1 // SPDX-License-Identifier: GPL-2.0
16 #include "map_symbol.h"
18 #include "mem-events.h"
29 #include <sys/types.h>
33 #include "linux/hash.h"
35 #include "bpf-event.h"
36 #include <internal/lib.h> // page_size
38 #include "arm64-frame-pointer-unwind-support.h"
40 #include <linux/ctype.h>
41 #include <symbol/kallsyms.h>
42 #include <linux/mman.h>
43 #include <linux/string.h>
44 #include <linux/zalloc.h>
46 static struct dso *machine__kernel_dso(struct machine *machine)
48 return map__dso(machine->vmlinux_map);
51 static void dsos__init(struct dsos *dsos)
53 INIT_LIST_HEAD(&dsos->head);
55 init_rwsem(&dsos->lock);
58 static int machine__set_mmap_name(struct machine *machine)
60 if (machine__is_host(machine))
61 machine->mmap_name = strdup("[kernel.kallsyms]");
62 else if (machine__is_default_guest(machine))
63 machine->mmap_name = strdup("[guest.kernel.kallsyms]");
64 else if (asprintf(&machine->mmap_name, "[guest.kernel.kallsyms.%d]",
66 machine->mmap_name = NULL;
68 return machine->mmap_name ? 0 : -ENOMEM;
71 static void thread__set_guest_comm(struct thread *thread, pid_t pid)
75 snprintf(comm, sizeof(comm), "[guest/%d]", pid);
76 thread__set_comm(thread, comm, 0);
79 int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
83 memset(machine, 0, sizeof(*machine));
84 machine->kmaps = maps__new(machine);
85 if (machine->kmaps == NULL)
88 RB_CLEAR_NODE(&machine->rb_node);
89 dsos__init(&machine->dsos);
91 threads__init(&machine->threads);
93 machine->vdso_info = NULL;
98 machine->id_hdr_size = 0;
99 machine->kptr_restrict_warned = false;
100 machine->comm_exec = false;
101 machine->kernel_start = 0;
102 machine->vmlinux_map = NULL;
104 machine->root_dir = strdup(root_dir);
105 if (machine->root_dir == NULL)
108 if (machine__set_mmap_name(machine))
111 if (pid != HOST_KERNEL_ID) {
112 struct thread *thread = machine__findnew_thread(machine, -1,
118 thread__set_guest_comm(thread, pid);
122 machine->current_tid = NULL;
127 zfree(&machine->kmaps);
128 zfree(&machine->root_dir);
129 zfree(&machine->mmap_name);
134 struct machine *machine__new_host(void)
136 struct machine *machine = malloc(sizeof(*machine));
138 if (machine != NULL) {
139 machine__init(machine, "", HOST_KERNEL_ID);
141 if (machine__create_kernel_maps(machine) < 0)
151 struct machine *machine__new_kallsyms(void)
153 struct machine *machine = machine__new_host();
156 * 1) We should switch to machine__load_kallsyms(), i.e. not explicitly
157 * ask for not using the kcore parsing code, once this one is fixed
158 * to create a map per module.
160 if (machine && machine__load_kallsyms(machine, "/proc/kallsyms") <= 0) {
161 machine__delete(machine);
168 static void dsos__purge(struct dsos *dsos)
172 down_write(&dsos->lock);
174 list_for_each_entry_safe(pos, n, &dsos->head, node) {
175 RB_CLEAR_NODE(&pos->rb_node);
177 list_del_init(&pos->node);
181 up_write(&dsos->lock);
184 static void dsos__exit(struct dsos *dsos)
187 exit_rwsem(&dsos->lock);
190 void machine__delete_threads(struct machine *machine)
192 threads__remove_all_threads(&machine->threads);
195 void machine__exit(struct machine *machine)
200 machine__destroy_kernel_maps(machine);
201 maps__zput(machine->kmaps);
202 dsos__exit(&machine->dsos);
203 machine__exit_vdso(machine);
204 zfree(&machine->root_dir);
205 zfree(&machine->mmap_name);
206 zfree(&machine->current_tid);
207 zfree(&machine->kallsyms_filename);
209 threads__exit(&machine->threads);
212 void machine__delete(struct machine *machine)
215 machine__exit(machine);
220 void machines__init(struct machines *machines)
222 machine__init(&machines->host, "", HOST_KERNEL_ID);
223 machines->guests = RB_ROOT_CACHED;
226 void machines__exit(struct machines *machines)
228 machine__exit(&machines->host);
232 struct machine *machines__add(struct machines *machines, pid_t pid,
233 const char *root_dir)
235 struct rb_node **p = &machines->guests.rb_root.rb_node;
236 struct rb_node *parent = NULL;
237 struct machine *pos, *machine = malloc(sizeof(*machine));
238 bool leftmost = true;
243 if (machine__init(machine, root_dir, pid) != 0) {
250 pos = rb_entry(parent, struct machine, rb_node);
259 rb_link_node(&machine->rb_node, parent, p);
260 rb_insert_color_cached(&machine->rb_node, &machines->guests, leftmost);
262 machine->machines = machines;
267 void machines__set_comm_exec(struct machines *machines, bool comm_exec)
271 machines->host.comm_exec = comm_exec;
273 for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
274 struct machine *machine = rb_entry(nd, struct machine, rb_node);
276 machine->comm_exec = comm_exec;
280 struct machine *machines__find(struct machines *machines, pid_t pid)
282 struct rb_node **p = &machines->guests.rb_root.rb_node;
283 struct rb_node *parent = NULL;
284 struct machine *machine;
285 struct machine *default_machine = NULL;
287 if (pid == HOST_KERNEL_ID)
288 return &machines->host;
292 machine = rb_entry(parent, struct machine, rb_node);
293 if (pid < machine->pid)
295 else if (pid > machine->pid)
300 default_machine = machine;
303 return default_machine;
306 struct machine *machines__findnew(struct machines *machines, pid_t pid)
309 const char *root_dir = "";
310 struct machine *machine = machines__find(machines, pid);
312 if (machine && (machine->pid == pid))
315 if ((pid != HOST_KERNEL_ID) &&
316 (pid != DEFAULT_GUEST_KERNEL_ID) &&
317 (symbol_conf.guestmount)) {
318 sprintf(path, "%s/%d", symbol_conf.guestmount, pid);
319 if (access(path, R_OK)) {
320 static struct strlist *seen;
323 seen = strlist__new(NULL, NULL);
325 if (!strlist__has_entry(seen, path)) {
326 pr_err("Can't access file %s\n", path);
327 strlist__add(seen, path);
335 machine = machines__add(machines, pid, root_dir);
340 struct machine *machines__find_guest(struct machines *machines, pid_t pid)
342 struct machine *machine = machines__find(machines, pid);
345 machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
350 * A common case for KVM test programs is that the test program acts as the
351 * hypervisor, creating, running and destroying the virtual machine, and
352 * providing the guest object code from its own object code. In this case,
353 * the VM is not running an OS, but only the functions loaded into it by the
354 * hypervisor test program, and conveniently, loaded at the same virtual
357 * Normally to resolve addresses, MMAP events are needed to map addresses
358 * back to the object code and debug symbols for that object code.
360 * Currently, there is no way to get such mapping information from guests
361 * but, in the scenario described above, the guest has the same mappings
362 * as the hypervisor, so support for that scenario can be achieved.
364 * To support that, copy the host thread's maps to the guest thread's maps.
365 * Note, we do not discover the guest until we encounter a guest event,
366 * which works well because it is not until then that we know that the host
367 * thread's maps have been set up.
369 * This function returns the guest thread. Apart from keeping the data
370 * structures sane, using a thread belonging to the guest machine, instead
371 * of the host thread, allows it to have its own comm (refer
372 * thread__set_guest_comm()).
374 static struct thread *findnew_guest_code(struct machine *machine,
375 struct machine *host_machine,
378 struct thread *host_thread;
379 struct thread *thread;
385 thread = machine__findnew_thread(machine, -1, pid);
389 /* Assume maps are set up if there are any */
390 if (!maps__empty(thread__maps(thread)))
393 host_thread = machine__find_thread(host_machine, -1, pid);
397 thread__set_guest_comm(thread, pid);
400 * Guest code can be found in hypervisor process at the same address
403 err = maps__copy_from(thread__maps(thread), thread__maps(host_thread));
404 thread__put(host_thread);
411 thread__zput(thread);
415 struct thread *machines__findnew_guest_code(struct machines *machines, pid_t pid)
417 struct machine *host_machine = machines__find(machines, HOST_KERNEL_ID);
418 struct machine *machine = machines__findnew(machines, pid);
420 return findnew_guest_code(machine, host_machine, pid);
423 struct thread *machine__findnew_guest_code(struct machine *machine, pid_t pid)
425 struct machines *machines = machine->machines;
426 struct machine *host_machine;
431 host_machine = machines__find(machines, HOST_KERNEL_ID);
433 return findnew_guest_code(machine, host_machine, pid);
436 void machines__process_guests(struct machines *machines,
437 machine__process_t process, void *data)
441 for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
442 struct machine *pos = rb_entry(nd, struct machine, rb_node);
447 void machines__set_id_hdr_size(struct machines *machines, u16 id_hdr_size)
449 struct rb_node *node;
450 struct machine *machine;
452 machines->host.id_hdr_size = id_hdr_size;
454 for (node = rb_first_cached(&machines->guests); node;
455 node = rb_next(node)) {
456 machine = rb_entry(node, struct machine, rb_node);
457 machine->id_hdr_size = id_hdr_size;
463 static void machine__update_thread_pid(struct machine *machine,
464 struct thread *th, pid_t pid)
466 struct thread *leader;
468 if (pid == thread__pid(th) || pid == -1 || thread__pid(th) != -1)
471 thread__set_pid(th, pid);
473 if (thread__pid(th) == thread__tid(th))
476 leader = machine__findnew_thread(machine, thread__pid(th), thread__pid(th));
480 if (!thread__maps(leader))
481 thread__set_maps(leader, maps__new(machine));
483 if (!thread__maps(leader))
486 if (thread__maps(th) == thread__maps(leader))
489 if (thread__maps(th)) {
491 * Maps are created from MMAP events which provide the pid and
492 * tid. Consequently there never should be any maps on a thread
493 * with an unknown pid. Just print an error if there are.
