1 #include <linux/kernel.h>
2 #include <traceevent/event-parse.h>
16 #include "perf_regs.h"
19 #include "thread-stack.h"
22 static int perf_session__deliver_event(struct perf_session
*session
,
23 union perf_event
*event
,
24 struct perf_sample
*sample
,
25 struct perf_tool
*tool
,
28 static int perf_session__open(struct perf_session
*session
)
30 struct perf_data_file
*file
= session
->file
;
32 if (perf_session__read_header(session
) < 0) {
33 pr_err("incompatible file format (rerun with -v to learn more)\n");
37 if (perf_data_file__is_pipe(file
))
40 if (perf_header__has_feat(&session
->header
, HEADER_STAT
))
43 if (!perf_evlist__valid_sample_type(session
->evlist
)) {
44 pr_err("non matching sample_type\n");
48 if (!perf_evlist__valid_sample_id_all(session
->evlist
)) {
49 pr_err("non matching sample_id_all\n");
53 if (!perf_evlist__valid_read_format(session
->evlist
)) {
54 pr_err("non matching read_format\n");
61 void perf_session__set_id_hdr_size(struct perf_session
*session
)
63 u16 id_hdr_size
= perf_evlist__id_hdr_size(session
->evlist
);
65 machines__set_id_hdr_size(&session
->machines
, id_hdr_size
);
68 int perf_session__create_kernel_maps(struct perf_session
*session
)
70 int ret
= machine__create_kernel_maps(&session
->machines
.host
);
73 ret
= machines__create_guest_kernel_maps(&session
->machines
);
77 static void perf_session__destroy_kernel_maps(struct perf_session
*session
)
79 machines__destroy_kernel_maps(&session
->machines
);
82 static bool perf_session__has_comm_exec(struct perf_session
*session
)
84 struct perf_evsel
*evsel
;
86 evlist__for_each(session
->evlist
, evsel
) {
87 if (evsel
->attr
.comm_exec
)
94 static void perf_session__set_comm_exec(struct perf_session
*session
)
96 bool comm_exec
= perf_session__has_comm_exec(session
);
98 machines__set_comm_exec(&session
->machines
, comm_exec
);
101 static int ordered_events__deliver_event(struct ordered_events
*oe
,
102 struct ordered_event
*event
)
104 struct perf_sample sample
;
105 struct perf_session
*session
= container_of(oe
, struct perf_session
,
107 int ret
= perf_evlist__parse_sample(session
->evlist
, event
->event
, &sample
);
110 pr_err("Can't parse sample, err = %d\n", ret
);
114 return perf_session__deliver_event(session
, event
->event
, &sample
,
115 session
->tool
, event
->file_offset
);
118 struct perf_session
*perf_session__new(struct perf_data_file
*file
,
119 bool repipe
, struct perf_tool
*tool
)
121 struct perf_session
*session
= zalloc(sizeof(*session
));
126 session
->repipe
= repipe
;
127 session
->tool
= tool
;
128 INIT_LIST_HEAD(&session
->auxtrace_index
);
129 machines__init(&session
->machines
);
130 ordered_events__init(&session
->ordered_events
, ordered_events__deliver_event
);
133 if (perf_data_file__open(file
))
136 session
->file
= file
;
138 if (perf_data_file__is_read(file
)) {
139 if (perf_session__open(session
) < 0)
142 perf_session__set_id_hdr_size(session
);
143 perf_session__set_comm_exec(session
);
146 session
->machines
.host
.env
= &perf_env
;
149 if (!file
|| perf_data_file__is_write(file
)) {
151 * In O_RDONLY mode this will be performed when reading the
152 * kernel MMAP event, in perf_event__process_mmap().
