1 // SPDX-License-Identifier: GPL-2.0
18 #include <sys/types.h>
22 #include "linux/hash.h"
25 #include "sane_ctype.h"
26 #include <symbol/kallsyms.h>
27 #include <linux/mman.h>
29 static void __machine__remove_thread(struct machine
*machine
, struct thread
*th
, bool lock
);
31 static void dsos__init(struct dsos
*dsos
)
33 INIT_LIST_HEAD(&dsos
->head
);
35 init_rwsem(&dsos
->lock
);
38 static void machine__threads_init(struct machine
*machine
)
42 for (i
= 0; i
< THREADS__TABLE_SIZE
; i
++) {
43 struct threads
*threads
= &machine
->threads
[i
];
44 threads
->entries
= RB_ROOT
;
45 init_rwsem(&threads
->lock
);
47 INIT_LIST_HEAD(&threads
->dead
);
48 threads
->last_match
= NULL
;
52 static int machine__set_mmap_name(struct machine
*machine
)
54 if (machine__is_host(machine
))
55 machine
->mmap_name
= strdup("[kernel.kallsyms]");
56 else if (machine__is_default_guest(machine
))
57 machine
->mmap_name
= strdup("[guest.kernel.kallsyms]");
58 else if (asprintf(&machine
->mmap_name
, "[guest.kernel.kallsyms.%d]",
60 machine
->mmap_name
= NULL
;
62 return machine
->mmap_name
? 0 : -ENOMEM
;
65 int machine__init(struct machine
*machine
, const char *root_dir
, pid_t pid
)
69 memset(machine
, 0, sizeof(*machine
));
70 map_groups__init(&machine
->kmaps
, machine
);
71 RB_CLEAR_NODE(&machine
->rb_node
);
72 dsos__init(&machine
->dsos
);
74 machine__threads_init(machine
);
76 machine
->vdso_info
= NULL
;
81 machine
->id_hdr_size
= 0;
82 machine
->kptr_restrict_warned
= false;
83 machine
->comm_exec
= false;
84 machine
->kernel_start
= 0;
85 machine
->vmlinux_map
= NULL
;
87 machine
->root_dir
= strdup(root_dir
);
88 if (machine
->root_dir
== NULL
)
91 if (machine__set_mmap_name(machine
))
94 if (pid
!= HOST_KERNEL_ID
) {
95 struct thread
*thread
= machine__findnew_thread(machine
, -1,
102 snprintf(comm
, sizeof(comm
), "[guest/%d]", pid
);
103 thread__set_comm(thread
, comm
, 0);
107 machine
->current_tid
= NULL
;
112 zfree(&machine
->root_dir
);
113 zfree(&machine
->mmap_name
);
118 struct machine
*machine__new_host(void)
120 struct machine
*machine
= malloc(sizeof(*machine
));
122 if (machine
!= NULL
) {
123 machine__init(machine
, "", HOST_KERNEL_ID
);
125 if (machine__create_kernel_maps(machine
) < 0)
135 struct machine
*machine__new_kallsyms(void)
137 struct machine
*machine
= machine__new_host();
140 * 1) We should switch to machine__load_kallsyms(), i.e. not explicitely
141 * ask for not using the kcore parsing code, once this one is fixed
142 * to create a map per module.
144 if (machine
&& machine__load_kallsyms(machine
, "/proc/kallsyms") <= 0) {
145 machine__delete(machine
);
152 static void dsos__purge(struct dsos
*dsos
)
156 down_write(&dsos
->lock
);
158 list_for_each_entry_safe(pos
, n
, &dsos
->head
, node
) {
159 RB_CLEAR_NODE(&pos
->rb_node
);
161 list_del_init(&pos
->node
);
165 up_write(&dsos
->lock
);
168 static void dsos__exit(struct dsos
*dsos
)
171 exit_rwsem(&dsos
->lock
);
174 void machine__delete_threads(struct machine
*machine
)
179 for (i
= 0; i
< THREADS__TABLE_SIZE
; i
++) {
180 struct threads
*threads
= &machine
->threads
[i
];
181 down_write(&threads
->lock
);
182 nd
= rb_first(&threads
->entries
);
184 struct thread
*t
= rb_entry(nd
, struct thread
, rb_node
);
187 __machine__remove_thread(machine
, t
, false);
189 up_write(&threads
->lock
);
193 void machine__exit(struct machine
*machine
)
200 machine__destroy_kernel_maps(machine
);
201 map_groups__exit(&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
);
208 for (i
= 0; i
< THREADS__TABLE_SIZE
; i
++) {
209 struct threads
*threads
= &machine
->threads
[i
];
210 exit_rwsem(&threads
->lock
);
214 void machine__delete(struct machine
*machine
)
217 machine__exit(machine
);
222 void machines__init(struct machines
*machines
)
224 machine__init(&machines
->host
, "", HOST_KERNEL_ID
);
225 machines
->guests
= RB_ROOT
;
228 void machines__exit(struct machines
*machines
)
230 machine__exit(&machines
->host
);
234 struct machine
*machines__add(struct machines
*machines
, pid_t pid
,
235 const char *root_dir
)
237 struct rb_node
**p
= &machines
->guests
.rb_node
;
238 struct rb_node
*parent
= NULL
;
239 struct machine
*pos
, *machine
= malloc(sizeof(*machine
));
244 if (machine__init(machine
, root_dir
, pid
) != 0) {
251 pos
= rb_entry(parent
, struct machine
, rb_node
);
258 rb_link_node(&machine
->rb_node
, parent
, p
);
259 rb_insert_color(&machine
->rb_node
, &machines
->guests
);
264 void machines__set_comm_exec(struct machines
*machines
, bool comm_exec
)
268 machines
->host
.comm_exec
= comm_exec
;
270 for (nd
= rb_first(&machines
->guests
); nd
; nd
= rb_next(nd
)) {
271 struct machine
*machine
= rb_entry(nd
, struct machine
, rb_node
);
273 machine
->comm_exec
= comm_exec
;
277 struct machine
*machines__find(struct machines
*machines
, pid_t pid
)
279 struct rb_node
**p
= &machines
->guests
.rb_node
;
280 struct rb_node
*parent
= NULL
;
281 struct machine
*machine
;
282 struct machine
*default_machine
= NULL
;
284 if (pid
== HOST_KERNEL_ID
)
285 return &machines
->host
;
289 machine
= rb_entry(parent
, struct machine
, rb_node
);
290 if (pid
< machine
->pid
)
292 else if (pid
> machine
->pid
)
297 default_machine
= machine
;
300 return default_machine
;
303 struct machine
*machines__findnew(struct machines
*machines
, pid_t pid
)
306 const char *root_dir
= "";
307 struct machine
*machine
= machines__find(machines
, pid
);
309 if (machine
&& (machine
->pid
== pid
))
312 if ((pid
!= HOST_KERNEL_ID
) &&
313 (pid
!= DEFAULT_GUEST_KERNEL_ID
) &&
314 (symbol_conf
.guestmount
)) {
315 sprintf(path
, "%s/%d", symbol_conf
.guestmount
, pid
);
316 if (access(path
, R_OK
)) {
317 static struct strlist
*seen
;
320 seen
= strlist__new(NULL
, NULL
);
322 if (!strlist__has_entry(seen
, path
)) {
323 pr_err("Can't access file %s\n", path
);
324 strlist__add(seen
, path
);
332 machine
= machines__add(machines
, pid
, root_dir
);
337 void machines__process_guests(struct machines
*machines
,
338 machine__process_t process
, void *data
)
342 for (nd
= rb_first(&machines
->guests
); nd
; nd
= rb_next(nd
)) {
343 struct machine
*pos
= rb_entry(nd
, struct machine
, rb_node
);
348 void machines__set_id_hdr_size(struct machines
*machines
, u16 id_hdr_size
)
350 struct rb_node
*node
;
351 struct machine
*machine
;
353 machines
->host
.id_hdr_size
= id_hdr_size
;
355 for (node
= rb_first(&machines
->guests
); node
; node
= rb_next(node
)) {
356 machine
= rb_entry(node
, struct machine
, rb_node
);
357 machine
->id_hdr_size
= id_hdr_size
;
363 static void machine__update_thread_pid(struct machine
*machine
,
364 struct thread
*th
, pid_t pid
)
366 struct thread
*leader
;
368 if (pid
== th
->pid_
|| pid
== -1 || th
->pid_
!= -1)
373 if (th
->pid_
== th
->tid
)
376 leader
= __machine__findnew_thread(machine
, th
->pid_
, th
->pid_
);
381 leader
->mg
= map_groups__new(machine
);
386 if (th
->mg
== leader
->mg
)
391 * Maps are created from MMAP events which provide the pid and
392 * tid. Consequently there never should be any maps on a thread
393 * with an unknown pid. Just print an error if there are.