495 if (!maps__empty(thread__maps(th)))
496 pr_err("Discarding thread maps for %d:%d\n",
497 thread__pid(th), thread__tid(th));
498 maps__put(thread__maps(th));
501 thread__set_maps(th, maps__get(thread__maps(leader)));
506 pr_err("Failed to join map groups for %d:%d\n", thread__pid(th), thread__tid(th));
511 * Caller must eventually drop thread->refcnt returned with a successful
512 * lookup/new thread inserted.
514 static struct thread *__machine__findnew_thread(struct machine *machine,
519 struct thread *th = threads__find(&machine->threads, tid);
523 machine__update_thread_pid(machine, th, pid);
529 th = threads__findnew(&machine->threads, pid, tid, &created);
532 * We have to initialize maps separately after rb tree is
535 * The reason is that we call machine__findnew_thread within
536 * thread__init_maps to find the thread leader and that would
537 * screwed the rb tree.
539 if (thread__init_maps(th, machine)) {
540 pr_err("Thread init failed thread %d\n", pid);
541 threads__remove(&machine->threads, th);
546 machine__update_thread_pid(machine, th, pid);
551 struct thread *machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid)
553 return __machine__findnew_thread(machine, pid, tid, /*create=*/true);
556 struct thread *machine__find_thread(struct machine *machine, pid_t pid,
559 return __machine__findnew_thread(machine, pid, tid, /*create=*/false);
563 * Threads are identified by pid and tid, and the idle task has pid == tid == 0.
564 * So here a single thread is created for that, but actually there is a separate
565 * idle task per cpu, so there should be one 'struct thread' per cpu, but there
566 * is only 1. That causes problems for some tools, requiring workarounds. For
567 * example get_idle_thread() in builtin-sched.c, or thread_stack__per_cpu().
569 struct thread *machine__idle_thread(struct machine *machine)
571 struct thread *thread = machine__findnew_thread(machine, 0, 0);
573 if (!thread || thread__set_comm(thread, "swapper", 0) ||
574 thread__set_namespaces(thread, 0, NULL))
575 pr_err("problem inserting idle task for machine pid %d\n", machine->pid);
580 struct comm *machine__thread_exec_comm(struct machine *machine,
581 struct thread *thread)
583 if (machine->comm_exec)
584 return thread__exec_comm(thread);
586 return thread__comm(thread);
589 int machine__process_comm_event(struct machine *machine, union perf_event *event,
590 struct perf_sample *sample)
592 struct thread *thread = machine__findnew_thread(machine,
595 bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
599 machine->comm_exec = true;
602 perf_event__fprintf_comm(event, stdout);
604 if (thread == NULL ||
605 __thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
606 dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
615 int machine__process_namespaces_event(struct machine *machine __maybe_unused,
616 union perf_event *event,
617 struct perf_sample *sample __maybe_unused)
619 struct thread *thread = machine__findnew_thread(machine,
620 event->namespaces.pid,
621 event->namespaces.tid);
624 WARN_ONCE(event->namespaces.nr_namespaces > NR_NAMESPACES,
625 "\nWARNING: kernel seems to support more namespaces than perf"
626 " tool.\nTry updating the perf tool..\n\n");
628 WARN_ONCE(event->namespaces.nr_namespaces < NR_NAMESPACES,
629 "\nWARNING: perf tool seems to support more namespaces than"
630 " the kernel.\nTry updating the kernel..\n\n");
633 perf_event__fprintf_namespaces(event, stdout);
635 if (thread == NULL ||
636 thread__set_namespaces(thread, sample->time, &event->namespaces)) {
637 dump_printf("problem processing PERF_RECORD_NAMESPACES, skipping event.\n");
646 int machine__process_cgroup_event(struct machine *machine,
647 union perf_event *event,
648 struct perf_sample *sample __maybe_unused)
653 perf_event__fprintf_cgroup(event, stdout);
655 cgrp = cgroup__findnew(machine->env, event->cgroup.id, event->cgroup.path);
662 int machine__process_lost_event(struct machine *machine __maybe_unused,
663 union perf_event *event, struct perf_sample *sample __maybe_unused)
665 dump_printf(": id:%" PRI_lu64 ": lost:%" PRI_lu64 "\n",
666 event->lost.id, event->lost.lost);
670 int machine__process_lost_samples_event(struct machine *machine __maybe_unused,
671 union perf_event *event, struct perf_sample *sample)
673 dump_printf(": id:%" PRIu64 ": lost samples :%" PRI_lu64 "\n",
674 sample->id, event->lost_samples.lost);
678 static struct dso *machine__findnew_module_dso(struct machine *machine,
680 const char *filename)
684 down_write(&machine->dsos.lock);
686 dso = __dsos__find(&machine->dsos, m->name, true);
688 dso = __dsos__addnew(&machine->dsos, m->name);
692 dso__set_module_info(dso, m, machine);
693 dso__set_long_name(dso, strdup(filename), true);
694 dso->kernel = DSO_SPACE__KERNEL;
699 up_write(&machine->dsos.lock);
703 int machine__process_aux_event(struct machine *machine __maybe_unused,
704 union perf_event *event)
707 perf_event__fprintf_aux(event, stdout);
711 int machine__process_itrace_start_event(struct machine *machine __maybe_unused,
712 union perf_event *event)
715 perf_event__fprintf_itrace_start(event, stdout);
719 int machine__process_aux_output_hw_id_event(struct machine *machine __maybe_unused,
720 union perf_event *event)
723 perf_event__fprintf_aux_output_hw_id(event, stdout);
727 int machine__process_switch_event(struct machine *machine __maybe_unused,
728 union perf_event *event)
731 perf_event__fprintf_switch(event, stdout);
735 static int machine__process_ksymbol_register(struct machine *machine,
736 union perf_event *event,
737 struct perf_sample *sample __maybe_unused)
741 struct map *map = maps__find(machine__kernel_maps(machine), event->ksymbol.addr);
745 dso = dso__new(event->ksymbol.name);
751 dso->kernel = DSO_SPACE__KERNEL;
752 map = map__new2(0, dso);
758 if (event->ksymbol.ksym_type == PERF_RECORD_KSYMBOL_TYPE_OOL) {
759 dso->binary_type = DSO_BINARY_TYPE__OOL;
760 dso->data.file_size = event->ksymbol.len;
761 dso__set_loaded(dso);
764 map__set_start(map, event->ksymbol.addr);
765 map__set_end(map, map__start(map) + event->ksymbol.len);
766 err = maps__insert(machine__kernel_maps(machine), map);
772 dso__set_loaded(dso);
774 if (is_bpf_image(event->ksymbol.name)) {
775 dso->binary_type = DSO_BINARY_TYPE__BPF_IMAGE;
776 dso__set_long_name(dso, "", false);
782 sym = symbol__new(map__map_ip(map, map__start(map)),
784 0, 0, event->ksymbol.name);
789 dso__insert_symbol(dso, sym);
795 static int machine__process_ksymbol_unregister(struct machine *machine,
796 union perf_event *event,
797 struct perf_sample *sample __maybe_unused)
802 map = maps__find(machine__kernel_maps(machine), event->ksymbol.addr);
806 if (!RC_CHK_EQUAL(map, machine->vmlinux_map))
807 maps__remove(machine__kernel_maps(machine), map);
809 struct dso *dso = map__dso(map);
811 sym = dso__find_symbol(dso, map__map_ip(map, map__start(map)));
813 dso__delete_symbol(dso, sym);
819 int machine__process_ksymbol(struct machine *machine __maybe_unused,
820 union perf_event *event,
821 struct perf_sample *sample)
824 perf_event__fprintf_ksymbol(event, stdout);
826 if (event->ksymbol.flags & PERF_RECORD_KSYMBOL_FLAGS_UNREGISTER)
827 return machine__process_ksymbol_unregister(machine, event,
829 return machine__process_ksymbol_register(machine, event, sample);
832 int machine__process_text_poke(struct machine *machine, union perf_event *event,
833 struct perf_sample *sample __maybe_unused)
835 struct map *map = maps__find(machine__kernel_maps(machine), event->text_poke.addr);
836 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
837 struct dso *dso = map ? map__dso(map) : NULL;
840 perf_event__fprintf_text_poke(event, machine, stdout);
842 if (!event->text_poke.new_len)
845 if (cpumode != PERF_RECORD_MISC_KERNEL) {
846 pr_debug("%s: unsupported cpumode - ignoring\n", __func__);
851 u8 *new_bytes = event->text_poke.bytes + event->text_poke.old_len;
855 * Kernel maps might be changed when loading symbols so loading
856 * must be done prior to using kernel maps.
