NULL)
2774					continue;
2775
2776				/*
2777				 * Check for overlap into the callchain.
2778				 * The return address is one off compared to
2779				 * the branch entry. To adjust for this
2780				 * assume the calling instruction is not longer
2781				 * than 8 bytes.
2782				 */
2783				if (i == skip_idx ||
2784				    chain->ips[first_call] >= PERF_CONTEXT_MAX)
2785					first_call++;
2786				else if (be[i].from < chain->ips[first_call] &&
2787				    be[i].from >= chain->ips[first_call] - 8)
2788					first_call++;
2789			} else
2790				be[i] = entries[branch->nr - i - 1];
2791		}
2792
2793		memset(iter, 0, sizeof(struct iterations) * nr);
2794		nr = remove_loops(be, nr, iter);
2795
2796		for (i = 0; i < nr; i++) {
2797			err = add_callchain_ip(thread, cursor, parent,
2798					       root_al,
2799					       NULL, be[i].to,
2800					       true, &be[i].flags,
2801					       NULL, be[i].from, symbols);
2802
2803			if (!err) {
2804				err = add_callchain_ip(thread, cursor, parent, root_al,
2805						       NULL, be[i].from,
2806						       true, &be[i].flags,
2807						       &iter[i], 0, symbols);
2808			}
2809			if (err == -EINVAL)
2810				break;
2811			if (err)
2812				return err;
2813		}
2814
2815		if (chain_nr == 0)
2816			return 0;
2817
2818		chain_nr -= nr;
2819	}
2820
2821check_calls:
2822	if (chain && callchain_param.order != ORDER_CALLEE) {
2823		err = find_prev_cpumode(chain, thread, cursor, parent, root_al,
2824					&cpumode, chain->nr - first_call, symbols);
2825		if (err)
2826			return (err < 0) ? err : 0;
2827	}
2828	for (i = first_call, nr_entries = 0;
2829	     i < chain_nr && nr_entries < max_stack; i++) {
2830		u64 ip;
2831
2832		if (callchain_param.order == ORDER_CALLEE)
2833			j = i;
2834		else
2835			j = chain->nr - i - 1;
2836
2837#ifdef HAVE_SKIP_CALLCHAIN_IDX
2838		if (j == skip_idx)
2839			continue;
2840#endif
2841		ip = chain->ips[j];
2842		if (ip < PERF_CONTEXT_MAX)
2843                       ++nr_entries;
2844		else if (callchain_param.order != ORDER_CALLEE) {
2845			err = find_prev_cpumode(chain, thread, cursor, parent,
2846						root_al, &cpumode, j, symbols);
2847			if (err)
2848				return (err < 0) ? err : 0;
2849			continue;
2850		}
2851
2852		/*
2853		 * PERF_CONTEXT_USER allows us to locate where the user stack ends.
2854		 * Depending on callchain_param.order and the position of PERF_CONTEXT_USER,
2855		 * the index will be different in order to add the missing frame
2856		 * at the right place.
2857		 */
2858
2859		usr_idx = callchain_param.order == ORDER_CALLEE ? j-2 : j-1;
2860
2861		if (usr_idx >= 0 && chain->ips[usr_idx] == PERF_CONTEXT_USER) {
2862
2863			leaf_frame_caller = get_leaf_frame_caller(sample, thread, usr_idx);
2864
2865			/*
2866			 * check if leaf_frame_Caller != ip to not add the same
2867			 * value twice.