154 if (perf_session__create_kernel_maps(session
) < 0)
155 pr_warning("Cannot read kernel map\n");
158 if (tool
&& tool
->ordering_requires_timestamps
&&
159 tool
->ordered_events
&& !perf_evlist__sample_id_all(session
->evlist
)) {
160 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
161 tool
->ordered_events
= false;
167 perf_data_file__close(file
);
169 perf_session__delete(session
);
174 static void perf_session__delete_threads(struct perf_session
*session
)
176 machine__delete_threads(&session
->machines
.host
);
179 void perf_session__delete(struct perf_session
*session
)
181 auxtrace__free(session
);
182 auxtrace_index__free(&session
->auxtrace_index
);
183 perf_session__destroy_kernel_maps(session
);
184 perf_session__delete_threads(session
);
185 perf_env__exit(&session
->header
.env
);
186 machines__exit(&session
->machines
);
188 perf_data_file__close(session
->file
);
192 static int process_event_synth_tracing_data_stub(struct perf_tool
*tool
194 union perf_event
*event
196 struct perf_session
*session
199 dump_printf(": unhandled!\n");
203 static int process_event_synth_attr_stub(struct perf_tool
*tool __maybe_unused
,
204 union perf_event
*event __maybe_unused
,
205 struct perf_evlist
**pevlist
208 dump_printf(": unhandled!\n");
212 static int process_event_synth_event_update_stub(struct perf_tool
*tool __maybe_unused
,
213 union perf_event
*event __maybe_unused
,
214 struct perf_evlist
**pevlist
218 perf_event__fprintf_event_update(event
, stdout
);
220 dump_printf(": unhandled!\n");
224 static int process_event_sample_stub(struct perf_tool
*tool __maybe_unused
,
225 union perf_event
*event __maybe_unused
,
226 struct perf_sample
*sample __maybe_unused
,
227 struct perf_evsel
*evsel __maybe_unused
,
228 struct machine
*machine __maybe_unused
)
230 dump_printf(": unhandled!\n");
234 static int process_event_stub(struct perf_tool
*tool __maybe_unused
,
235 union perf_event
*event __maybe_unused
,
236 struct perf_sample
*sample __maybe_unused
,
237 struct machine
*machine __maybe_unused
)
239 dump_printf(": unhandled!\n");
243 static int process_finished_round_stub(struct perf_tool
*tool __maybe_unused
,
244 union perf_event
*event __maybe_unused
,
245 struct ordered_events
*oe __maybe_unused
)
247 dump_printf(": unhandled!\n");
251 static int process_finished_round(struct perf_tool
*tool
,
252 union perf_event
*event
,
253 struct ordered_events
*oe
);
255 static int skipn(int fd
, off_t n
)
261 ret
= read(fd
, buf
, min(n
, (off_t
)sizeof(buf
)));
270 static s64
process_event_auxtrace_stub(struct perf_tool
*tool __maybe_unused
,
271 union perf_event
*event
,
272 struct perf_session
*session
275 dump_printf(": unhandled!\n");
276 if (perf_data_file__is_pipe(session
->file
))
277 skipn(perf_data_file__fd(session
->file
), event
->auxtrace
.size
);
278 return event
->auxtrace
.size
;
281 static int process_event_op2_stub(struct perf_tool
*tool __maybe_unused
,
282 union perf_event
*event __maybe_unused
,
283 struct perf_session
*session __maybe_unused
)
285 dump_printf(": unhandled!\n");
291 int process_event_thread_map_stub(struct perf_tool
*tool __maybe_unused
,
292 union perf_event
*event __maybe_unused
,
293 struct perf_session
*session __maybe_unused
)
296 perf_event__fprintf_thread_map(event
, stdout
);
298 dump_printf(": unhandled!\n");
303 int process_event_cpu_map_stub(struct perf_tool
*tool __maybe_unused
,
304 union perf_event
*event __maybe_unused
,
305 struct perf_session
*session __maybe_unused
)
308 perf_event__fprintf_cpu_map(event
, stdout
);
310 dump_printf(": unhandled!\n");
315 int process_event_stat_config_stub(struct perf_tool
*tool __maybe_unused
,
316 union perf_event
*event __maybe_unused
,
317 struct perf_session
*session __maybe_unused
)
320 perf_event__fprintf_stat_config(event
, stdout
);
322 dump_printf(": unhandled!\n");