859 ret = dso__data_write_cache_addr(dso, map, machine,
860 event->text_poke.addr,
862 event->text_poke.new_len);
863 if (ret != event->text_poke.new_len)
864 pr_debug("Failed to write kernel text poke at %#" PRI_lx64 "\n",
865 event->text_poke.addr);
867 pr_debug("Failed to find kernel text poke address map for %#" PRI_lx64 "\n",
868 event->text_poke.addr);
875 static struct map *machine__addnew_module_map(struct machine *machine, u64 start,
876 const char *filename)
878 struct map *map = NULL;
883 if (kmod_path__parse_name(&m, filename))
886 dso = machine__findnew_module_dso(machine, &m, filename);
890 map = map__new2(start, dso);
894 err = maps__insert(machine__kernel_maps(machine), map);
895 /* If maps__insert failed, return NULL. */
901 /* put the dso here, corresponding to machine__findnew_module_dso */
907 size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
910 size_t ret = __dsos__fprintf(&machines->host.dsos.head, fp);
912 for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
913 struct machine *pos = rb_entry(nd, struct machine, rb_node);
914 ret += __dsos__fprintf(&pos->dsos.head, fp);
920 size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
921 bool (skip)(struct dso *dso, int parm), int parm)
923 return __dsos__fprintf_buildid(&m->dsos.head, fp, skip, parm);
926 size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
927 bool (skip)(struct dso *dso, int parm), int parm)
930 size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
932 for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
933 struct machine *pos = rb_entry(nd, struct machine, rb_node);
934 ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
939 size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
943 struct dso *kdso = machine__kernel_dso(machine);
945 if (kdso->has_build_id) {
946 char filename[PATH_MAX];
947 if (dso__build_id_filename(kdso, filename, sizeof(filename),
949 printed += fprintf(fp, "[0] %s\n", filename);
952 for (i = 0; i < vmlinux_path__nr_entries; ++i)
953 printed += fprintf(fp, "[%d] %s\n",
954 i + kdso->has_build_id, vmlinux_path[i]);
959 struct machine_fprintf_cb_args {
964 static int machine_fprintf_cb(struct thread *thread, void *data)
966 struct machine_fprintf_cb_args *args = data;
968 /* TODO: handle fprintf errors. */
969 args->printed += thread__fprintf(thread, args->fp);
973 size_t machine__fprintf(struct machine *machine, FILE *fp)
975 struct machine_fprintf_cb_args args = {
979 size_t ret = fprintf(fp, "Threads: %zu\n", threads__nr(&machine->threads));
981 machine__for_each_thread(machine, machine_fprintf_cb, &args);
982 return ret + args.printed;
985 static struct dso *machine__get_kernel(struct machine *machine)
987 const char *vmlinux_name = machine->mmap_name;
990 if (machine__is_host(machine)) {
991 if (symbol_conf.vmlinux_name)
992 vmlinux_name = symbol_conf.vmlinux_name;
994 kernel = machine__findnew_kernel(machine, vmlinux_name,
995 "[kernel]", DSO_SPACE__KERNEL);
997 if (symbol_conf.default_guest_vmlinux_name)
998 vmlinux_name = symbol_conf.default_guest_vmlinux_name;
1000 kernel = machine__findnew_kernel(machine, vmlinux_name,
1002 DSO_SPACE__KERNEL_GUEST);
1005 if (kernel != NULL && (!kernel->has_build_id))
1006 dso__read_running_kernel_build_id(kernel, machine);
1011 void machine__get_kallsyms_filename(struct machine *machine, char *buf,
1014 if (machine__is_default_guest(machine))
1015 scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
1017 scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
1020 const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
1022 /* Figure out the start address of kernel map from /proc/kallsyms.
1023 * Returns the name of the start symbol in *symbol_name. Pass in NULL as
1024 * symbol_name if it's not that important.
1026 static int machine__get_running_kernel_start(struct machine *machine,
1027 const char **symbol_name,
1028 u64 *start, u64 *end)
1030 char filename[PATH_MAX];
1035 machine__get_kallsyms_filename(machine, filename, PATH_MAX);
1037 if (symbol__restricted_filename(filename, "/proc/kallsyms"))
1040 for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
1041 err = kallsyms__get_function_start(filename, name, &addr);
1050 *symbol_name = name;
1054 err = kallsyms__get_symbol_start(filename, "_edata", &addr);
1056 err = kallsyms__get_function_start(filename, "_etext", &addr);
1063 int machine__create_extra_kernel_map(struct machine *machine,
1065 struct extra_kernel_map *xm)
1071 map = map__new2(xm->start, kernel);
1075 map__set_end(map, xm->end);
1076 map__set_pgoff(map, xm->pgoff);
1078 kmap = map__kmap(map);
1080 strlcpy(kmap->name, xm->name, KMAP_NAME_LEN);
1082 err = maps__insert(machine__kernel_maps(machine), map);
1085 pr_debug2("Added extra kernel map %s %" PRIx64 "-%" PRIx64 "\n",
1086 kmap->name, map__start(map), map__end(map));
1094 static u64 find_entry_trampoline(struct dso *dso)
1096 /* Duplicates are removed so lookup all aliases */
1097 const char *syms[] = {
1098 "_entry_trampoline",
1099 "__entry_trampoline_start",
1100 "entry_SYSCALL_64_trampoline",
1102 struct symbol *sym = dso__first_symbol(dso);
1105 for (; sym; sym = dso__next_symbol(sym)) {
1106 if (sym->binding != STB_GLOBAL)
1108 for (i = 0; i < ARRAY_SIZE(syms); i++) {
1109 if (!strcmp(sym->name, syms[i]))
1118 * These values can be used for kernels that do not have symbols for the entry
1119 * trampolines in kallsyms.
1121 #define X86_64_CPU_ENTRY_AREA_PER_CPU 0xfffffe0000000000ULL
1122 #define X86_64_CPU_ENTRY_AREA_SIZE 0x2c000
1123 #define X86_64_ENTRY_TRAMPOLINE 0x6000
1125 struct machine__map_x86_64_entry_trampolines_args {
1130 static int machine__map_x86_64_entry_trampolines_cb(struct map *map, void *data)
1132 struct machine__map_x86_64_entry_trampolines_args *args = data;
1133 struct map *dest_map;
1134 struct kmap *kmap = __map__kmap(map);
1136 if (!kmap || !is_entry_trampoline(kmap->name))
1139 dest_map = maps__find(args->kmaps, map__pgoff(map));
1140 if (RC_CHK_ACCESS(dest_map) != RC_CHK_ACCESS(map))
1141 map__set_pgoff(map, map__map_ip(dest_map, map__pgoff(map)));
1148 /* Map x86_64 PTI entry trampolines */
1149 int machine__map_x86_64_entry_trampolines(struct machine *machine,
1152 struct machine__map_x86_64_entry_trampolines_args args = {
1153 .kmaps = machine__kernel_maps(machine),
1156 int nr_cpus_avail, cpu;
1160 * In the vmlinux case, pgoff is a virtual address which must now be
1161 * mapped to a vmlinux offset.
1163 maps__for_each_map(args.kmaps, machine__map_x86_64_entry_trampolines_cb, &args);
1165 if (args.found || machine->trampolines_mapped)
1168 pgoff = find_entry_trampoline(kernel);
1172 nr_cpus_avail = machine__nr_cpus_avail(machine);
1174 /* Add a 1 page map for each CPU's entry trampoline */
1175 for (cpu = 0; cpu < nr_cpus_avail; cpu++) {
1176 u64 va = X86_64_CPU_ENTRY_AREA_PER_CPU +
1177 cpu * X86_64_CPU_ENTRY_AREA_SIZE +
1178 X86_64_ENTRY_TRAMPOLINE;
1179 struct extra_kernel_map xm = {
1181 .end = va + page_size,
1185 strlcpy(xm.name, ENTRY_TRAMPOLINE_NAME, KMAP_NAME_LEN);
1187 if (machine__create_extra_kernel_map(machine, kernel, &xm) < 0)
1191 machine->trampolines_mapped = nr_cpus_avail;
1196 int __weak machine__create_extra_kernel_maps(struct machine *machine __maybe_unused,
1197 struct dso *kernel __maybe_unused)
1203 __machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
1205 /* In case of renewal the kernel map, destroy previous one */
1206 machine__destroy_kernel_maps(machine);
1208 map__put(machine->vmlinux_map);
1209 machine->vmlinux_map = map__new2(0, kernel);
1210 if (machine->vmlinux_map == NULL)
1213 map__set_mapping_type(machine->vmlinux_map, MAPPING_TYPE__IDENTITY);
1214 return maps__insert(machine__kernel_maps(machine), machine->vmlinux_map);
1217 void machine__destroy_kernel_maps(struct machine *machine)
1220 struct map *map = machine__kernel_map(machine);
1225 kmap = map__kmap(map);
1226 maps__remove(machine__kernel_maps(machine), map);
1227 if (kmap && kmap->ref_reloc_sym) {
1228 zfree((char **)&kmap->ref_reloc_sym->name);
1229 zfree(&kmap->ref_reloc_sym);
1232 map__zput(machine->vmlinux_map);
1235 int machines__create_guest_kernel_maps(struct machines *machines)
1238 struct dirent **namelist = NULL;
1240 char path[PATH_MAX];
1244 if (symbol_conf.default_guest_vmlinux_name ||
1245 symbol_conf.default_guest_modules ||
1246 symbol_conf.default_guest_kallsyms) {
1247 machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
1250 if (symbol_conf.guestmount) {
1251 items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
1254 for (i = 0; i < items; i++) {
1255 if (!isdigit(namelist[i]->d_name[0])) {
1256 /* Filter out . and .. */
1259 pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
1260 if ((*endp != '\0') ||
1261 (endp == namelist[i]->d_name) ||
1262 (errno == ERANGE)) {
1263 pr_debug("invalid directory (%s). Skipping.\n",
1264 namelist[i]->d_name);
1267 sprintf(path, "%s/%s/proc/kallsyms",
1268 symbol_conf.guestmount,
1269 namelist[i]->d_name);
1270 ret = access(path, R_OK);
1272 pr_debug("Can't access file %s\n", path);
1275 machines__create_kernel_maps(machines, pid);
1284 void machines__destroy_kernel_maps(struct machines *machines)
1286 struct rb_node *next = rb_first_cached(&machines->guests);
1288 machine__destroy_kernel_maps(&machines->host);
1291 struct machine *pos = rb_entry(next, struct machine, rb_node);
1293 next = rb_next(&pos->rb_node);
1294 rb_erase_cached(&pos->rb_node, &machines->guests);
1295 machine__delete(pos);
1299 int machines__create_kernel_maps(struct machines *machines, pid_t pid)
1301 struct machine *machine = machines__findnew(machines, pid);
1303 if (machine == NULL)
1306 return machine__create_kernel_maps(machine);
1309 int machine__load_kallsyms(struct machine *machine, const char *filename)
1311 struct map *map = machine__kernel_map(machine);
1312 struct dso *dso = map__dso(map);
1313 int ret = __dso__load_kallsyms(dso, filename, map, true);
1316 dso__set_loaded(dso);
1318 * Since /proc/kallsyms will have multiple sessions for the
1319 * kernel, with modules between them, fixup the end of all
1322 maps__fixup_end(machine__kernel_maps(machine));
1328 int machine__load_vmlinux_path(struct machine *machine)
1330 struct map *map = machine__kernel_map(machine);
1331 struct dso *dso = map__dso(map);
1332 int ret = dso__load_vmlinux_path(dso, map);
1335 dso__set_loaded(dso);
1340 static char *get_kernel_version(const char *root_dir)
1342 char version[PATH_MAX];
1345 const char *prefix = "Linux version ";
1347 sprintf(version, "%s/proc/version", root_dir);
1348 file = fopen(version, "r");
1352 tmp = fgets(version, sizeof(version), file);
1357 name = strstr(version, prefix);
1360 name += strlen(prefix);
1361 tmp = strchr(name, ' ');
1365 return strdup(name);
1368 static bool is_kmod_dso(struct dso *dso)
1370 return dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
1371 dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE;
1374 static int maps__set_module_path(struct maps *maps, const char *path, struct kmod_path *m)
1378 struct map *map = maps__find_by_name(maps, m->name);
1383 long_name = strdup(path);
1384 if (long_name == NULL) {
1389 dso = map__dso(map);
1390 dso__set_long_name(dso, long_name, true);
1391 dso__kernel_module_get_build_id(dso, "");
1394 * Full name could reveal us kmod compression, so
1395 * we need to update the symtab_type if needed.