395 if (!map_groups__empty(th
->mg
))
396 pr_err("Discarding thread maps for %d:%d\n",
398 map_groups__put(th
->mg
);
401 th
->mg
= map_groups__get(leader
->mg
);
406 pr_err("Failed to join map groups for %d:%d\n", th
->pid_
, th
->tid
);
411 * Front-end cache - TID lookups come in blocks,
412 * so most of the time we dont have to look up
415 static struct thread
*
416 __threads__get_last_match(struct threads
*threads
, struct machine
*machine
,
421 th
= threads
->last_match
;
423 if (th
->tid
== tid
) {
424 machine__update_thread_pid(machine
, th
, pid
);
425 return thread__get(th
);
428 threads
->last_match
= NULL
;
434 static struct thread
*
435 threads__get_last_match(struct threads
*threads
, struct machine
*machine
,
438 struct thread
*th
= NULL
;
440 if (perf_singlethreaded
)
441 th
= __threads__get_last_match(threads
, machine
, pid
, tid
);
447 __threads__set_last_match(struct threads
*threads
, struct thread
*th
)
449 threads
->last_match
= th
;
453 threads__set_last_match(struct threads
*threads
, struct thread
*th
)
455 if (perf_singlethreaded
)
456 __threads__set_last_match(threads
, th
);
460 * Caller must eventually drop thread->refcnt returned with a successful
461 * lookup/new thread inserted.
463 static struct thread
*____machine__findnew_thread(struct machine
*machine
,
464 struct threads
*threads
,
465 pid_t pid
, pid_t tid
,
468 struct rb_node
**p
= &threads
->entries
.rb_node
;
469 struct rb_node
*parent
= NULL
;
472 th
= threads__get_last_match(threads
, machine
, pid
, tid
);
478 th
= rb_entry(parent
, struct thread
, rb_node
);
480 if (th
->tid
== tid
) {
481 threads__set_last_match(threads
, th
);
482 machine__update_thread_pid(machine
, th
, pid
);
483 return thread__get(th
);
495 th
= thread__new(pid
, tid
);
497 rb_link_node(&th
->rb_node
, parent
, p
);
498 rb_insert_color(&th
->rb_node
, &threads
->entries
);
501 * We have to initialize map_groups separately
502 * after rb tree is updated.
504 * The reason is that we call machine__findnew_thread
505 * within thread__init_map_groups to find the thread
506 * leader and that would screwed the rb tree.
508 if (thread__init_map_groups(th
, machine
)) {
509 rb_erase_init(&th
->rb_node
, &threads
->entries
);
510 RB_CLEAR_NODE(&th
->rb_node
);
515 * It is now in the rbtree, get a ref
518 threads__set_last_match(threads
, th
);
525 struct thread
*__machine__findnew_thread(struct machine
*machine
, pid_t pid
, pid_t tid
)
527 return ____machine__findnew_thread(machine
, machine__threads(machine
, tid
), pid
, tid
, true);
530 struct thread
*machine__findnew_thread(struct machine
*machine
, pid_t pid
,
533 struct threads
*threads
= machine__threads(machine
, tid
);
536 down_write(&threads
->lock
);
537 th
= __machine__findnew_thread(machine
, pid
, tid
);
538 up_write(&threads
->lock
);
542 struct thread
*machine__find_thread(struct machine
*machine
, pid_t pid
,
545 struct threads
*threads
= machine__threads(machine
, tid
);
548 down_read(&threads
->lock
);
549 th
= ____machine__findnew_thread(machine
, threads
, pid
, tid
, false);
550 up_read(&threads
->lock
);
554 struct comm
*machine__thread_exec_comm(struct machine
*machine
,
555 struct thread
*thread
)
557 if (machine
->comm_exec
)
558 return thread__exec_comm(thread
);
560 return thread__comm(thread
);
563 int machine__process_comm_event(struct machine
*machine
, union perf_event
*event
,
564 struct perf_sample
*sample
)
566 struct thread
*thread
= machine__findnew_thread(machine
,
569 bool exec
= event
->header
.misc
& PERF_RECORD_MISC_COMM_EXEC
;
573 machine
->comm_exec
= true;
576 perf_event__fprintf_comm(event
, stdout
);
578 if (thread
== NULL
||
579 __thread__set_comm(thread
, event
->comm
.comm
, sample
->time
, exec
)) {
580 dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
589 int machine__process_namespaces_event(struct machine
*machine __maybe_unused
,
590 union perf_event
*event
,
591 struct perf_sample
*sample __maybe_unused
)
593 struct thread
*thread
= machine__findnew_thread(machine
,
594 event
->namespaces
.pid
,
595 event
->namespaces
.tid
);
598 WARN_ONCE(event
->namespaces
.nr_namespaces
> NR_NAMESPACES
,
599 "\nWARNING: kernel seems to support more namespaces than perf"
600 " tool.\nTry updating the perf tool..\n\n");
602 WARN_ONCE(event
->namespaces
.nr_namespaces
< NR_NAMESPACES
,
603 "\nWARNING: perf tool seems to support more namespaces than"
604 " the kernel.\nTry updating the kernel..\n\n");
607 perf_event__fprintf_namespaces(event
, stdout
);
609 if (thread
== NULL
||
610 thread__set_namespaces(thread
, sample
->time
, &event
->namespaces
)) {
611 dump_printf("problem processing PERF_RECORD_NAMESPACES, skipping event.\n");
620 int machine__process_lost_event(struct machine
*machine __maybe_unused
,
621 union perf_event
*event
, struct perf_sample
*sample __maybe_unused
)
623 dump_printf(": id:%" PRIu64
": lost:%" PRIu64
"\n",
624 event
->lost
.id
, event
->lost
.lost
);
628 int machine__process_lost_samples_event(struct machine
*machine __maybe_unused
,
629 union perf_event
*event
, struct perf_sample
*sample
)
631 dump_printf(": id:%" PRIu64
": lost samples :%" PRIu64
"\n",
632 sample
->id
, event
->lost_samples
.lost
);
636 static struct dso
*machine__findnew_module_dso(struct machine
*machine
,
638 const char *filename
)
642 down_write(&machine
->dsos
.lock
);
644 dso
= __dsos__find(&machine
->dsos
, m
->name
, true);
646 dso
= __dsos__addnew(&machine
->dsos
, m
->name
);
650 dso__set_module_info(dso
, m
, machine
);
651 dso__set_long_name(dso
, strdup(filename
), true);
656 up_write(&machine
->dsos
.lock
);
660 int machine__process_aux_event(struct machine
*machine __maybe_unused
,
661 union perf_event
*event
)
664 perf_event__fprintf_aux(event
, stdout
);
668 int machine__process_itrace_start_event(struct machine
*machine __maybe_unused
,
669 union perf_event
*event
)
672 perf_event__fprintf_itrace_start(event
, stdout
);
676 int machine__process_switch_event(struct machine
*machine __maybe_unused
,
677 union perf_event
*event
)
680 perf_event__fprintf_switch(event
, stdout
);