2868			 */
2869
2870			if (leaf_frame_caller && leaf_frame_caller != ip) {
2871
2872				err = add_callchain_ip(thread, cursor, parent,
2873						root_al, &cpumode, leaf_frame_caller,
2874						false, NULL, NULL, 0, symbols);
2875				if (err)
2876					return (err < 0) ? err : 0;
2877			}
2878		}
2879
2880		err = add_callchain_ip(thread, cursor, parent,
2881				       root_al, &cpumode, ip,
2882				       false, NULL, NULL, 0, symbols);
2883
2884		if (err)
2885			return (err < 0) ? err : 0;
2886	}
2887
2888	return 0;
2889}
2890
2891static int append_inlines(struct callchain_cursor *cursor, struct map_symbol *ms, u64 ip)
2892{
2893	struct symbol *sym = ms->sym;
2894	struct map *map = ms->map;
2895	struct inline_node *inline_node;
2896	struct inline_list *ilist;
2897	struct dso *dso;
2898	u64 addr;
2899	int ret = 1;
2900	struct map_symbol ilist_ms;
2901
2902	if (!symbol_conf.inline_name || !map || !sym)
2903		return ret;
2904
2905	addr = map__dso_map_ip(map, ip);
2906	addr = map__rip_2objdump(map, addr);
2907	dso = map__dso(map);
2908
2909	inline_node = inlines__tree_find(dso__inlined_nodes(dso), addr);
2910	if (!inline_node) {
2911		inline_node = dso__parse_addr_inlines(dso, addr, sym);
2912		if (!inline_node)
2913			return ret;
2914		inlines__tree_insert(dso__inlined_nodes(dso), inline_node);
2915	}
2916
2917	ilist_ms = (struct map_symbol) {
2918		.maps = maps__get(ms->maps),
2919		.map = map__get(map),
2920	};
2921	list_for_each_entry(ilist, &inline_node->val, list) {
2922		ilist_ms.sym = ilist->symbol;
2923		ret = callchain_cursor_append(cursor, ip, &ilist_ms, false,
2924					      NULL, 0, 0, 0, ilist->srcline);
2925
2926		if (ret != 0)
2927			return ret;
2928	}
2929	map_symbol__exit(&ilist_ms);
2930
2931	return ret;
2932}
2933
2934static int unwind_entry(struct unwind_entry *entry, void *arg)
2935{
2936	struct callchain_cursor *cursor = arg;
2937	const char *srcline = NULL;
2938	u64 addr = entry->ip;
2939
2940	if (symbol_conf.hide_unresolved && entry->ms.sym == NULL)
2941		return 0;
2942
2943	if (append_inlines(cursor, &entry->ms, entry->ip) == 0)
2944		return 0;
2945
2946	/*
2947	 * Convert entry->ip from a virtual address to an offset in
2948	 * its corresponding binary.
2949	 */
2950	if (entry->ms.map)
2951		addr = map__dso_map_ip(entry->ms.map, entry->ip);
2952
2953	srcline = callchain_srcline(&entry->ms, addr);
2954	return callchain_cursor_append(cursor, entry->ip, &entry->ms,
2955				       false, NULL, 0, 0, 0, srcline);
2956}
2957
2958static int thread__resolve_callchain_unwind(struct thread *thread,
2959					    struct callchain_cursor *cursor,
2960					    struct evsel *evsel,
2961					    struct perf_sample *sample,
2962					    int max_stack, bool symbols)
2963{
2964	/* Can we do dwarf post unwind? */
2965	if (!((evsel->core.attr.sample_type & PERF_SAMPLE_REGS_USER) &&
2966	      (evsel->core.attr.sample_type & PERF_SAMPLE_STACK_USER)))
2967		return 0;
2968
2969	/* Bail out if nothing was captured. */
2970	if (!sample->user_regs || !sample->user_regs->regs ||
2971	    !sample->user_stack.size)
2972		return 0;
2973
2974	if (!symbols)
2975		pr_debug("Not resolving symbols with an unwinder isn't currently supported\n");
2976
2977	return unwind__get_entries(unwind_entry, cursor,
2978				   thread, sample, max_stack, false);
2979}
2980
2981int __thread__resolve_callchain(struct thread *thread,
2982				struct callchain_cursor *cursor,
2983				struct evsel *evsel,
2984				struct perf_sample *sample,
2985				struct symbol **parent,
2986				struct addr_location *root_al,
2987				int max_stack,
2988				bool symbols)
2989{
2990	int ret = 0;
2991
2992	if (cursor == NULL)
2993		return -ENOMEM;
2994
2995	callchain_cursor_reset(cursor);
2996
2997	if (callchain_param.order == ORDER_CALLEE) {
2998		ret = thread__resolve_callchain_sample(thread, cursor,
2999						       evsel, sample,
3000						       parent, root_al,
3001						       max_stack, symbols);
3002		if (ret)
3003			return ret;
3004		ret = thread__resolve_callchain_unwind(thread, cursor,
3005						       evsel, sample,
3006						       max_stack, symbols);
3007	} else {