1397 if (m->comp && is_kmod_dso(dso)) {
1399 dso->comp = m->comp;
1405 static int maps__set_modules_path_dir(struct maps *maps, const char *dir_name, int depth)
1407 struct dirent *dent;
1408 DIR *dir = opendir(dir_name);
1412 pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
1416 while ((dent = readdir(dir)) != NULL) {
1417 char path[PATH_MAX];
1420 /*sshfs might return bad dent->d_type, so we have to stat*/
1421 path__join(path, sizeof(path), dir_name, dent->d_name);
1422 if (stat(path, &st))
1425 if (S_ISDIR(st.st_mode)) {
1426 if (!strcmp(dent->d_name, ".") ||
1427 !strcmp(dent->d_name, ".."))
1430 /* Do not follow top-level source and build symlinks */
1432 if (!strcmp(dent->d_name, "source") ||
1433 !strcmp(dent->d_name, "build"))
1437 ret = maps__set_modules_path_dir(maps, path, depth + 1);
1443 ret = kmod_path__parse_name(&m, dent->d_name);
1448 ret = maps__set_module_path(maps, path, &m);
1462 static int machine__set_modules_path(struct machine *machine)
1465 char modules_path[PATH_MAX];
1467 version = get_kernel_version(machine->root_dir);
1471 snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
1472 machine->root_dir, version);
1475 return maps__set_modules_path_dir(machine__kernel_maps(machine), modules_path, 0);
1477 int __weak arch__fix_module_text_start(u64 *start __maybe_unused,
1478 u64 *size __maybe_unused,
1479 const char *name __maybe_unused)
1484 static int machine__create_module(void *arg, const char *name, u64 start,
1487 struct machine *machine = arg;
1490 if (arch__fix_module_text_start(&start, &size, name) < 0)
1493 map = machine__addnew_module_map(machine, start, name);
1496 map__set_end(map, start + size);
1498 dso__kernel_module_get_build_id(map__dso(map), machine->root_dir);
1503 static int machine__create_modules(struct machine *machine)
1505 const char *modules;
1506 char path[PATH_MAX];
1508 if (machine__is_default_guest(machine)) {
1509 modules = symbol_conf.default_guest_modules;
1511 snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
1515 if (symbol__restricted_filename(modules, "/proc/modules"))
1518 if (modules__parse(modules, machine, machine__create_module))
1521 if (!machine__set_modules_path(machine))
1524 pr_debug("Problems setting modules path maps, continuing anyway...\n");
1529 static void machine__set_kernel_mmap(struct machine *machine,
1532 map__set_start(machine->vmlinux_map, start);
1533 map__set_end(machine->vmlinux_map, end);
1535 * Be a bit paranoid here, some perf.data file came with
1536 * a zero sized synthesized MMAP event for the kernel.
1538 if (start == 0 && end == 0)
1539 map__set_end(machine->vmlinux_map, ~0ULL);
1542 static int machine__update_kernel_mmap(struct machine *machine,
1545 struct map *orig, *updated;
1548 orig = machine->vmlinux_map;
1549 updated = map__get(orig);
1551 machine->vmlinux_map = updated;
1552 machine__set_kernel_mmap(machine, start, end);
1553 maps__remove(machine__kernel_maps(machine), orig);
1554 err = maps__insert(machine__kernel_maps(machine), updated);
1560 int machine__create_kernel_maps(struct machine *machine)
1562 struct dso *kernel = machine__get_kernel(machine);
1563 const char *name = NULL;
1564 u64 start = 0, end = ~0ULL;
1570 ret = __machine__create_kernel_maps(machine, kernel);
1574 if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
1575 if (machine__is_host(machine))
1576 pr_debug("Problems creating module maps, "
1577 "continuing anyway...\n");
1579 pr_debug("Problems creating module maps for guest %d, "
1580 "continuing anyway...\n", machine->pid);
1583 if (!machine__get_running_kernel_start(machine, &name, &start, &end)) {
1585 map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map, name, start)) {
1586 machine__destroy_kernel_maps(machine);
1592 * we have a real start address now, so re-order the kmaps
1593 * assume it's the last in the kmaps
1595 ret = machine__update_kernel_mmap(machine, start, end);
1600 if (machine__create_extra_kernel_maps(machine, kernel))
1601 pr_debug("Problems creating extra kernel maps, continuing anyway...\n");
1604 /* update end address of the kernel map using adjacent module address */
1605 struct map *next = maps__find_next_entry(machine__kernel_maps(machine),
1606 machine__kernel_map(machine));
1609 machine__set_kernel_mmap(machine, start, map__start(next));
1619 static bool machine__uses_kcore(struct machine *machine)
1623 list_for_each_entry(dso, &machine->dsos.head, node) {
1624 if (dso__is_kcore(dso))
1631 static bool perf_event__is_extra_kernel_mmap(struct machine *machine,
1632 struct extra_kernel_map *xm)
1634 return machine__is(machine, "x86_64") &&
1635 is_entry_trampoline(xm->name);
1638 static int machine__process_extra_kernel_map(struct machine *machine,
1639 struct extra_kernel_map *xm)
1641 struct dso *kernel = machine__kernel_dso(machine);
1646 return machine__create_extra_kernel_map(machine, kernel, xm);
1649 static int machine__process_kernel_mmap_event(struct machine *machine,
1650 struct extra_kernel_map *xm,
1651 struct build_id *bid)
1653 enum dso_space_type dso_space;
1654 bool is_kernel_mmap;
1655 const char *mmap_name = machine->mmap_name;
1657 /* If we have maps from kcore then we do not need or want any others */
1658 if (machine__uses_kcore(machine))
1661 if (machine__is_host(machine))
1662 dso_space = DSO_SPACE__KERNEL;
1664 dso_space = DSO_SPACE__KERNEL_GUEST;
1666 is_kernel_mmap = memcmp(xm->name, mmap_name, strlen(mmap_name) - 1) == 0;
1667 if (!is_kernel_mmap && !machine__is_host(machine)) {
1669 * If the event was recorded inside the guest and injected into
1670 * the host perf.data file, then it will match a host mmap_name,
1671 * so try that - see machine__set_mmap_name().
1673 mmap_name = "[kernel.kallsyms]";
1674 is_kernel_mmap = memcmp(xm->name, mmap_name, strlen(mmap_name) - 1) == 0;
1676 if (xm->name[0] == '/' ||
1677 (!is_kernel_mmap && xm->name[0] == '[')) {
1678 struct map *map = machine__addnew_module_map(machine, xm->start, xm->name);
1683 map__set_end(map, map__start(map) + xm->end - xm->start);
1685 if (build_id__is_defined(bid))
1686 dso__set_build_id(map__dso(map), bid);
1689 } else if (is_kernel_mmap) {
1690 const char *symbol_name = xm->name + strlen(mmap_name);
1692 * Should be there already, from the build-id table in
1695 struct dso *kernel = NULL;
1698 down_read(&machine->dsos.lock);
1700 list_for_each_entry(dso, &machine->dsos.head, node) {
1703 * The cpumode passed to is_kernel_module is not the
1704 * cpumode of *this* event. If we insist on passing
1705 * correct cpumode to is_kernel_module, we should
1706 * record the cpumode when we adding this dso to the
1709 * However we don't really need passing correct
1710 * cpumode. We know the correct cpumode must be kernel
1711 * mode (if not, we should not link it onto kernel_dsos
1714 * Therefore, we pass PERF_RECORD_MISC_CPUMODE_UNKNOWN.
1715 * is_kernel_module() treats it as a kernel cpumode.
1719 is_kernel_module(dso->long_name,
1720 PERF_RECORD_MISC_CPUMODE_UNKNOWN))
1724 kernel = dso__get(dso);
1728 up_read(&machine->dsos.lock);
1731 kernel = machine__findnew_dso(machine, machine->mmap_name);
1735 kernel->kernel = dso_space;
1736 if (__machine__create_kernel_maps(machine, kernel) < 0) {
1741 if (strstr(kernel->long_name, "vmlinux"))
1742 dso__set_short_name(kernel, "[kernel.vmlinux]", false);
1744 if (machine__update_kernel_mmap(machine, xm->start, xm->end) < 0) {
1749 if (build_id__is_defined(bid))
1750 dso__set_build_id(kernel, bid);
1753 * Avoid using a zero address (kptr_restrict) for the ref reloc
1754 * symbol. Effectively having zero here means that at record
1755 * time /proc/sys/kernel/kptr_restrict was non zero.