684 static void dso__adjust_kmod_long_name(struct dso
*dso
, const char *filename
)
686 const char *dup_filename
;
688 if (!filename
|| !dso
|| !dso
->long_name
)
690 if (dso
->long_name
[0] != '[')
692 if (!strchr(filename
, '/'))
695 dup_filename
= strdup(filename
);
699 dso__set_long_name(dso
, dup_filename
, true);
702 struct map
*machine__findnew_module_map(struct machine
*machine
, u64 start
,
703 const char *filename
)
705 struct map
*map
= NULL
;
706 struct dso
*dso
= NULL
;
709 if (kmod_path__parse_name(&m
, filename
))
712 map
= map_groups__find_by_name(&machine
->kmaps
, m
.name
);
715 * If the map's dso is an offline module, give dso__load()
716 * a chance to find the file path of that module by fixing
719 dso__adjust_kmod_long_name(map
->dso
, filename
);
723 dso
= machine__findnew_module_dso(machine
, &m
, filename
);
727 map
= map__new2(start
, dso
);
731 map_groups__insert(&machine
->kmaps
, map
);
733 /* Put the map here because map_groups__insert alread got it */
736 /* put the dso here, corresponding to machine__findnew_module_dso */
742 size_t machines__fprintf_dsos(struct machines
*machines
, FILE *fp
)
745 size_t ret
= __dsos__fprintf(&machines
->host
.dsos
.head
, fp
);
747 for (nd
= rb_first(&machines
->guests
); nd
; nd
= rb_next(nd
)) {
748 struct machine
*pos
= rb_entry(nd
, struct machine
, rb_node
);
749 ret
+= __dsos__fprintf(&pos
->dsos
.head
, fp
);
755 size_t machine__fprintf_dsos_buildid(struct machine
*m
, FILE *fp
,
756 bool (skip
)(struct dso
*dso
, int parm
), int parm
)
758 return __dsos__fprintf_buildid(&m
->dsos
.head
, fp
, skip
, parm
);
761 size_t machines__fprintf_dsos_buildid(struct machines
*machines
, FILE *fp
,
762 bool (skip
)(struct dso
*dso
, int parm
), int parm
)
765 size_t ret
= machine__fprintf_dsos_buildid(&machines
->host
, fp
, skip
, parm
);
767 for (nd
= rb_first(&machines
->guests
); nd
; nd
= rb_next(nd
)) {
768 struct machine
*pos
= rb_entry(nd
, struct machine
, rb_node
);
769 ret
+= machine__fprintf_dsos_buildid(pos
, fp
, skip
, parm
);
774 size_t machine__fprintf_vmlinux_path(struct machine
*machine
, FILE *fp
)
778 struct dso
*kdso
= machine__kernel_map(machine
)->dso
;
780 if (kdso
->has_build_id
) {
781 char filename
[PATH_MAX
];
782 if (dso__build_id_filename(kdso
, filename
, sizeof(filename
),
784 printed
+= fprintf(fp
, "[0] %s\n", filename
);
787 for (i
= 0; i
< vmlinux_path__nr_entries
; ++i
)
788 printed
+= fprintf(fp
, "[%d] %s\n",
789 i
+ kdso
->has_build_id
, vmlinux_path
[i
]);
794 size_t machine__fprintf(struct machine
*machine
, FILE *fp
)
800 for (i
= 0; i
< THREADS__TABLE_SIZE
; i
++) {
801 struct threads
*threads
= &machine
->threads
[i
];
803 down_read(&threads
->lock
);
805 ret
= fprintf(fp
, "Threads: %u\n", threads
->nr
);
807 for (nd
= rb_first(&threads
->entries
); nd
; nd
= rb_next(nd
)) {
808 struct thread
*pos
= rb_entry(nd
, struct thread
, rb_node
);
810 ret
+= thread__fprintf(pos
, fp
);
813 up_read(&threads
->lock
);
818 static struct dso
*machine__get_kernel(struct machine
*machine
)
820 const char *vmlinux_name
= machine
->mmap_name
;
823 if (machine__is_host(machine
)) {
824 if (symbol_conf
.vmlinux_name
)
825 vmlinux_name
= symbol_conf
.vmlinux_name
;
827 kernel
= machine__findnew_kernel(machine
, vmlinux_name
,
828 "[kernel]", DSO_TYPE_KERNEL
);
830 if (symbol_conf
.default_guest_vmlinux_name
)
831 vmlinux_name
= symbol_conf
.default_guest_vmlinux_name
;
833 kernel
= machine__findnew_kernel(machine
, vmlinux_name
,
835 DSO_TYPE_GUEST_KERNEL
);
838 if (kernel
!= NULL
&& (!kernel
->has_build_id
))
839 dso__read_running_kernel_build_id(kernel
, machine
);
844 struct process_args
{
848 void machine__get_kallsyms_filename(struct machine
*machine
, char *buf
,
851 if (machine__is_default_guest(machine
))
852 scnprintf(buf
, bufsz
, "%s", symbol_conf
.default_guest_kallsyms
);
854 scnprintf(buf
, bufsz
, "%s/proc/kallsyms", machine
->root_dir
);
857 const char *ref_reloc_sym_names
[] = {"_text", "_stext", NULL
};
859 /* Figure out the start address of kernel map from /proc/kallsyms.
860 * Returns the name of the start symbol in *symbol_name. Pass in NULL as
861 * symbol_name if it's not that important.
863 static int machine__get_running_kernel_start(struct machine
*machine
,
864 const char **symbol_name
, u64
*start
)
866 char filename
[PATH_MAX
];
871 machine__get_kallsyms_filename(machine
, filename
, PATH_MAX
);
873 if (symbol__restricted_filename(filename
, "/proc/kallsyms"))
876 for (i
= 0; (name
= ref_reloc_sym_names
[i
]) != NULL
; i
++) {
877 err
= kallsyms__get_function_start(filename
, name
, &addr
);
892 int machine__create_extra_kernel_map(struct machine
*machine
,
894 struct extra_kernel_map
*xm
)
899 map
= map__new2(xm
->start
, kernel
);
904 map
->pgoff
= xm
->pgoff
;
906 kmap
= map__kmap(map
);
908 kmap
->kmaps
= &machine
->kmaps
;
909 strlcpy(kmap
->name
, xm
->name
, KMAP_NAME_LEN
);
911 map_groups__insert(&machine
->kmaps
, map
);
913 pr_debug2("Added extra kernel map %s %" PRIx64
"-%" PRIx64
"\n",
914 kmap
->name
, map
->start
, map
->end
);
921 static u64
find_entry_trampoline(struct dso
*dso
)
923 /* Duplicates are removed so lookup all aliases */
924 const char *syms
[] = {
926 "__entry_trampoline_start",
927 "entry_SYSCALL_64_trampoline",
929 struct symbol
*sym
= dso__first_symbol(dso
);
932 for (; sym
; sym
= dso__next_symbol(sym
)) {
933 if (sym
->binding
!= STB_GLOBAL
)
935 for (i
= 0; i
< ARRAY_SIZE(syms
); i
++) {
936 if (!strcmp(sym
->name
, syms
[i
]))
945 * These values can be used for kernels that do not have symbols for the entry
946 * trampolines in kallsyms.
948 #define X86_64_CPU_ENTRY_AREA_PER_CPU 0xfffffe0000000000ULL
949 #define X86_64_CPU_ENTRY_AREA_SIZE 0x2c000
950 #define X86_64_ENTRY_TRAMPOLINE 0x6000
952 /* Map x86_64 PTI entry trampolines */
953 int machine__map_x86_64_entry_trampolines(struct machine
*machine
,
956 struct map_groups
*kmaps
= &machine
->kmaps
;
957 struct maps
*maps
= &kmaps
->maps
;
958 int nr_cpus_avail
, cpu
;
964 * In the vmlinux case, pgoff is a virtual address which must now be
965 * mapped to a vmlinux offset.