3008		ret = thread__resolve_callchain_unwind(thread, cursor,
3009						       evsel, sample,
3010						       max_stack, symbols);
3011		if (ret)
3012			return ret;
3013		ret = thread__resolve_callchain_sample(thread, cursor,
3014						       evsel, sample,
3015						       parent, root_al,
3016						       max_stack, symbols);
3017	}
3018
3019	return ret;
3020}
3021
3022int machine__for_each_thread(struct machine *machine,
3023			     int (*fn)(struct thread *thread, void *p),
3024			     void *priv)
3025{
3026	return threads__for_each_thread(&machine->threads, fn, priv);
3027}
3028
3029int machines__for_each_thread(struct machines *machines,
3030			      int (*fn)(struct thread *thread, void *p),
3031			      void *priv)
3032{
3033	struct rb_node *nd;
3034	int rc = 0;
3035
3036	rc = machine__for_each_thread(&machines->host, fn, priv);
3037	if (rc != 0)
3038		return rc;
3039
3040	for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
3041		struct machine *machine = rb_entry(nd, struct machine, rb_node);
3042
3043		rc = machine__for_each_thread(machine, fn, priv);
3044		if (rc != 0)
3045			return rc;
3046	}
3047	return rc;
3048}
3049
3050
3051static int thread_list_cb(struct thread *thread, void *data)
3052{
3053	struct list_head *list = data;
3054	struct thread_list *entry = malloc(sizeof(*entry));
3055
3056	if (!entry)
3057		return -ENOMEM;
3058
3059	entry->thread = thread__get(thread);
3060	list_add_tail(&entry->list, list);
3061	return 0;
3062}
3063
3064int machine__thread_list(struct machine *machine, struct list_head *list)
3065{
3066	return machine__for_each_thread(machine, thread_list_cb, list);
3067}
3068
3069void thread_list__delete(struct list_head *list)
3070{
3071	struct thread_list *pos, *next;
3072
3073	list_for_each_entry_safe(pos, next, list, list) {
3074		thread__zput(pos->thread);
3075		list_del(&pos->list);
3076		free(pos);
3077	}
3078}
3079
3080pid_t machine__get_current_tid(struct machine *machine, int cpu)
3081{
3082	if (cpu < 0 || (size_t)cpu >= machine->current_tid_sz)
3083		return -1;
3084
3085	return machine->current_tid[cpu];
3086}
3087
3088int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
3089			     pid_t tid)
3090{
3091	struct thread *thread;
3092	const pid_t init_val = -1;
3093
3094	if (cpu < 0)
3095		return -EINVAL;
3096
3097	if (realloc_array_as_needed(machine->current_tid,
3098				    machine->current_tid_sz,
3099				    (unsigned int)cpu,
3100				    &init_val))
3101		return -ENOMEM;
3102
3103	machine->current_tid[cpu] = tid;
3104
3105	thread = machine__findnew_thread(machine, pid, tid);
3106	if (!thread)
3107		return -ENOMEM;
3108
3109	thread__set_cpu(thread, cpu);
3110	thread__put(thread);
3111
3112	return 0;
3113}
3114
3115/*
3116 * Compares the raw arch string. N.B. see instead perf_env__arch() or
3117 * machine__normalized_is() if a normalized arch is needed.
3118 */
3119bool machine__is(struct machine *machine, const char *arch)
3120{
3121	return machine && !strcmp(perf_env__raw_arch(machine->env), arch);
3122}
3123
3124bool machine__normalized_is(struct machine *machine, const char *arch)
3125{
3126	return machine && !strcmp(perf_env__arch(machine->env), arch);
3127}
3128
3129int machine__nr_cpus_avail(struct machine *machine)
3130{
3131	return machine ? perf_env__nr_cpus_avail(machine->env) : 0;
3132}
3133
3134int machine__get_kernel_start(struct machine *machine)
3135{
3136	struct map *map = machine__kernel_map(machine);
3137	int err = 0;
3138
3139	/*
3140	 * The only addresses above 2^63 are kernel addresses of a 64-bit
3141	 * kernel.  Note that addresses are unsigned so that on a 32-bit system
3142	 * all addresses including kernel addresses are less than 2^32.  In
3143	 * that case (32-bit system), if the kernel mapping is unknown, all
3144	 * addresses will be assumed to be in user space - see
3145	 * machine__kernel_ip().
3146	 */
3147	machine->kernel_start = 1ULL << 63;
3148	if (map) {
3149		err = map__load(map);
3150		/*
3151		 * On x86_64, PTI entry trampolines are less than the
3152		 * start of kernel text, but still above 2^63. So leave
3153		 * kernel_start = 1ULL << 63 for x86_64.