1757 if (xm->pgoff != 0) {
1758 map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map,
1763 if (machine__is_default_guest(machine)) {
1765 * preload dso of guest kernel and modules
1767 dso__load(kernel, machine__kernel_map(machine));
1770 } else if (perf_event__is_extra_kernel_mmap(machine, xm)) {
1771 return machine__process_extra_kernel_map(machine, xm);
1778 int machine__process_mmap2_event(struct machine *machine,
1779 union perf_event *event,
1780 struct perf_sample *sample)
1782 struct thread *thread;
1784 struct dso_id dso_id = {
1785 .maj = event->mmap2.maj,
1786 .min = event->mmap2.min,
1787 .ino = event->mmap2.ino,
1788 .ino_generation = event->mmap2.ino_generation,
1790 struct build_id __bid, *bid = NULL;
1794 perf_event__fprintf_mmap2(event, stdout);
1796 if (event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID) {
1798 build_id__init(bid, event->mmap2.build_id, event->mmap2.build_id_size);
1801 if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1802 sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1803 struct extra_kernel_map xm = {
1804 .start = event->mmap2.start,
1805 .end = event->mmap2.start + event->mmap2.len,
1806 .pgoff = event->mmap2.pgoff,
1809 strlcpy(xm.name, event->mmap2.filename, KMAP_NAME_LEN);
1810 ret = machine__process_kernel_mmap_event(machine, &xm, bid);
1816 thread = machine__findnew_thread(machine, event->mmap2.pid,
1821 map = map__new(machine, event->mmap2.start,
1822 event->mmap2.len, event->mmap2.pgoff,
1823 &dso_id, event->mmap2.prot,
1824 event->mmap2.flags, bid,
1825 event->mmap2.filename, thread);
1828 goto out_problem_map;
1830 ret = thread__insert_map(thread, map);
1832 goto out_problem_insert;
1834 thread__put(thread);
1841 thread__put(thread);
1843 dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
1847 int machine__process_mmap_event(struct machine *machine, union perf_event *event,
1848 struct perf_sample *sample)
1850 struct thread *thread;
1856 perf_event__fprintf_mmap(event, stdout);
1858 if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1859 sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1860 struct extra_kernel_map xm = {
1861 .start = event->mmap.start,
1862 .end = event->mmap.start + event->mmap.len,
1863 .pgoff = event->mmap.pgoff,
1866 strlcpy(xm.name, event->mmap.filename, KMAP_NAME_LEN);
1867 ret = machine__process_kernel_mmap_event(machine, &xm, NULL);
1873 thread = machine__findnew_thread(machine, event->mmap.pid,
1878 if (!(event->header.misc & PERF_RECORD_MISC_MMAP_DATA))
1881 map = map__new(machine, event->mmap.start,
1882 event->mmap.len, event->mmap.pgoff,
1883 NULL, prot, 0, NULL, event->mmap.filename, thread);
1886 goto out_problem_map;
1888 ret = thread__insert_map(thread, map);
1890 goto out_problem_insert;
1892 thread__put(thread);
1899 thread__put(thread);
1901 dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
1905 void machine__remove_thread(struct machine *machine, struct thread *th)
1907 return threads__remove(&machine->threads, th);
1910 int machine__process_fork_event(struct machine *machine, union perf_event *event,
1911 struct perf_sample *sample)
1913 struct thread *thread = machine__find_thread(machine,
1916 struct thread *parent = machine__findnew_thread(machine,
1919 bool do_maps_clone = true;
1923 perf_event__fprintf_task(event, stdout);
1926 * There may be an existing thread that is not actually the parent,
1927 * either because we are processing events out of order, or because the
1928 * (fork) event that would have removed the thread was lost. Assume the
1929 * latter case and continue on as best we can.
1931 if (thread__pid(parent) != (pid_t)event->fork.ppid) {
1932 dump_printf("removing erroneous parent thread %d/%d\n",
1933 thread__pid(parent), thread__tid(parent));
1934 machine__remove_thread(machine, parent);
1935 thread__put(parent);
1936 parent = machine__findnew_thread(machine, event->fork.ppid,
1940 /* if a thread currently exists for the thread id remove it */
1941 if (thread != NULL) {
1942 machine__remove_thread(machine, thread);
1943 thread__put(thread);
1946 thread = machine__findnew_thread(machine, event->fork.pid,
1949 * When synthesizing FORK events, we are trying to create thread
1950 * objects for the already running tasks on the machine.
1952 * Normally, for a kernel FORK event, we want to clone the parent's
1953 * maps because that is what the kernel just did.
1955 * But when synthesizing, this should not be done. If we do, we end up
1956 * with overlapping maps as we process the synthesized MMAP2 events that
1957 * get delivered shortly thereafter.
1959 * Use the FORK event misc flags in an internal way to signal this
1960 * situation, so we can elide the map clone when appropriate.
1962 if (event->fork.header.misc & PERF_RECORD_MISC_FORK_EXEC)
1963 do_maps_clone = false;
1965 if (thread == NULL || parent == NULL ||
1966 thread__fork(thread, parent, sample->time, do_maps_clone) < 0) {
1967 dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
1970 thread__put(thread);
1971 thread__put(parent);
1976 int machine__process_exit_event(struct machine *machine, union perf_event *event,
1977 struct perf_sample *sample __maybe_unused)
1979 struct thread *thread = machine__find_thread(machine,
1984 perf_event__fprintf_task(event, stdout);
1986 if (thread != NULL) {
1987 if (symbol_conf.keep_exited_threads)
1988 thread__set_exited(thread, /*exited=*/true);
1990 machine__remove_thread(machine, thread);
1992 thread__put(thread);
1996 int machine__process_event(struct machine *machine, union perf_event *event,
1997 struct perf_sample *sample)
2001 switch (event->header.type) {
2002 case PERF_RECORD_COMM:
2003 ret = machine__process_comm_event(machine, event, sample); break;
2004 case PERF_RECORD_MMAP:
2005 ret = machine__process_mmap_event(machine, event, sample); break;
2006 case PERF_RECORD_NAMESPACES:
2007 ret = machine__process_namespaces_event(machine, event, sample); break;
2008 case PERF_RECORD_CGROUP:
2009 ret = machine__process_cgroup_event(machine, event, sample); break;
2010 case PERF_RECORD_MMAP2:
2011 ret = machine__process_mmap2_event(machine, event, sample); break;
2012 case PERF_RECORD_FORK:
2013 ret = machine__process_fork_event(machine, event, sample); break;
2014 case PERF_RECORD_EXIT:
2015 ret = machine__process_exit_event(machine, event, sample); break;
2016 case PERF_RECORD_LOST:
2017 ret = machine__process_lost_event(machine, event, sample); break;
2018 case PERF_RECORD_AUX:
2019 ret = machine__process_aux_event(machine, event); break;
2020 case PERF_RECORD_ITRACE_START:
2021 ret = machine__process_itrace_start_event(machine, event); break;
2022 case PERF_RECORD_LOST_SAMPLES:
2023 ret = machine__process_lost_samples_event(machine, event, sample); break;
2024 case PERF_RECORD_SWITCH:
2025 case PERF_RECORD_SWITCH_CPU_WIDE:
2026 ret = machine__process_switch_event(machine, event); break;
2027 case PERF_RECORD_KSYMBOL:
2028 ret = machine__process_ksymbol(machine, event, sample); break;
2029 case PERF_RECORD_BPF_EVENT:
2030 ret = machine__process_bpf(machine, event, sample); break;
2031 case PERF_RECORD_TEXT_POKE:
2032 ret = machine__process_text_poke(machine, event, sample); break;
2033 case PERF_RECORD_AUX_OUTPUT_HW_ID:
2034 ret = machine__process_aux_output_hw_id_event(machine, event); break;
2043 static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
2045 return regexec(regex, sym->name, 0, NULL, 0) == 0;
2048 static void ip__resolve_ams(struct thread *thread,
2049 struct addr_map_symbol *ams,
2052 struct addr_location al;
2054 addr_location__init(&al);
2056 * We cannot use the header.misc hint to determine whether a
2057 * branch stack address is user, kernel, guest, hypervisor.
2058 * Branches may straddle the kernel/user/hypervisor boundaries.