967 for (map
= maps__first(maps
); map
; map
= map__next(map
)) {
968 struct kmap
*kmap
= __map__kmap(map
);
969 struct map
*dest_map
;
971 if (!kmap
|| !is_entry_trampoline(kmap
->name
))
974 dest_map
= map_groups__find(kmaps
, map
->pgoff
);
976 map
->pgoff
= dest_map
->map_ip(dest_map
, map
->pgoff
);
979 if (found
|| machine
->trampolines_mapped
)
982 pgoff
= find_entry_trampoline(kernel
);
986 nr_cpus_avail
= machine__nr_cpus_avail(machine
);
988 /* Add a 1 page map for each CPU's entry trampoline */
989 for (cpu
= 0; cpu
< nr_cpus_avail
; cpu
++) {
990 u64 va
= X86_64_CPU_ENTRY_AREA_PER_CPU
+
991 cpu
* X86_64_CPU_ENTRY_AREA_SIZE
+
992 X86_64_ENTRY_TRAMPOLINE
;
993 struct extra_kernel_map xm
= {
995 .end
= va
+ page_size
,
999 strlcpy(xm
.name
, ENTRY_TRAMPOLINE_NAME
, KMAP_NAME_LEN
);
1001 if (machine__create_extra_kernel_map(machine
, kernel
, &xm
) < 0)
1005 machine
->trampolines_mapped
= nr_cpus_avail
;
1010 int __weak
machine__create_extra_kernel_maps(struct machine
*machine __maybe_unused
,
1011 struct dso
*kernel __maybe_unused
)
1017 __machine__create_kernel_maps(struct machine
*machine
, struct dso
*kernel
)
1022 /* In case of renewal the kernel map, destroy previous one */
1023 machine__destroy_kernel_maps(machine
);
1025 machine
->vmlinux_map
= map__new2(0, kernel
);
1026 if (machine
->vmlinux_map
== NULL
)
1029 machine
->vmlinux_map
->map_ip
= machine
->vmlinux_map
->unmap_ip
= identity__map_ip
;
1030 map
= machine__kernel_map(machine
);
1031 kmap
= map__kmap(map
);
1035 kmap
->kmaps
= &machine
->kmaps
;
1036 map_groups__insert(&machine
->kmaps
, map
);
1041 void machine__destroy_kernel_maps(struct machine
*machine
)
1044 struct map
*map
= machine__kernel_map(machine
);
1049 kmap
= map__kmap(map
);
1050 map_groups__remove(&machine
->kmaps
, map
);
1051 if (kmap
&& kmap
->ref_reloc_sym
) {
1052 zfree((char **)&kmap
->ref_reloc_sym
->name
);
1053 zfree(&kmap
->ref_reloc_sym
);
1056 map__zput(machine
->vmlinux_map
);
1059 int machines__create_guest_kernel_maps(struct machines
*machines
)
1062 struct dirent
**namelist
= NULL
;
1064 char path
[PATH_MAX
];
1068 if (symbol_conf
.default_guest_vmlinux_name
||
1069 symbol_conf
.default_guest_modules
||
1070 symbol_conf
.default_guest_kallsyms
) {
1071 machines__create_kernel_maps(machines
, DEFAULT_GUEST_KERNEL_ID
);
1074 if (symbol_conf
.guestmount
) {
1075 items
= scandir(symbol_conf
.guestmount
, &namelist
, NULL
, NULL
);
1078 for (i
= 0; i
< items
; i
++) {
1079 if (!isdigit(namelist
[i
]->d_name
[0])) {
1080 /* Filter out . and .. */
1083 pid
= (pid_t
)strtol(namelist
[i
]->d_name
, &endp
, 10);
1084 if ((*endp
!= '\0') ||
1085 (endp
== namelist
[i
]->d_name
) ||
1086 (errno
== ERANGE
)) {
1087 pr_debug("invalid directory (%s). Skipping.\n",
1088 namelist
[i
]->d_name
);
1091 sprintf(path
, "%s/%s/proc/kallsyms",
1092 symbol_conf
.guestmount
,
1093 namelist
[i
]->d_name
);
1094 ret
= access(path
, R_OK
);
1096 pr_debug("Can't access file %s\n", path
);
1099 machines__create_kernel_maps(machines
, pid
);
1108 void machines__destroy_kernel_maps(struct machines
*machines
)
1110 struct rb_node
*next
= rb_first(&machines
->guests
);
1112 machine__destroy_kernel_maps(&machines
->host
);
1115 struct machine
*pos
= rb_entry(next
, struct machine
, rb_node
);
1117 next
= rb_next(&pos
->rb_node
);
1118 rb_erase(&pos
->rb_node
, &machines
->guests
);
1119 machine__delete(pos
);
1123 int machines__create_kernel_maps(struct machines
*machines
, pid_t pid
)
1125 struct machine
*machine
= machines__findnew(machines
, pid
);
1127 if (machine
== NULL
)
1130 return machine__create_kernel_maps(machine
);
1133 int machine__load_kallsyms(struct machine
*machine
, const char *filename
)
1135 struct map
*map
= machine__kernel_map(machine
);
1136 int ret
= __dso__load_kallsyms(map
->dso
, filename
, map
, true);
1139 dso__set_loaded(map
->dso
);
1141 * Since /proc/kallsyms will have multiple sessions for the
1142 * kernel, with modules between them, fixup the end of all
1145 map_groups__fixup_end(&machine
->kmaps
);
1151 int machine__load_vmlinux_path(struct machine
*machine
)
1153 struct map
*map
= machine__kernel_map(machine
);
1154 int ret
= dso__load_vmlinux_path(map
->dso
, map
);
1157 dso__set_loaded(map
->dso
);
1162 static char *get_kernel_version(const char *root_dir
)
1164 char version
[PATH_MAX
];
1167 const char *prefix
= "Linux version ";
1169 sprintf(version
, "%s/proc/version", root_dir
);
1170 file
= fopen(version
, "r");
1175 tmp
= fgets(version
, sizeof(version
), file
);
1178 name
= strstr(version
, prefix
);
1181 name
+= strlen(prefix
);
1182 tmp
= strchr(name
, ' ');
1186 return strdup(name
);
1189 static bool is_kmod_dso(struct dso
*dso
)
1191 return dso
->symtab_type
== DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE
||
1192 dso
->symtab_type
== DSO_BINARY_TYPE__GUEST_KMODULE
;
1195 static int map_groups__set_module_path(struct map_groups
*mg
, const char *path
,
1196 struct kmod_path
*m
)
1199 struct map
*map
= map_groups__find_by_name(mg
, m
->name
);
1204 long_name
= strdup(path
);
1205 if (long_name
== NULL
)
1208 dso__set_long_name(map
->dso
, long_name
, true);
1209 dso__kernel_module_get_build_id(map
->dso
, "");
1212 * Full name could reveal us kmod compression, so
1213 * we need to update the symtab_type if needed.
1215 if (m
->comp
&& is_kmod_dso(map
->dso
)) {
1216 map
->dso
->symtab_type
++;
1217 map
->dso
->comp
= m
->comp
;
1223 static int map_groups__set_modules_path_dir(struct map_groups
*mg
,
1224 const char *dir_name
, int depth
)
1226 struct dirent
*dent
;
1227 DIR *dir
= opendir(dir_name
);
1231 pr_debug("%s: cannot open %s dir\n", __func__
, dir_name
);
1235 while ((dent
= readdir(dir
)) != NULL
) {
1236 char path
[PATH_MAX
];
1239 /*sshfs might return bad dent->d_type, so we have to stat*/
1240 snprintf(path
, sizeof(path
), "%s/%s", dir_name
, dent
->d_name
);
1241 if (stat(path
, &st
))
1244 if (S_ISDIR(st
.st_mode
)) {
1245 if (!strcmp(dent
->d_name
, ".") ||
1246 !strcmp(dent
->d_name
, ".."))
1249 /* Do not follow top-level source and build symlinks */
1251 if (!strcmp(dent
->d_name
, "source") ||
1252 !strcmp(dent
->d_name
, "build"))
1256 ret
= map_groups__set_modules_path_dir(mg
, path
,
1263 ret
= kmod_path__parse_name(&m
, dent
->d_name
);
1268 ret
= map_groups__set_module_path(mg
, path
, &m
);
1282 static int machine__set_modules_path(struct machine
*machine
)
1285 char modules_path
[PATH_MAX
];
1287 version
= get_kernel_version(machine
->root_dir
);
1291 snprintf(modules_path
, sizeof(modules_path
), "%s/lib/modules/%s",
1292 machine
->root_dir
, version
);
1295 return map_groups__set_modules_path_dir(&machine
->kmaps
, modules_path
, 0);
1297 int __weak
arch__fix_module_text_start(u64
*start __maybe_unused
,
1298 const char *name __maybe_unused
)
1303 static int machine__create_module(void *arg
, const char *name
, u64 start
,
1306 struct machine
*machine
= arg
;
1309 if (arch__fix_module_text_start(&start
, name
) < 0)
1312 map
= machine__findnew_module_map(machine
, start
, name
);
1315 map
->end
= start
+ size
;
1317 dso__kernel_module_get_build_id(map
->dso
, machine
->root_dir
);
1322 static int machine__create_modules(struct machine
*machine
)
1324 const char *modules
;
1325 char path
[PATH_MAX
];
1327 if (machine__is_default_guest(machine
)) {
1328 modules
= symbol_conf
.default_guest_modules
;
1330 snprintf(path
, PATH_MAX
, "%s/proc/modules", machine
->root_dir
);
1334 if (symbol__restricted_filename(modules
, "/proc/modules"))
1337 if (modules__parse(modules
, machine
, machine__create_module
))
1340 if (!machine__set_modules_path(machine
))
1343 pr_debug("Problems setting modules path maps, continuing anyway...\n");
1348 static void machine__set_kernel_mmap(struct machine
*machine
,
1351 machine
->vmlinux_map
->start
= start
;
1352 machine
->vmlinux_map
->end
= end
;
1354 * Be a bit paranoid here, some perf.data file came with
1355 * a zero sized synthesized MMAP event for the kernel.