3154		 */
3155		if (!err && !machine__is(machine, "x86_64"))
3156			machine->kernel_start = map__start(map);
3157	}
3158	return err;
3159}
3160
3161u8 machine__addr_cpumode(struct machine *machine, u8 cpumode, u64 addr)
3162{
3163	u8 addr_cpumode = cpumode;
3164	bool kernel_ip;
3165
3166	if (!machine->single_address_space)
3167		goto out;
3168
3169	kernel_ip = machine__kernel_ip(machine, addr);
3170	switch (cpumode) {
3171	case PERF_RECORD_MISC_KERNEL:
3172	case PERF_RECORD_MISC_USER:
3173		addr_cpumode = kernel_ip ? PERF_RECORD_MISC_KERNEL :
3174					   PERF_RECORD_MISC_USER;
3175		break;
3176	case PERF_RECORD_MISC_GUEST_KERNEL:
3177	case PERF_RECORD_MISC_GUEST_USER:
3178		addr_cpumode = kernel_ip ? PERF_RECORD_MISC_GUEST_KERNEL :
3179					   PERF_RECORD_MISC_GUEST_USER;
3180		break;
3181	default:
3182		break;
3183	}
3184out:
3185	return addr_cpumode;
3186}
3187
3188struct dso *machine__findnew_dso_id(struct machine *machine, const char *filename,
3189				    const struct dso_id *id)
3190{
3191	return dsos__findnew_id(&machine->dsos, filename, id);
3192}
3193
3194struct dso *machine__findnew_dso(struct machine *machine, const char *filename)
3195{
3196	return machine__findnew_dso_id(machine, filename, &dso_id_empty);
3197}
3198
3199char *machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
3200{
3201	struct machine *machine = vmachine;
3202	struct map *map;
3203	struct symbol *sym = machine__find_kernel_symbol(machine, *addrp, &map);
3204
3205	if (sym == NULL)
3206		return NULL;
3207
3208	*modp = __map__is_kmodule(map) ? (char *)dso__short_name(map__dso(map)) : NULL;
3209	*addrp = map__unmap_ip(map, sym->start);
3210	return sym->name;
3211}
3212
3213struct machine__for_each_dso_cb_args {
3214	struct machine *machine;
3215	machine__dso_t fn;
3216	void *priv;
3217};
3218
3219static int machine__for_each_dso_cb(struct dso *dso, void *data)
3220{
3221	struct machine__for_each_dso_cb_args *args = data;
3222
3223	return args->fn(dso, args->machine, args->priv);
3224}
3225
3226int machine__for_each_dso(struct machine *machine, machine__dso_t fn, void *priv)
3227{
3228	struct machine__for_each_dso_cb_args args = {
3229		.machine = machine,
3230		.fn = fn,
3231		.priv = priv,
3232	};
3233
3234	return dsos__for_each_dso(&machine->dsos, machine__for_each_dso_cb, &args);
3235}
3236
3237int machine__for_each_kernel_map(struct machine *machine, machine__map_t fn, void *priv)
3238{
3239	struct maps *maps = machine__kernel_maps(machine);
3240
3241	return maps__for_each_map(maps, fn, priv);
3242}
3243
3244bool machine__is_lock_function(struct machine *machine, u64 addr)
3245{
3246	if (!machine->sched.text_start) {
3247		struct map *kmap;
3248		struct symbol *sym = machine__find_kernel_symbol_by_name(machine, "__sched_text_start", &kmap);
3249
3250		if (!sym) {
3251			/* to avoid retry */
3252			machine->sched.text_start = 1;
3253			return false;
3254		}
3255
3256		machine->sched.text_start = map__unmap_ip(kmap, sym->start);
3257
3258		/* should not fail from here */
3259		sym = machine__find_kernel_symbol_by_name(machine, "__sched_text_end", &kmap);
3260		machine->sched.text_end = map__unmap_ip(kmap, sym->start);
3261
3262		sym = machine__find_kernel_symbol_by_name(machine, "__lock_text_start", &kmap);
3263		machine->lock.text_start = map__unmap_ip(kmap, sym->start);
3264
3265		sym = machine__find_kernel_symbol_by_name(machine, "__lock_text_end", &kmap);
3266		machine->lock.text_end = map__unmap_ip(kmap, sym->start);
3267
3268		sym = machine__find_kernel_symbol_by_name(machine, "__traceiter_contention_begin", &kmap);
3269		if (sym) {
3270			machine->traceiter.text_start = map__unmap_ip(kmap, sym->start);
3271			machine->traceiter.text_end = map__unmap_ip(kmap, sym->end);
3272		}
3273		sym = machine__find_kernel_symbol_by_name(machine, "trace_contention_begin", &kmap);
3274		if (sym) {
3275			machine->trace.text_start = map__unmap_ip(kmap, sym->start);
3276			machine->trace.text_end = map__unmap_ip(kmap, sym->end);
3277		}
3278	}
3279
3280	/* failed to get kernel symbols */
3281	if (machine->sched.text_start == 1)
3282		return false;
3283
3284	/* mutex and rwsem functions are in sched text section */
3285	if (machine->sched.text_start <= addr && addr < machine->sched.text_end)
3286		return true;
3287
3288	/* spinlock functions are in lock text section */
3289	if (machine->lock.text_start <= addr && addr < machine->lock.text_end)
3290		return true;
3291
3292	/* traceiter functions currently don't have their own section
3293	 * but we consider them lock functions
3294	 */
3295	if (machine->traceiter.text_start != 0) {
3296		if (machine->traceiter.text_start <= addr && addr < machine->traceiter.text_end)
3297			return true;
3298	}
3299
3300	if (machine->trace.text_start != 0) {
3301		if (machine->trace.text_start <= addr && addr < machine->trace.text_end)
3302			return true;
3303	}
3304
3305	return false;
3306}
3307
3308int machine__hit_all_dsos(struct machine *machine)
3309{
3310	return dsos__hit_all(&machine->dsos);
3311}