2059 * Thus, we have to try consecutively until we find a match
2060 * or else, the symbol is unknown
2062 thread__find_cpumode_addr_location(thread, ip, &al);
2065 ams->al_addr = al.addr;
2066 ams->al_level = al.level;
2067 ams->ms.maps = maps__get(al.maps);
2068 ams->ms.sym = al.sym;
2069 ams->ms.map = map__get(al.map);
2071 ams->data_page_size = 0;
2072 addr_location__exit(&al);
2075 static void ip__resolve_data(struct thread *thread,
2076 u8 m, struct addr_map_symbol *ams,
2077 u64 addr, u64 phys_addr, u64 daddr_page_size)
2079 struct addr_location al;
2081 addr_location__init(&al);
2083 thread__find_symbol(thread, m, addr, &al);
2086 ams->al_addr = al.addr;
2087 ams->al_level = al.level;
2088 ams->ms.maps = maps__get(al.maps);
2089 ams->ms.sym = al.sym;
2090 ams->ms.map = map__get(al.map);
2091 ams->phys_addr = phys_addr;
2092 ams->data_page_size = daddr_page_size;
2093 addr_location__exit(&al);
2096 struct mem_info *sample__resolve_mem(struct perf_sample *sample,
2097 struct addr_location *al)
2099 struct mem_info *mi = mem_info__new();
2104 ip__resolve_ams(al->thread, &mi->iaddr, sample->ip);
2105 ip__resolve_data(al->thread, al->cpumode, &mi->daddr,
2106 sample->addr, sample->phys_addr,
2107 sample->data_page_size);
2108 mi->data_src.val = sample->data_src;
2113 static char *callchain_srcline(struct map_symbol *ms, u64 ip)
2115 struct map *map = ms->map;
2116 char *srcline = NULL;
2119 if (!map || callchain_param.key == CCKEY_FUNCTION)
2122 dso = map__dso(map);
2123 srcline = srcline__tree_find(&dso->srclines, ip);
2125 bool show_sym = false;
2126 bool show_addr = callchain_param.key == CCKEY_ADDRESS;
2128 srcline = get_srcline(dso, map__rip_2objdump(map, ip),
2129 ms->sym, show_sym, show_addr, ip);
2130 srcline__tree_insert(&dso->srclines, ip, srcline);
2141 static int add_callchain_ip(struct thread *thread,
2142 struct callchain_cursor *cursor,
2143 struct symbol **parent,
2144 struct addr_location *root_al,
2148 struct branch_flags *flags,
2149 struct iterations *iter,
2152 struct map_symbol ms = {};
2153 struct addr_location al;
2154 int nr_loop_iter = 0, err = 0;
2155 u64 iter_cycles = 0;
2156 const char *srcline = NULL;
2158 addr_location__init(&al);
2163 thread__find_cpumode_addr_location(thread, ip, &al);
2165 if (ip >= PERF_CONTEXT_MAX) {
2167 case PERF_CONTEXT_HV:
2168 *cpumode = PERF_RECORD_MISC_HYPERVISOR;
2170 case PERF_CONTEXT_KERNEL:
2171 *cpumode = PERF_RECORD_MISC_KERNEL;
2173 case PERF_CONTEXT_USER:
2174 *cpumode = PERF_RECORD_MISC_USER;
2177 pr_debug("invalid callchain context: "
2178 "%"PRId64"\n", (s64) ip);
2180 * It seems the callchain is corrupted.
2183 callchain_cursor_reset(cursor);
2189 thread__find_symbol(thread, *cpumode, ip, &al);
2192 if (al.sym != NULL) {
2193 if (perf_hpp_list.parent && !*parent &&
2194 symbol__match_regex(al.sym, &parent_regex))
2196 else if (have_ignore_callees && root_al &&
2197 symbol__match_regex(al.sym, &ignore_callees_regex)) {
2198 /* Treat this symbol as the root,
2199 forgetting its callees. */
2200 addr_location__copy(root_al, &al);
2201 callchain_cursor_reset(cursor);
2205 if (symbol_conf.hide_unresolved && al.sym == NULL)
2209 nr_loop_iter = iter->nr_loop_iter;
2210 iter_cycles = iter->cycles;
2213 ms.maps = maps__get(al.maps);
2214 ms.map = map__get(al.map);
2216 srcline = callchain_srcline(&ms, al.addr);
2217 err = callchain_cursor_append(cursor, ip, &ms,
2218 branch, flags, nr_loop_iter,
2219 iter_cycles, branch_from, srcline);
2221 addr_location__exit(&al);
2222 map_symbol__exit(&ms);
2226 struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
2227 struct addr_location *al)
2230 const struct branch_stack *bs = sample->branch_stack;
2231 struct branch_entry *entries = perf_sample__branch_entries(sample);
2232 struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
2237 for (i = 0; i < bs->nr; i++) {
2238 ip__resolve_ams(al->thread, &bi[i].to, entries[i].to);
2239 ip__resolve_ams(al->thread, &bi[i].from, entries[i].from);
2240 bi[i].flags = entries[i].flags;
2245 static void save_iterations(struct iterations *iter,
2246 struct branch_entry *be, int nr)
2250 iter->nr_loop_iter++;
2253 for (i = 0; i < nr; i++)
2254 iter->cycles += be[i].flags.cycles;
2259 #define NO_ENTRY 0xff
2261 #define PERF_MAX_BRANCH_DEPTH 127
2264 static int remove_loops(struct branch_entry *l, int nr,
2265 struct iterations *iter)
2268 unsigned char chash[CHASHSZ];
2270 memset(chash, NO_ENTRY, sizeof(chash));
2272 BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
2274 for (i = 0; i < nr; i++) {
2275 int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
2277 /* no collision handling for now */
2278 if (chash[h] == NO_ENTRY) {
2280 } else if (l[chash[h]].from == l[i].from) {
2281 bool is_loop = true;
2282 /* check if it is a real loop */
2284 for (j = chash[h]; j < i && i + off < nr; j++, off++)
2285 if (l[j].from != l[i + off].from) {
2292 save_iterations(iter + i + off,
2295 memmove(iter + i, iter + i + off,
2298 memmove(l + i, l + i + off,
2309 static int lbr_callchain_add_kernel_ip(struct thread *thread,
2310 struct callchain_cursor *cursor,
2311 struct perf_sample *sample,
2312 struct symbol **parent,
2313 struct addr_location *root_al,
2315 bool callee, int end)
2317 struct ip_callchain *chain = sample->callchain;
2318 u8 cpumode = PERF_RECORD_MISC_USER;
2322 for (i = 0; i < end + 1; i++) {
2323 err = add_callchain_ip(thread, cursor, parent,
2324 root_al, &cpumode, chain->ips[i],
2325 false, NULL, NULL, branch_from);
2332 for (i = end; i >= 0; i--) {
2333 err = add_callchain_ip(thread, cursor, parent,
2334 root_al, &cpumode, chain->ips[i],
2335 false, NULL, NULL, branch_from);
2343 static void save_lbr_cursor_node(struct thread *thread,
2344 struct callchain_cursor *cursor,
2347 struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2352 if (cursor->pos == cursor->nr) {
2353 lbr_stitch->prev_lbr_cursor[idx].valid = false;
2358 cursor->curr = cursor->first;
2360 cursor->curr = cursor->curr->next;
2361 memcpy(&lbr_stitch->prev_lbr_cursor[idx], cursor->curr,
2362 sizeof(struct callchain_cursor_node));
2364 lbr_stitch->prev_lbr_cursor[idx].valid = true;
2368 static int lbr_callchain_add_lbr_ip(struct thread *thread,
2369 struct callchain_cursor *cursor,
2370 struct perf_sample *sample,
2371 struct symbol **parent,
2372 struct addr_location *root_al,
2376 struct branch_stack *lbr_stack = sample->branch_stack;
2377 struct branch_entry *entries = perf_sample__branch_entries(sample);
2378 u8 cpumode = PERF_RECORD_MISC_USER;
2379 int lbr_nr = lbr_stack->nr;
2380 struct branch_flags *flags;
2385 * The curr and pos are not used in writing session. They are cleared
2386 * in callchain_cursor_commit() when the writing session is closed.
2387 * Using curr and pos to track the current cursor node.
2389 if (thread__lbr_stitch(thread)) {
2390 cursor->curr = NULL;
2391 cursor->pos = cursor->nr;
2393 cursor->curr = cursor->first;
2394 for (i = 0; i < (int)(cursor->nr - 1); i++)
2395 cursor->curr = cursor->curr->next;
2400 /* Add LBR ip from first entries.to */
2402 flags = &entries[0].flags;
2403 *branch_from = entries[0].from;
2404 err = add_callchain_ip(thread, cursor, parent,
2405 root_al, &cpumode, ip,
2412 * The number of cursor node increases.
2413 * Move the current cursor node.
2414 * But does not need to save current cursor node for entry 0.
2415 * It's impossible to stitch the whole LBRs of previous sample.
2417 if (thread__lbr_stitch(thread) && (cursor->pos != cursor->nr)) {
2419 cursor->curr = cursor->first;
2421 cursor->curr = cursor->curr->next;
2425 /* Add LBR ip from entries.from one by one. */
2426 for (i = 0; i < lbr_nr; i++) {
2427 ip = entries[i].from;
2428 flags = &entries[i].flags;
2429 err = add_callchain_ip(thread, cursor, parent,
2430 root_al, &cpumode, ip,
2435 save_lbr_cursor_node(thread, cursor, i);
2440 /* Add LBR ip from entries.from one by one. */
2441 for (i = lbr_nr - 1; i >= 0; i--) {
2442 ip = entries[i].from;
2443 flags = &entries[i].flags;
2444 err = add_callchain_ip(thread, cursor, parent,
2445 root_al, &cpumode, ip,
2450 save_lbr_cursor_node(thread, cursor, i);
2454 /* Add LBR ip from first entries.to */
2456 flags = &entries[0].flags;
2457 *branch_from = entries[0].from;
2458 err = add_callchain_ip(thread, cursor, parent,
2459 root_al, &cpumode, ip,
2469 static int lbr_callchain_add_stitched_lbr_ip(struct thread *thread,
2470 struct callchain_cursor *cursor)
2472 struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2473 struct callchain_cursor_node *cnode;
2474 struct stitch_list *stitch_node;
2477 list_for_each_entry(stitch_node, &lbr_stitch->lists, node) {
2478 cnode = &stitch_node->cursor;
2480 err = callchain_cursor_append(cursor, cnode->ip,
2483 &cnode->branch_flags,
2484 cnode->nr_loop_iter,
2494 static struct stitch_list *get_stitch_node(struct thread *thread)
2496 struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2497 struct stitch_list *stitch_node;
2499 if (!list_empty(&lbr_stitch->free_lists)) {
2500 stitch_node = list_first_entry(&lbr_stitch->free_lists,
2501 struct stitch_list, node);
2502 list_del(&stitch_node->node);
2507 return malloc(sizeof(struct stitch_list));
2510 static bool has_stitched_lbr(struct thread *thread,
2511 struct perf_sample *cur,
2512 struct perf_sample *prev,
2513 unsigned int max_lbr,
2516 struct branch_stack *cur_stack = cur->branch_stack;
2517 struct branch_entry *cur_entries = perf_sample__branch_entries(cur);
2518 struct branch_stack *prev_stack = prev->branch_stack;
2519 struct branch_entry *prev_entries = perf_sample__branch_entries(prev);
2520 struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2521 int i, j, nr_identical_branches = 0;
2522 struct stitch_list *stitch_node;
2523 u64 cur_base, distance;
2525 if (!cur_stack || !prev_stack)
2528 /* Find the physical index of the base-of-stack for current sample. */
2529 cur_base = max_lbr - cur_stack->nr + cur_stack->hw_idx + 1;
2531 distance = (prev_stack->hw_idx > cur_base) ? (prev_stack->hw_idx - cur_base) :
2532 (max_lbr + prev_stack->hw_idx - cur_base);
2533 /* Previous sample has shorter stack. Nothing can be stitched. */
2534 if (distance + 1 > prev_stack->nr)
2538 * Check if there are identical LBRs between two samples.