1357 if (start
== 0 && end
== 0)
1358 machine
->vmlinux_map
->end
= ~0ULL;
1361 int machine__create_kernel_maps(struct machine
*machine
)
1363 struct dso
*kernel
= machine__get_kernel(machine
);
1364 const char *name
= NULL
;
1372 ret
= __machine__create_kernel_maps(machine
, kernel
);
1376 if (symbol_conf
.use_modules
&& machine__create_modules(machine
) < 0) {
1377 if (machine__is_host(machine
))
1378 pr_debug("Problems creating module maps, "
1379 "continuing anyway...\n");
1381 pr_debug("Problems creating module maps for guest %d, "
1382 "continuing anyway...\n", machine
->pid
);
1385 if (!machine__get_running_kernel_start(machine
, &name
, &addr
)) {
1387 map__set_kallsyms_ref_reloc_sym(machine
->vmlinux_map
, name
, addr
)) {
1388 machine__destroy_kernel_maps(machine
);
1393 /* we have a real start address now, so re-order the kmaps */
1394 map
= machine__kernel_map(machine
);
1397 map_groups__remove(&machine
->kmaps
, map
);
1399 /* assume it's the last in the kmaps */
1400 machine__set_kernel_mmap(machine
, addr
, ~0ULL);
1402 map_groups__insert(&machine
->kmaps
, map
);
1406 if (machine__create_extra_kernel_maps(machine
, kernel
))
1407 pr_debug("Problems creating extra kernel maps, continuing anyway...\n");
1409 /* update end address of the kernel map using adjacent module address */
1410 map
= map__next(machine__kernel_map(machine
));
1412 machine__set_kernel_mmap(machine
, addr
, map
->start
);
1418 static bool machine__uses_kcore(struct machine
*machine
)
1422 list_for_each_entry(dso
, &machine
->dsos
.head
, node
) {
1423 if (dso__is_kcore(dso
))
1430 static bool perf_event__is_extra_kernel_mmap(struct machine
*machine
,
1431 union perf_event
*event
)
1433 return machine__is(machine
, "x86_64") &&
1434 is_entry_trampoline(event
->mmap
.filename
);
1437 static int machine__process_extra_kernel_map(struct machine
*machine
,
1438 union perf_event
*event
)
1440 struct map
*kernel_map
= machine__kernel_map(machine
);
1441 struct dso
*kernel
= kernel_map
? kernel_map
->dso
: NULL
;
1442 struct extra_kernel_map xm
= {
1443 .start
= event
->mmap
.start
,
1444 .end
= event
->mmap
.start
+ event
->mmap
.len
,
1445 .pgoff
= event
->mmap
.pgoff
,
1451 strlcpy(xm
.name
, event
->mmap
.filename
, KMAP_NAME_LEN
);
1453 return machine__create_extra_kernel_map(machine
, kernel
, &xm
);
1456 static int machine__process_kernel_mmap_event(struct machine
*machine
,
1457 union perf_event
*event
)
1460 enum dso_kernel_type kernel_type
;
1461 bool is_kernel_mmap
;
1463 /* If we have maps from kcore then we do not need or want any others */
1464 if (machine__uses_kcore(machine
))
1467 if (machine__is_host(machine
))
1468 kernel_type
= DSO_TYPE_KERNEL
;
1470 kernel_type
= DSO_TYPE_GUEST_KERNEL
;
1472 is_kernel_mmap
= memcmp(event
->mmap
.filename
,
1474 strlen(machine
->mmap_name
) - 1) == 0;
1475 if (event
->mmap
.filename
[0] == '/' ||
1476 (!is_kernel_mmap
&& event
->mmap
.filename
[0] == '[')) {
1477 map
= machine__findnew_module_map(machine
, event
->mmap
.start
,
1478 event
->mmap
.filename
);
1482 map
->end
= map
->start
+ event
->mmap
.len
;
1483 } else if (is_kernel_mmap
) {
1484 const char *symbol_name
= (event
->mmap
.filename
+
1485 strlen(machine
->mmap_name
));
1487 * Should be there already, from the build-id table in
1490 struct dso
*kernel
= NULL
;
1493 down_read(&machine
->dsos
.lock
);
1495 list_for_each_entry(dso
, &machine
->dsos
.head
, node
) {
1498 * The cpumode passed to is_kernel_module is not the
1499 * cpumode of *this* event. If we insist on passing
1500 * correct cpumode to is_kernel_module, we should
1501 * record the cpumode when we adding this dso to the
1504 * However we don't really need passing correct
1505 * cpumode. We know the correct cpumode must be kernel
1506 * mode (if not, we should not link it onto kernel_dsos
1509 * Therefore, we pass PERF_RECORD_MISC_CPUMODE_UNKNOWN.
1510 * is_kernel_module() treats it as a kernel cpumode.
1514 is_kernel_module(dso
->long_name
,
1515 PERF_RECORD_MISC_CPUMODE_UNKNOWN
))
1523 up_read(&machine
->dsos
.lock
);
1526 kernel
= machine__findnew_dso(machine
, machine
->mmap_name
);
1530 kernel
->kernel
= kernel_type
;
1531 if (__machine__create_kernel_maps(machine
, kernel
) < 0) {
1536 if (strstr(kernel
->long_name
, "vmlinux"))
1537 dso__set_short_name(kernel
, "[kernel.vmlinux]", false);
1539 machine__set_kernel_mmap(machine
, event
->mmap
.start
,
1540 event
->mmap
.start
+ event
->mmap
.len
);
1543 * Avoid using a zero address (kptr_restrict) for the ref reloc
1544 * symbol. Effectively having zero here means that at record
1545 * time /proc/sys/kernel/kptr_restrict was non zero.
1547 if (event
->mmap
.pgoff
!= 0) {
1548 map__set_kallsyms_ref_reloc_sym(machine
->vmlinux_map
,
1553 if (machine__is_default_guest(machine
)) {
1555 * preload dso of guest kernel and modules
1557 dso__load(kernel
, machine__kernel_map(machine
));
1559 } else if (perf_event__is_extra_kernel_mmap(machine
, event
)) {
1560 return machine__process_extra_kernel_map(machine
, event
);
1567 int machine__process_mmap2_event(struct machine
*machine
,
1568 union perf_event
*event
,
1569 struct perf_sample
*sample
)
1571 struct thread
*thread
;
1576 perf_event__fprintf_mmap2(event
, stdout
);
1578 if (sample
->cpumode
== PERF_RECORD_MISC_GUEST_KERNEL
||
1579 sample
->cpumode
== PERF_RECORD_MISC_KERNEL
) {
1580 ret
= machine__process_kernel_mmap_event(machine
, event
);
1586 thread
= machine__findnew_thread(machine
, event
->mmap2
.pid
,
1591 map
= map__new(machine
, event
->mmap2
.start
,
1592 event
->mmap2
.len
, event
->mmap2
.pgoff
,
1594 event
->mmap2
.min
, event
->mmap2
.ino
,
1595 event
->mmap2
.ino_generation
,
1598 event
->mmap2
.filename
, thread
);
1601 goto out_problem_map
;
1603 ret
= thread__insert_map(thread
, map
);
1605 goto out_problem_insert
;
1607 thread__put(thread
);
1614 thread__put(thread
);
1616 dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
1620 int machine__process_mmap_event(struct machine
*machine
, union perf_event
*event
,
1621 struct perf_sample
*sample
)
1623 struct thread
*thread
;
1629 perf_event__fprintf_mmap(event
, stdout
);
1631 if (sample
->cpumode
== PERF_RECORD_MISC_GUEST_KERNEL
||
1632 sample
->cpumode
== PERF_RECORD_MISC_KERNEL
) {
1633 ret
= machine__process_kernel_mmap_event(machine
, event
);
1639 thread
= machine__findnew_thread(machine
, event
->mmap
.pid
,
1644 if (!(event
->header
.misc
& PERF_RECORD_MISC_MMAP_DATA
))
1647 map
= map__new(machine
, event
->mmap
.start
,
1648 event
->mmap
.len
, event
->mmap
.pgoff
,
1649 0, 0, 0, 0, prot
, 0,
1650 event
->mmap
.filename
,
1654 goto out_problem_map
;
1656 ret
= thread__insert_map(thread
, map
);
1658 goto out_problem_insert
;
1660 thread__put(thread
);
1667 thread__put(thread
);
1669 dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
1673 static void __machine__remove_thread(struct machine
*machine
, struct thread
*th
, bool lock
)
1675 struct threads
*threads
= machine__threads(machine
, th
->tid
);
1677 if (threads
->last_match
== th
)
1678 threads__set_last_match(threads
, NULL
);
1680 BUG_ON(refcount_read(&th
->refcnt
) == 0);
1682 down_write(&threads
->lock
);
1683 rb_erase_init(&th
->rb_node
, &threads
->entries
);
1684 RB_CLEAR_NODE(&th
->rb_node
);
1687 * Move it first to the dead_threads list, then drop the reference,
1688 * if this is the last reference, then the thread__delete destructor
1689 * will be called and we will remove it from the dead_threads list.
1691 list_add_tail(&th
->node
, &threads
->dead
);
1693 up_write(&threads
->lock
);
1697 void machine__remove_thread(struct machine
*machine
, struct thread
*th
)
1699 return __machine__remove_thread(machine
, th
, true);
1702 int machine__process_fork_event(struct machine
*machine
, union perf_event
*event
,
1703 struct perf_sample
*sample
)
1705 struct thread
*thread
= machine__find_thread(machine
,
1708 struct thread
*parent
= machine__findnew_thread(machine
,
1711 bool do_maps_clone
= true;
1715 perf_event__fprintf_task(event
, stdout
);
1718 * There may be an existing thread that is not actually the parent,
1719 * either because we are processing events out of order, or because the
1720 * (fork) event that would have removed the thread was lost. Assume the
1721 * latter case and continue on as best we can.