2539 * Identical LBRs must have same from, to and flags values. Also,
2540 * they have to be saved in the same LBR registers (same physical
2543 * Starts from the base-of-stack of current sample.
2545 for (i = distance, j = cur_stack->nr - 1; (i >= 0) && (j >= 0); i--, j--) {
2546 if ((prev_entries[i].from != cur_entries[j].from) ||
2547 (prev_entries[i].to != cur_entries[j].to) ||
2548 (prev_entries[i].flags.value != cur_entries[j].flags.value))
2550 nr_identical_branches++;
2553 if (!nr_identical_branches)
2557 * Save the LBRs between the base-of-stack of previous sample
2558 * and the base-of-stack of current sample into lbr_stitch->lists.
2559 * These LBRs will be stitched later.
2561 for (i = prev_stack->nr - 1; i > (int)distance; i--) {
2563 if (!lbr_stitch->prev_lbr_cursor[i].valid)
2566 stitch_node = get_stitch_node(thread);
2570 memcpy(&stitch_node->cursor, &lbr_stitch->prev_lbr_cursor[i],
2571 sizeof(struct callchain_cursor_node));
2574 list_add(&stitch_node->node, &lbr_stitch->lists);
2576 list_add_tail(&stitch_node->node, &lbr_stitch->lists);
2582 static bool alloc_lbr_stitch(struct thread *thread, unsigned int max_lbr)
2584 if (thread__lbr_stitch(thread))
2587 thread__set_lbr_stitch(thread, zalloc(sizeof(struct lbr_stitch)));
2588 if (!thread__lbr_stitch(thread))
2591 thread__lbr_stitch(thread)->prev_lbr_cursor =
2592 calloc(max_lbr + 1, sizeof(struct callchain_cursor_node));
2593 if (!thread__lbr_stitch(thread)->prev_lbr_cursor)
2594 goto free_lbr_stitch;
2596 INIT_LIST_HEAD(&thread__lbr_stitch(thread)->lists);
2597 INIT_LIST_HEAD(&thread__lbr_stitch(thread)->free_lists);
2602 free(thread__lbr_stitch(thread));
2603 thread__set_lbr_stitch(thread, NULL);
2605 pr_warning("Failed to allocate space for stitched LBRs. Disable LBR stitch\n");
2606 thread__set_lbr_stitch_enable(thread, false);
2611 * Resolve LBR callstack chain sample
2613 * 1 on success get LBR callchain information
2614 * 0 no available LBR callchain information, should try fp
2615 * negative error code on other errors.
2617 static int resolve_lbr_callchain_sample(struct thread *thread,
2618 struct callchain_cursor *cursor,
2619 struct perf_sample *sample,
2620 struct symbol **parent,
2621 struct addr_location *root_al,
2623 unsigned int max_lbr)
2625 bool callee = (callchain_param.order == ORDER_CALLEE);
2626 struct ip_callchain *chain = sample->callchain;
2627 int chain_nr = min(max_stack, (int)chain->nr), i;
2628 struct lbr_stitch *lbr_stitch;
2629 bool stitched_lbr = false;
2630 u64 branch_from = 0;
2633 for (i = 0; i < chain_nr; i++) {
2634 if (chain->ips[i] == PERF_CONTEXT_USER)
2638 /* LBR only affects the user callchain */
2642 if (thread__lbr_stitch_enable(thread) && !sample->no_hw_idx &&
2643 (max_lbr > 0) && alloc_lbr_stitch(thread, max_lbr)) {
2644 lbr_stitch = thread__lbr_stitch(thread);
2646 stitched_lbr = has_stitched_lbr(thread, sample,
2647 &lbr_stitch->prev_sample,
2650 if (!stitched_lbr && !list_empty(&lbr_stitch->lists)) {
2651 list_replace_init(&lbr_stitch->lists,
2652 &lbr_stitch->free_lists);
2654 memcpy(&lbr_stitch->prev_sample, sample, sizeof(*sample));
2659 err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2660 parent, root_al, branch_from,
2665 err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2666 root_al, &branch_from, true);
2671 err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2678 err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2682 err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2683 root_al, &branch_from, false);
2688 err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2689 parent, root_al, branch_from,
2697 return (err < 0) ? err : 0;
2700 static int find_prev_cpumode(struct ip_callchain *chain, struct thread *thread,
2701 struct callchain_cursor *cursor,
2702 struct symbol **parent,
2703 struct addr_location *root_al,
2704 u8 *cpumode, int ent)
2708 while (--ent >= 0) {
2709 u64 ip = chain->ips[ent];
2711 if (ip >= PERF_CONTEXT_MAX) {
2712 err = add_callchain_ip(thread, cursor, parent,
2713 root_al, cpumode, ip,
2714 false, NULL, NULL, 0);
2721 static u64 get_leaf_frame_caller(struct perf_sample *sample,
2722 struct thread *thread, int usr_idx)
2724 if (machine__normalized_is(maps__machine(thread__maps(thread)), "arm64"))
2725 return get_leaf_frame_caller_aarch64(sample, thread, usr_idx);
2730 static int thread__resolve_callchain_sample(struct thread *thread,
2731 struct callchain_cursor *cursor,
2732 struct evsel *evsel,
2733 struct perf_sample *sample,
2734 struct symbol **parent,
2735 struct addr_location *root_al,
2738 struct branch_stack *branch = sample->branch_stack;
2739 struct branch_entry *entries = perf_sample__branch_entries(sample);
2740 struct ip_callchain *chain = sample->callchain;
2742 u8 cpumode = PERF_RECORD_MISC_USER;
2743 int i, j, err, nr_entries, usr_idx;
2746 u64 leaf_frame_caller;
2749 chain_nr = chain->nr;
2751 if (evsel__has_branch_callstack(evsel)) {
2752 struct perf_env *env = evsel__env(evsel);
2754 err = resolve_lbr_callchain_sample(thread, cursor, sample, parent,
2756 !env ? 0 : env->max_branches);
2758 return (err < 0) ? err : 0;
2762 * Based on DWARF debug information, some architectures skip
2763 * a callchain entry saved by the kernel.
2765 skip_idx = arch_skip_callchain_idx(thread, chain);
2768 * Add branches to call stack for easier browsing. This gives
2769 * more context for a sample than just the callers.
2771 * This uses individual histograms of paths compared to the
2772 * aggregated histograms the normal LBR mode uses.
2774 * Limitations for now:
2775 * - No extra filters
2776 * - No annotations (should annotate somehow)
2779 if (branch && callchain_param.branch_callstack) {
2780 int nr = min(max_stack, (int)branch->nr);
2781 struct branch_entry be[nr];
2782 struct iterations iter[nr];
2784 if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
2785 pr_warning("corrupted branch chain. skipping...\n");
2789 for (i = 0; i < nr; i++) {
2790 if (callchain_param.order == ORDER_CALLEE) {
2797 * Check for overlap into the callchain.
2798 * The return address is one off compared to
2799 * the branch entry. To adjust for this
2800 * assume the calling instruction is not longer
2803 if (i == skip_idx ||
2804 chain->ips[first_call] >= PERF_CONTEXT_MAX)
2806 else if (be[i].from < chain->ips[first_call] &&
2807 be[i].from >= chain->ips[first_call] - 8)
2810 be[i] = entries[branch->nr - i - 1];
2813 memset(iter, 0, sizeof(struct iterations) * nr);
2814 nr = remove_loops(be, nr, iter);
2816 for (i = 0; i < nr; i++) {
2817 err = add_callchain_ip(thread, cursor, parent,
2824 err = add_callchain_ip(thread, cursor, parent, root_al,
2841 if (chain && callchain_param.order != ORDER_CALLEE) {
2842 err = find_prev_cpumode(chain, thread, cursor, parent, root_al,
2843 &cpumode, chain->nr - first_call);
2845 return (err < 0) ? err : 0;
2847 for (i = first_call, nr_entries = 0;
2848 i < chain_nr && nr_entries < max_stack; i++) {
2851 if (callchain_param.order == ORDER_CALLEE)
2854 j = chain->nr - i - 1;
2856 #ifdef HAVE_SKIP_CALLCHAIN_IDX
2861 if (ip < PERF_CONTEXT_MAX)
2863 else if (callchain_param.order != ORDER_CALLEE) {
2864 err = find_prev_cpumode(chain, thread, cursor, parent,
2865 root_al, &cpumode, j);
2867 return (err < 0) ? err : 0;
2872 * PERF_CONTEXT_USER allows us to locate where the user stack ends.
2873 * Depending on callchain_param.order and the position of PERF_CONTEXT_USER,
2874 * the index will be different in order to add the missing frame
2875 * at the right place.