1723 if (parent
->pid_
!= (pid_t
)event
->fork
.ppid
) {
1724 dump_printf("removing erroneous parent thread %d/%d\n",
1725 parent
->pid_
, parent
->tid
);
1726 machine__remove_thread(machine
, parent
);
1727 thread__put(parent
);
1728 parent
= machine__findnew_thread(machine
, event
->fork
.ppid
,
1732 /* if a thread currently exists for the thread id remove it */
1733 if (thread
!= NULL
) {
1734 machine__remove_thread(machine
, thread
);
1735 thread__put(thread
);
1738 thread
= machine__findnew_thread(machine
, event
->fork
.pid
,
1741 * When synthesizing FORK events, we are trying to create thread
1742 * objects for the already running tasks on the machine.
1744 * Normally, for a kernel FORK event, we want to clone the parent's
1745 * maps because that is what the kernel just did.
1747 * But when synthesizing, this should not be done. If we do, we end up
1748 * with overlapping maps as we process the sythesized MMAP2 events that
1749 * get delivered shortly thereafter.
1751 * Use the FORK event misc flags in an internal way to signal this
1752 * situation, so we can elide the map clone when appropriate.
1754 if (event
->fork
.header
.misc
& PERF_RECORD_MISC_FORK_EXEC
)
1755 do_maps_clone
= false;
1757 if (thread
== NULL
|| parent
== NULL
||
1758 thread__fork(thread
, parent
, sample
->time
, do_maps_clone
) < 0) {
1759 dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
1762 thread__put(thread
);
1763 thread__put(parent
);
1768 int machine__process_exit_event(struct machine
*machine
, union perf_event
*event
,
1769 struct perf_sample
*sample __maybe_unused
)
1771 struct thread
*thread
= machine__find_thread(machine
,
1776 perf_event__fprintf_task(event
, stdout
);
1778 if (thread
!= NULL
) {
1779 thread__exited(thread
);
1780 thread__put(thread
);
1786 int machine__process_event(struct machine
*machine
, union perf_event
*event
,
1787 struct perf_sample
*sample
)
1791 switch (event
->header
.type
) {
1792 case PERF_RECORD_COMM
:
1793 ret
= machine__process_comm_event(machine
, event
, sample
); break;
1794 case PERF_RECORD_MMAP
:
1795 ret
= machine__process_mmap_event(machine
, event
, sample
); break;
1796 case PERF_RECORD_NAMESPACES
:
1797 ret
= machine__process_namespaces_event(machine
, event
, sample
); break;
1798 case PERF_RECORD_MMAP2
:
1799 ret
= machine__process_mmap2_event(machine
, event
, sample
); break;
1800 case PERF_RECORD_FORK
:
1801 ret
= machine__process_fork_event(machine
, event
, sample
); break;
1802 case PERF_RECORD_EXIT
:
1803 ret
= machine__process_exit_event(machine
, event
, sample
); break;
1804 case PERF_RECORD_LOST
:
1805 ret
= machine__process_lost_event(machine
, event
, sample
); break;
1806 case PERF_RECORD_AUX
:
1807 ret
= machine__process_aux_event(machine
, event
); break;
1808 case PERF_RECORD_ITRACE_START
:
1809 ret
= machine__process_itrace_start_event(machine
, event
); break;
1810 case PERF_RECORD_LOST_SAMPLES
:
1811 ret
= machine__process_lost_samples_event(machine
, event
, sample
); break;
1812 case PERF_RECORD_SWITCH
:
1813 case PERF_RECORD_SWITCH_CPU_WIDE
:
1814 ret
= machine__process_switch_event(machine
, event
); break;
1823 static bool symbol__match_regex(struct symbol
*sym
, regex_t
*regex
)
1825 if (!regexec(regex
, sym
->name
, 0, NULL
, 0))
1830 static void ip__resolve_ams(struct thread
*thread
,
1831 struct addr_map_symbol
*ams
,
1834 struct addr_location al
;
1836 memset(&al
, 0, sizeof(al
));
1838 * We cannot use the header.misc hint to determine whether a
1839 * branch stack address is user, kernel, guest, hypervisor.
1840 * Branches may straddle the kernel/user/hypervisor boundaries.
1841 * Thus, we have to try consecutively until we find a match
1842 * or else, the symbol is unknown
1844 thread__find_cpumode_addr_location(thread
, ip
, &al
);
1847 ams
->al_addr
= al
.addr
;
1853 static void ip__resolve_data(struct thread
*thread
,
1854 u8 m
, struct addr_map_symbol
*ams
,
1855 u64 addr
, u64 phys_addr
)
1857 struct addr_location al
;
1859 memset(&al
, 0, sizeof(al
));
1861 thread__find_symbol(thread
, m
, addr
, &al
);
1864 ams
->al_addr
= al
.addr
;
1867 ams
->phys_addr
= phys_addr
;
1870 struct mem_info
*sample__resolve_mem(struct perf_sample
*sample
,
1871 struct addr_location
*al
)
1873 struct mem_info
*mi
= mem_info__new();
1878 ip__resolve_ams(al
->thread
, &mi
->iaddr
, sample
->ip
);
1879 ip__resolve_data(al
->thread
, al
->cpumode
, &mi
->daddr
,
1880 sample
->addr
, sample
->phys_addr
);
1881 mi
->data_src
.val
= sample
->data_src
;
1886 static char *callchain_srcline(struct map
*map
, struct symbol
*sym
, u64 ip
)
1888 char *srcline
= NULL
;
1890 if (!map
|| callchain_param
.key
== CCKEY_FUNCTION
)
1893 srcline
= srcline__tree_find(&map
->dso
->srclines
, ip
);
1895 bool show_sym
= false;
1896 bool show_addr
= callchain_param
.key
== CCKEY_ADDRESS
;
1898 srcline
= get_srcline(map
->dso
, map__rip_2objdump(map
, ip
),
1899 sym
, show_sym
, show_addr
, ip
);
1900 srcline__tree_insert(&map
->dso
->srclines
, ip
, srcline
);
1911 static int add_callchain_ip(struct thread
*thread
,
1912 struct callchain_cursor
*cursor
,
1913 struct symbol
**parent
,
1914 struct addr_location
*root_al
,
1918 struct branch_flags
*flags
,
1919 struct iterations
*iter
,
1922 struct addr_location al
;
1923 int nr_loop_iter
= 0;
1924 u64 iter_cycles
= 0;
1925 const char *srcline
= NULL
;
1930 thread__find_cpumode_addr_location(thread
, ip
, &al
);
1932 if (ip
>= PERF_CONTEXT_MAX
) {
1934 case PERF_CONTEXT_HV
:
1935 *cpumode
= PERF_RECORD_MISC_HYPERVISOR
;
1937 case PERF_CONTEXT_KERNEL
:
1938 *cpumode
= PERF_RECORD_MISC_KERNEL
;
1940 case PERF_CONTEXT_USER
:
1941 *cpumode
= PERF_RECORD_MISC_USER
;
1944 pr_debug("invalid callchain context: "
1945 "%"PRId64
"\n", (s64
) ip
);
1947 * It seems the callchain is corrupted.