2878 usr_idx = callchain_param.order == ORDER_CALLEE ? j-2 : j-1;
2880 if (usr_idx >= 0 && chain->ips[usr_idx] == PERF_CONTEXT_USER) {
2882 leaf_frame_caller = get_leaf_frame_caller(sample, thread, usr_idx);
2885 * check if leaf_frame_Caller != ip to not add the same
2889 if (leaf_frame_caller && leaf_frame_caller != ip) {
2891 err = add_callchain_ip(thread, cursor, parent,
2892 root_al, &cpumode, leaf_frame_caller,
2893 false, NULL, NULL, 0);
2895 return (err < 0) ? err : 0;
2899 err = add_callchain_ip(thread, cursor, parent,
2900 root_al, &cpumode, ip,
2901 false, NULL, NULL, 0);
2904 return (err < 0) ? err : 0;
2910 static int append_inlines(struct callchain_cursor *cursor, struct map_symbol *ms, u64 ip)
2912 struct symbol *sym = ms->sym;
2913 struct map *map = ms->map;
2914 struct inline_node *inline_node;
2915 struct inline_list *ilist;
2919 struct map_symbol ilist_ms;
2921 if (!symbol_conf.inline_name || !map || !sym)
2924 addr = map__dso_map_ip(map, ip);
2925 addr = map__rip_2objdump(map, addr);
2926 dso = map__dso(map);
2928 inline_node = inlines__tree_find(&dso->inlined_nodes, addr);
2930 inline_node = dso__parse_addr_inlines(dso, addr, sym);
2933 inlines__tree_insert(&dso->inlined_nodes, inline_node);
2936 ilist_ms = (struct map_symbol) {
2937 .maps = maps__get(ms->maps),
2938 .map = map__get(map),
2940 list_for_each_entry(ilist, &inline_node->val, list) {
2941 ilist_ms.sym = ilist->symbol;
2942 ret = callchain_cursor_append(cursor, ip, &ilist_ms, false,
2943 NULL, 0, 0, 0, ilist->srcline);
2948 map_symbol__exit(&ilist_ms);
2953 static int unwind_entry(struct unwind_entry *entry, void *arg)
2955 struct callchain_cursor *cursor = arg;
2956 const char *srcline = NULL;
2957 u64 addr = entry->ip;
2959 if (symbol_conf.hide_unresolved && entry->ms.sym == NULL)
2962 if (append_inlines(cursor, &entry->ms, entry->ip) == 0)
2966 * Convert entry->ip from a virtual address to an offset in
2967 * its corresponding binary.
2970 addr = map__dso_map_ip(entry->ms.map, entry->ip);
2972 srcline = callchain_srcline(&entry->ms, addr);
2973 return callchain_cursor_append(cursor, entry->ip, &entry->ms,
2974 false, NULL, 0, 0, 0, srcline);
2977 static int thread__resolve_callchain_unwind(struct thread *thread,
2978 struct callchain_cursor *cursor,
2979 struct evsel *evsel,
2980 struct perf_sample *sample,
2983 /* Can we do dwarf post unwind? */
2984 if (!((evsel->core.attr.sample_type & PERF_SAMPLE_REGS_USER) &&
2985 (evsel->core.attr.sample_type & PERF_SAMPLE_STACK_USER)))
2988 /* Bail out if nothing was captured. */
2989 if ((!sample->user_regs.regs) ||
2990 (!sample->user_stack.size))
2993 return unwind__get_entries(unwind_entry, cursor,
2994 thread, sample, max_stack, false);
2997 int thread__resolve_callchain(struct thread *thread,
2998 struct callchain_cursor *cursor,
2999 struct evsel *evsel,
3000 struct perf_sample *sample,
3001 struct symbol **parent,
3002 struct addr_location *root_al,
3010 callchain_cursor_reset(cursor);
3012 if (callchain_param.order == ORDER_CALLEE) {
3013 ret = thread__resolve_callchain_sample(thread, cursor,
3019 ret = thread__resolve_callchain_unwind(thread, cursor,
3023 ret = thread__resolve_callchain_unwind(thread, cursor,
3028 ret = thread__resolve_callchain_sample(thread, cursor,
3037 int machine__for_each_thread(struct machine *machine,
3038 int (*fn)(struct thread *thread, void *p),
3041 return threads__for_each_thread(&machine->threads, fn, priv);
3044 int machines__for_each_thread(struct machines *machines,
3045 int (*fn)(struct thread *thread, void *p),
3051 rc = machine__for_each_thread(&machines->host, fn, priv);
3055 for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
3056 struct machine *machine = rb_entry(nd, struct machine, rb_node);
3058 rc = machine__for_each_thread(machine, fn, priv);
3066 static int thread_list_cb(struct thread *thread, void *data)
3068 struct list_head *list = data;
3069 struct thread_list *entry = malloc(sizeof(*entry));
3074 entry->thread = thread__get(thread);
3075 list_add_tail(&entry->list, list);
3079 int machine__thread_list(struct machine *machine, struct list_head *list)
3081 return machine__for_each_thread(machine, thread_list_cb, list);
3084 void thread_list__delete(struct list_head *list)
3086 struct thread_list *pos, *next;
3088 list_for_each_entry_safe(pos, next, list, list) {
3089 thread__zput(pos->thread);
3090 list_del(&pos->list);
3095 pid_t machine__get_current_tid(struct machine *machine, int cpu)
3097 if (cpu < 0 || (size_t)cpu >= machine->current_tid_sz)
3100 return machine->current_tid[cpu];
3103 int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
3106 struct thread *thread;
3107 const pid_t init_val = -1;
3112 if (realloc_array_as_needed(machine->current_tid,
3113 machine->current_tid_sz,
3118 machine->current_tid[cpu] = tid;
3120 thread = machine__findnew_thread(machine, pid, tid);
3124 thread__set_cpu(thread, cpu);
3125 thread__put(thread);
3131 * Compares the raw arch string. N.B. see instead perf_env__arch() or
3132 * machine__normalized_is() if a normalized arch is needed.
3134 bool machine__is(struct machine *machine, const char *arch)
3136 return machine && !strcmp(perf_env__raw_arch(machine->env), arch);
3139 bool machine__normalized_is(struct machine *machine, const char *arch)
3141 return machine && !strcmp(perf_env__arch(machine->env), arch);
3144 int machine__nr_cpus_avail(struct machine *machine)
3146 return machine ? perf_env__nr_cpus_avail(machine->env) : 0;
3149 int machine__get_kernel_start(struct machine *machine)
3151 struct map *map = machine__kernel_map(machine);
3155 * The only addresses above 2^63 are kernel addresses of a 64-bit
3156 * kernel. Note that addresses are unsigned so that on a 32-bit system
3157 * all addresses including kernel addresses are less than 2^32. In
3158 * that case (32-bit system), if the kernel mapping is unknown, all
3159 * addresses will be assumed to be in user space - see
3160 * machine__kernel_ip().
3162 machine->kernel_start = 1ULL << 63;
3164 err = map__load(map);
3166 * On x86_64, PTI entry trampolines are less than the
3167 * start of kernel text, but still above 2^63. So leave
3168 * kernel_start = 1ULL << 63 for x86_64.
3170 if (!err && !machine__is(machine, "x86_64"))
3171 machine->kernel_start = map__start(map);
3176 u8 machine__addr_cpumode(struct machine *machine, u8 cpumode, u64 addr)
3178 u8 addr_cpumode = cpumode;
3181 if (!machine->single_address_space)
3184 kernel_ip = machine__kernel_ip(machine, addr);
3186 case PERF_RECORD_MISC_KERNEL:
3187 case PERF_RECORD_MISC_USER:
3188 addr_cpumode = kernel_ip ? PERF_RECORD_MISC_KERNEL :
3189 PERF_RECORD_MISC_USER;
3191 case PERF_RECORD_MISC_GUEST_KERNEL:
3192 case PERF_RECORD_MISC_GUEST_USER:
3193 addr_cpumode = kernel_ip ? PERF_RECORD_MISC_GUEST_KERNEL :
3194 PERF_RECORD_MISC_GUEST_USER;
3200 return addr_cpumode;
3203 struct dso *machine__findnew_dso_id(struct machine *machine, const char *filename, struct dso_id *id)
3205 return dsos__findnew_id(&machine->dsos, filename, id);
3208 struct dso *machine__findnew_dso(struct machine *machine, const char *filename)
3210 return machine__findnew_dso_id(machine, filename, NULL);
3213 char *machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
3215 struct machine *machine = vmachine;
3217 struct symbol *sym = machine__find_kernel_symbol(machine, *addrp, &map);
3222 *modp = __map__is_kmodule(map) ? (char *)map__dso(map)->short_name : NULL;
3223 *addrp = map__unmap_ip(map, sym->start);
3227 int machine__for_each_dso(struct machine *machine, machine__dso_t fn, void *priv)
3232 list_for_each_entry(pos, &machine->dsos.head, node) {
3233 if (fn(pos, machine, priv))
3239 int machine__for_each_kernel_map(struct machine *machine, machine__map_t fn, void *priv)
3241 struct maps *maps = machine__kernel_maps(machine);
3243 return maps__for_each_map(maps, fn, priv);
3246 bool machine__is_lock_function(struct machine *machine, u64 addr)
3248 if (!machine->sched.text_start) {
3250 struct symbol *sym = machine__find_kernel_symbol_by_name(machine, "__sched_text_start", &kmap);
3253 /* to avoid retry */
3254 machine->sched.text_start = 1;
3258 machine->sched.text_start = map__unmap_ip(kmap, sym->start);
3260 /* should not fail from here */
3261 sym = machine__find_kernel_symbol_by_name(machine, "__sched_text_end", &kmap);
3262 machine->sched.text_end = map__unmap_ip(kmap, sym->start);
3264 sym = machine__find_kernel_symbol_by_name(machine, "__lock_text_start", &kmap);
3265 machine->lock.text_start = map__unmap_ip(kmap, sym->start);
3267 sym = machine__find_kernel_symbol_by_name(machine, "__lock_text_end", &kmap);
3268 machine->lock.text_end = map__unmap_ip(kmap, sym->start);
3271 /* failed to get kernel symbols */
3272 if (machine->sched.text_start == 1)
3275 /* mutex and rwsem functions are in sched text section */
3276 if (machine->sched.text_start <= addr && addr < machine->sched.text_end)
3279 /* spinlock functions are in lock text section */
3280 if (machine->lock.text_start <= addr && addr < machine->lock.text_end)