1950 callchain_cursor_reset(cursor
);
1955 thread__find_symbol(thread
, *cpumode
, ip
, &al
);
1958 if (al
.sym
!= NULL
) {
1959 if (perf_hpp_list
.parent
&& !*parent
&&
1960 symbol__match_regex(al
.sym
, &parent_regex
))
1962 else if (have_ignore_callees
&& root_al
&&
1963 symbol__match_regex(al
.sym
, &ignore_callees_regex
)) {
1964 /* Treat this symbol as the root,
1965 forgetting its callees. */
1967 callchain_cursor_reset(cursor
);
1971 if (symbol_conf
.hide_unresolved
&& al
.sym
== NULL
)
1975 nr_loop_iter
= iter
->nr_loop_iter
;
1976 iter_cycles
= iter
->cycles
;
1979 srcline
= callchain_srcline(al
.map
, al
.sym
, al
.addr
);
1980 return callchain_cursor_append(cursor
, ip
, al
.map
, al
.sym
,
1981 branch
, flags
, nr_loop_iter
,
1982 iter_cycles
, branch_from
, srcline
);
1985 struct branch_info
*sample__resolve_bstack(struct perf_sample
*sample
,
1986 struct addr_location
*al
)
1989 const struct branch_stack
*bs
= sample
->branch_stack
;
1990 struct branch_info
*bi
= calloc(bs
->nr
, sizeof(struct branch_info
));
1995 for (i
= 0; i
< bs
->nr
; i
++) {
1996 ip__resolve_ams(al
->thread
, &bi
[i
].to
, bs
->entries
[i
].to
);
1997 ip__resolve_ams(al
->thread
, &bi
[i
].from
, bs
->entries
[i
].from
);
1998 bi
[i
].flags
= bs
->entries
[i
].flags
;
2003 static void save_iterations(struct iterations
*iter
,
2004 struct branch_entry
*be
, int nr
)
2008 iter
->nr_loop_iter
= nr
;
2011 for (i
= 0; i
< nr
; i
++)
2012 iter
->cycles
+= be
[i
].flags
.cycles
;
2017 #define NO_ENTRY 0xff
2019 #define PERF_MAX_BRANCH_DEPTH 127
2022 static int remove_loops(struct branch_entry
*l
, int nr
,
2023 struct iterations
*iter
)
2026 unsigned char chash
[CHASHSZ
];
2028 memset(chash
, NO_ENTRY
, sizeof(chash
));
2030 BUG_ON(PERF_MAX_BRANCH_DEPTH
> 255);
2032 for (i
= 0; i
< nr
; i
++) {
2033 int h
= hash_64(l
[i
].from
, CHASHBITS
) % CHASHSZ
;
2035 /* no collision handling for now */
2036 if (chash
[h
] == NO_ENTRY
) {
2038 } else if (l
[chash
[h
]].from
== l
[i
].from
) {
2039 bool is_loop
= true;
2040 /* check if it is a real loop */
2042 for (j
= chash
[h
]; j
< i
&& i
+ off
< nr
; j
++, off
++)
2043 if (l
[j
].from
!= l
[i
+ off
].from
) {
2050 save_iterations(iter
+ i
+ off
,
2053 memmove(iter
+ i
, iter
+ i
+ off
,
2056 memmove(l
+ i
, l
+ i
+ off
,
2068 * Recolve LBR callstack chain sample
2070 * 1 on success get LBR callchain information
2071 * 0 no available LBR callchain information, should try fp
2072 * negative error code on other errors.
2074 static int resolve_lbr_callchain_sample(struct thread
*thread
,
2075 struct callchain_cursor
*cursor
,
2076 struct perf_sample
*sample
,
2077 struct symbol
**parent
,
2078 struct addr_location
*root_al
,
2081 struct ip_callchain
*chain
= sample
->callchain
;
2082 int chain_nr
= min(max_stack
, (int)chain
->nr
), i
;
2083 u8 cpumode
= PERF_RECORD_MISC_USER
;
2084 u64 ip
, branch_from
= 0;
2086 for (i
= 0; i
< chain_nr
; i
++) {
2087 if (chain
->ips
[i
] == PERF_CONTEXT_USER
)
2091 /* LBR only affects the user callchain */
2092 if (i
!= chain_nr
) {
2093 struct branch_stack
*lbr_stack
= sample
->branch_stack
;
2094 int lbr_nr
= lbr_stack
->nr
, j
, k
;
2096 struct branch_flags
*flags
;
2098 * LBR callstack can only get user call chain.
2099 * The mix_chain_nr is kernel call chain
2100 * number plus LBR user call chain number.
2101 * i is kernel call chain number,
2102 * 1 is PERF_CONTEXT_USER,
2103 * lbr_nr + 1 is the user call chain number.
2104 * For details, please refer to the comments
2105 * in callchain__printf
2107 int mix_chain_nr
= i
+ 1 + lbr_nr
+ 1;
2109 for (j
= 0; j
< mix_chain_nr
; j
++) {
2114 if (callchain_param
.order
== ORDER_CALLEE
) {
2117 else if (j
> i
+ 1) {
2119 ip
= lbr_stack
->entries
[k
].from
;
2121 flags
= &lbr_stack
->entries
[k
].flags
;
2123 ip
= lbr_stack
->entries
[0].to
;
2125 flags
= &lbr_stack
->entries
[0].flags
;
2127 lbr_stack
->entries
[0].from
;
2132 ip
= lbr_stack
->entries
[k
].from
;
2134 flags
= &lbr_stack
->entries
[k
].flags
;
2136 else if (j
> lbr_nr
)
2137 ip
= chain
->ips
[i
+ 1 - (j
- lbr_nr
)];
2139 ip
= lbr_stack
->entries
[0].to
;
2141 flags
= &lbr_stack
->entries
[0].flags
;
2143 lbr_stack
->entries
[0].from
;
2147 err
= add_callchain_ip(thread
, cursor
, parent
,
2148 root_al
, &cpumode
, ip
,
2149 branch
, flags
, NULL
,
2152 return (err
< 0) ? err
: 0;
2160 static int find_prev_cpumode(struct ip_callchain
*chain
, struct thread
*thread
,
2161 struct callchain_cursor
*cursor
,
2162 struct symbol
**parent
,
2163 struct addr_location
*root_al
,
2164 u8
*cpumode
, int ent
)
2168 while (--ent
>= 0) {
2169 u64 ip
= chain
->ips
[ent
];
2171 if (ip
>= PERF_CONTEXT_MAX
) {
2172 err
= add_callchain_ip(thread
, cursor
, parent
,
2173 root_al
, cpumode
, ip
,
2174 false, NULL
, NULL
, 0);
2181 static int thread__resolve_callchain_sample(struct thread
*thread
,
2182 struct callchain_cursor
*cursor
,
2183 struct perf_evsel
*evsel
,
2184 struct perf_sample
*sample
,
2185 struct symbol
**parent
,
2186 struct addr_location
*root_al
,
2189 struct branch_stack
*branch
= sample
->branch_stack
;
2190 struct ip_callchain
*chain
= sample
->callchain
;
2192 u8 cpumode
= PERF_RECORD_MISC_USER
;
2193 int i
, j
, err
, nr_entries
;
2198 chain_nr
= chain
->nr
;
2200 if (perf_evsel__has_branch_callstack(evsel
)) {
2201 err
= resolve_lbr_callchain_sample(thread
, cursor
, sample
, parent
,
2202 root_al
, max_stack
);
2204 return (err
< 0) ? err
: 0;
2208 * Based on DWARF debug information, some architectures skip
2209 * a callchain entry saved by the kernel.
2211 skip_idx
= arch_skip_callchain_idx(thread
, chain
);
2214 * Add branches to call stack for easier browsing. This gives
2215 * more context for a sample than just the callers.
2217 * This uses individual histograms of paths compared to the
2218 * aggregated histograms the normal LBR mode uses.
2220 * Limitations for now:
2221 * - No extra filters
2222 * - No annotations (should annotate somehow)
2225 if (branch
&& callchain_param
.branch_callstack
) {
2226 int nr
= min(max_stack
, (int)branch
->nr
);
2227 struct branch_entry be
[nr
];
2228 struct iterations iter
[nr
];
2230 if (branch
->nr
> PERF_MAX_BRANCH_DEPTH
) {
2231 pr_warning("corrupted branch chain. skipping...\n");
2235 for (i
= 0; i
< nr
; i
++) {
2236 if (callchain_param
.order
== ORDER_CALLEE
) {
2237 be
[i
] = branch
->entries
[i
];
2243 * Check for overlap into the callchain.
2244 * The return address is one off compared to
2245 * the branch entry. To adjust for this
2246 * assume the calling instruction is not longer
2249 if (i
== skip_idx
||
2250 chain
->ips
[first_call
] >= PERF_CONTEXT_MAX
)
2252 else if (be
[i
].from
< chain
->ips
[first_call
] &&
2253 be
[i
].from
>= chain
->ips
[first_call
] - 8)
2256 be
[i
] = branch
->entries
[branch
->nr
- i
- 1];
2259 memset(iter
, 0, sizeof(struct iterations
) * nr
);
2260 nr
= remove_loops(be
, nr
, iter
);
2262 for (i
= 0; i
< nr
; i
++) {
2263 err
= add_callchain_ip(thread
, cursor
, parent
,
2270 err
= add_callchain_ip(thread
, cursor
, parent
, root_al
,
2287 if (callchain_param
.order
!= ORDER_CALLEE
) {
2288 err
= find_prev_cpumode(chain
, thread
, cursor
, parent
, root_al
,
2289 &cpumode
, chain
->nr
- first_call
);
2291 return (err
< 0) ? err
: 0;
2293 for (i
= first_call
, nr_entries
= 0;
2294 i
< chain_nr
&& nr_entries
< max_stack
; i
++) {
2297 if (callchain_param
.order
== ORDER_CALLEE
)
2300 j
= chain
->nr
- i
- 1;
2302 #ifdef HAVE_SKIP_CALLCHAIN_IDX
2307 if (ip
< PERF_CONTEXT_MAX
)
2309 else if (callchain_param
.order
!= ORDER_CALLEE
) {
2310 err
= find_prev_cpumode(chain
, thread
, cursor
, parent
,
2311 root_al
, &cpumode
, j
);
2313 return (err
< 0) ? err
: 0;
2317 err
= add_callchain_ip(thread
, cursor
, parent
,
2318 root_al
, &cpumode
, ip
,
2319 false, NULL
, NULL
, 0);
2322 return (err
< 0) ? err
: 0;
2328 static int append_inlines(struct callchain_cursor
*cursor
,
2329 struct map
*map
, struct symbol
*sym
, u64 ip
)
2331 struct inline_node
*inline_node
;
2332 struct inline_list
*ilist
;
2336 if (!symbol_conf
.inline_name
|| !map
|| !sym
)
2339 addr
= map__map_ip(map
, ip
);
2340 addr
= map__rip_2objdump(map
, addr
);
2342 inline_node
= inlines__tree_find(&map
->dso
->inlined_nodes
, addr
);
2344 inline_node
= dso__parse_addr_inlines(map
->dso
, addr
, sym
);
2347 inlines__tree_insert(&map
->dso
->inlined_nodes
, inline_node
);
2350 list_for_each_entry(ilist
, &inline_node
->val
, list
) {
2351 ret
= callchain_cursor_append(cursor
, ip
, map
,
2352 ilist
->symbol
, false,
2353 NULL
, 0, 0, 0, ilist
->srcline
);
2362 static int unwind_entry(struct unwind_entry
*entry
, void *arg
)
2364 struct callchain_cursor
*cursor
= arg
;
2365 const char *srcline
= NULL
;
2366 u64 addr
= entry
->ip
;
2368 if (symbol_conf
.hide_unresolved
&& entry
->sym
== NULL
)
2371 if (append_inlines(cursor
, entry
->map
, entry
->sym
, entry
->ip
) == 0)
2375 * Convert entry->ip from a virtual address to an offset in
2376 * its corresponding binary.
2379 addr
= map__map_ip(entry
->map
, entry
->ip
);
2381 srcline
= callchain_srcline(entry
->map
, entry
->sym
, addr
);
2382 return callchain_cursor_append(cursor
, entry
->ip
,
2383 entry
->map
, entry
->sym
,
2384 false, NULL
, 0, 0, 0, srcline
);
2387 static int thread__resolve_callchain_unwind(struct thread
*thread
,
2388 struct callchain_cursor
*cursor
,
2389 struct perf_evsel
*evsel
,
2390 struct perf_sample
*sample
,
2393 /* Can we do dwarf post unwind? */
2394 if (!((evsel
->attr
.sample_type
& PERF_SAMPLE_REGS_USER
) &&
2395 (evsel
->attr
.sample_type
& PERF_SAMPLE_STACK_USER
)))
2398 /* Bail out if nothing was captured. */
2399 if ((!sample
->user_regs
.regs
) ||
2400 (!sample
->user_stack
.size
))
2403 return unwind__get_entries(unwind_entry
, cursor
,
2404 thread
, sample
, max_stack
);
2407 int thread__resolve_callchain(struct thread
*thread
,
2408 struct callchain_cursor
*cursor
,
2409 struct perf_evsel
*evsel
,
2410 struct perf_sample
*sample
,
2411 struct symbol
**parent
,
2412 struct addr_location
*root_al
,
2417 callchain_cursor_reset(cursor
);
2419 if (callchain_param
.order
== ORDER_CALLEE
) {
2420 ret
= thread__resolve_callchain_sample(thread
, cursor
,
2426 ret
= thread__resolve_callchain_unwind(thread
, cursor
,
2430 ret
= thread__resolve_callchain_unwind(thread
, cursor
,
2435 ret
= thread__resolve_callchain_sample(thread
, cursor
,
2444 int machine__for_each_thread(struct machine
*machine
,
2445 int (*fn
)(struct thread
*thread
, void *p
),
2448 struct threads
*threads
;
2450 struct thread
*thread
;
2454 for (i
= 0; i
< THREADS__TABLE_SIZE
; i
++) {
2455 threads
= &machine
->threads
[i
];
2456 for (nd
= rb_first(&threads
->entries
); nd
; nd
= rb_next(nd
)) {
2457 thread
= rb_entry(nd
, struct thread
, rb_node
);
2458 rc
= fn(thread
, priv
);
2463 list_for_each_entry(thread
, &threads
->dead
, node
) {
2464 rc
= fn(thread
, priv
);
2472 int machines__for_each_thread(struct machines
*machines
,
2473 int (*fn
)(struct thread
*thread
, void *p
),
2479 rc
= machine__for_each_thread(&machines
->host
, fn
, priv
);
2483 for (nd
= rb_first(&machines
->guests
); nd
; nd
= rb_next(nd
)) {
2484 struct machine
*machine
= rb_entry(nd
, struct machine
, rb_node
);
2486 rc
= machine__for_each_thread(machine
, fn
, priv
);
2493 int __machine__synthesize_threads(struct machine
*machine
, struct perf_tool
*tool
,
2494 struct target
*target
, struct thread_map
*threads
,
2495 perf_event__handler_t process
, bool data_mmap
,
2496 unsigned int proc_map_timeout
,
2497 unsigned int nr_threads_synthesize
)
2499 if (target__has_task(target
))
2500 return perf_event__synthesize_thread_map(tool
, threads
, process
, machine
, data_mmap
, proc_map_timeout
);
2501 else if (target__has_cpu(target
))
2502 return perf_event__synthesize_threads(tool
, process
,
2505 nr_threads_synthesize
);
2506 /* command specified */
2510 pid_t
machine__get_current_tid(struct machine
*machine
, int cpu
)
2512 if (cpu
< 0 || cpu
>= MAX_NR_CPUS
|| !machine
->current_tid
)
2515 return machine
->current_tid
[cpu
];
2518 int machine__set_current_tid(struct machine
*machine
, int cpu
, pid_t pid
,
2521 struct thread
*thread
;
2526 if (!machine
->current_tid
) {
2529 machine
->current_tid
= calloc(MAX_NR_CPUS
, sizeof(pid_t
));
2530 if (!machine
->current_tid
)
2532 for (i
= 0; i
< MAX_NR_CPUS
; i
++)
2533 machine
->current_tid
[i
] = -1;
2536 if (cpu
>= MAX_NR_CPUS
) {
2537 pr_err("Requested CPU %d too large. ", cpu
);
2538 pr_err("Consider raising MAX_NR_CPUS\n");
2542 machine
->current_tid
[cpu
] = tid
;
2544 thread
= machine__findnew_thread(machine
, pid
, tid
);
2549 thread__put(thread
);
2555 * Compares the raw arch string. N.B. see instead perf_env__arch() if a
2556 * normalized arch is needed.
2558 bool machine__is(struct machine
*machine
, const char *arch
)
2560 return machine
&& !strcmp(perf_env__raw_arch(machine
->env
), arch
);
2563 int machine__nr_cpus_avail(struct machine
*machine
)
2565 return machine
? perf_env__nr_cpus_avail(machine
->env
) : 0;
2568 int machine__get_kernel_start(struct machine
*machine
)
2570 struct map
*map
= machine__kernel_map(machine
);
2574 * The only addresses above 2^63 are kernel addresses of a 64-bit
2575 * kernel. Note that addresses are unsigned so that on a 32-bit system
2576 * all addresses including kernel addresses are less than 2^32. In
2577 * that case (32-bit system), if the kernel mapping is unknown, all
2578 * addresses will be assumed to be in user space - see
2579 * machine__kernel_ip().
2581 machine
->kernel_start
= 1ULL << 63;
2583 err
= map__load(map
);
2585 * On x86_64, PTI entry trampolines are less than the
2586 * start of kernel text, but still above 2^63. So leave
2587 * kernel_start = 1ULL << 63 for x86_64.
2589 if (!err
&& !machine__is(machine
, "x86_64"))
2590 machine
->kernel_start
= map
->start
;
2595 struct dso
*machine__findnew_dso(struct machine
*machine
, const char *filename
)
2597 return dsos__findnew(&machine
->dsos
, filename
);
2600 char *machine__resolve_kernel_addr(void *vmachine
, unsigned long long *addrp
, char **modp
)
2602 struct machine
*machine
= vmachine
;
2604 struct symbol
*sym
= machine__find_kernel_symbol(machine
, *addrp
, &map
);
2609 *modp
= __map__is_kmodule(map
) ? (char *)map
->dso
->short_name
: NULL
;
2610 *addrp
= map
->unmap_ip(map
, sym
->start
);