326 static int process_stat_stub(struct perf_tool
*tool __maybe_unused
,
327 union perf_event
*event __maybe_unused
,
328 struct perf_session
*perf_session
332 perf_event__fprintf_stat(event
, stdout
);
334 dump_printf(": unhandled!\n");
338 static int process_stat_round_stub(struct perf_tool
*tool __maybe_unused
,
339 union perf_event
*event __maybe_unused
,
340 struct perf_session
*perf_session
344 perf_event__fprintf_stat_round(event
, stdout
);
346 dump_printf(": unhandled!\n");
350 void perf_tool__fill_defaults(struct perf_tool
*tool
)
352 if (tool
->sample
== NULL
)
353 tool
->sample
= process_event_sample_stub
;
354 if (tool
->mmap
== NULL
)
355 tool
->mmap
= process_event_stub
;
356 if (tool
->mmap2
== NULL
)
357 tool
->mmap2
= process_event_stub
;
358 if (tool
->comm
== NULL
)
359 tool
->comm
= process_event_stub
;
360 if (tool
->fork
== NULL
)
361 tool
->fork
= process_event_stub
;
362 if (tool
->exit
== NULL
)
363 tool
->exit
= process_event_stub
;
364 if (tool
->lost
== NULL
)
365 tool
->lost
= perf_event__process_lost
;
366 if (tool
->lost_samples
== NULL
)
367 tool
->lost_samples
= perf_event__process_lost_samples
;
368 if (tool
->aux
== NULL
)
369 tool
->aux
= perf_event__process_aux
;
370 if (tool
->itrace_start
== NULL
)
371 tool
->itrace_start
= perf_event__process_itrace_start
;
372 if (tool
->context_switch
== NULL
)
373 tool
->context_switch
= perf_event__process_switch
;
374 if (tool
->read
== NULL
)
375 tool
->read
= process_event_sample_stub
;
376 if (tool
->throttle
== NULL
)
377 tool
->throttle
= process_event_stub
;
378 if (tool
->unthrottle
== NULL
)
379 tool
->unthrottle
= process_event_stub
;
380 if (tool
->attr
== NULL
)
381 tool
->attr
= process_event_synth_attr_stub
;
382 if (tool
->event_update
== NULL
)
383 tool
->event_update
= process_event_synth_event_update_stub
;
384 if (tool
->tracing_data
== NULL
)
385 tool
->tracing_data
= process_event_synth_tracing_data_stub
;
386 if (tool
->build_id
== NULL
)
387 tool
->build_id
= process_event_op2_stub
;
388 if (tool
->finished_round
== NULL
) {
389 if (tool
->ordered_events
)
390 tool
->finished_round
= process_finished_round
;
392 tool
->finished_round
= process_finished_round_stub
;
394 if (tool
->id_index
== NULL
)
395 tool
->id_index
= process_event_op2_stub
;
396 if (tool
->auxtrace_info
== NULL
)
397 tool
->auxtrace_info
= process_event_op2_stub
;
398 if (tool
->auxtrace
== NULL
)
399 tool
->auxtrace
= process_event_auxtrace_stub
;
400 if (tool
->auxtrace_error
== NULL
)
401 tool
->auxtrace_error
= process_event_op2_stub
;
402 if (tool
->thread_map
== NULL
)
403 tool
->thread_map
= process_event_thread_map_stub
;
404 if (tool
->cpu_map
== NULL
)
405 tool
->cpu_map
= process_event_cpu_map_stub
;
406 if (tool
->stat_config
== NULL
)
407 tool
->stat_config
= process_event_stat_config_stub
;
408 if (tool
->stat
== NULL
)
409 tool
->stat
= process_stat_stub
;
410 if (tool
->stat_round
== NULL
)
411 tool
->stat_round
= process_stat_round_stub
;
414 static void swap_sample_id_all(union perf_event
*event
, void *data
)
416 void *end
= (void *) event
+ event
->header
.size
;
417 int size
= end
- data
;
419 BUG_ON(size
% sizeof(u64
));
420 mem_bswap_64(data
, size
);
423 static void perf_event__all64_swap(union perf_event
*event
,
424 bool sample_id_all __maybe_unused
)
426 struct perf_event_header
*hdr
= &event
->header
;
427 mem_bswap_64(hdr
+ 1, event
->header
.size
- sizeof(*hdr
));
430 static void perf_event__comm_swap(union perf_event
*event
, bool sample_id_all
)
432 event
->comm
.pid
= bswap_32(event
->comm
.pid
);
433 event
->comm
.tid
= bswap_32(event
->comm
.tid
);
436 void *data
= &event
->comm
.comm
;
438 data
+= PERF_ALIGN(strlen(data
) + 1, sizeof(u64
));
439 swap_sample_id_all(event
, data
);
443 static void perf_event__mmap_swap(union perf_event
*event
,
446 event
->mmap
.pid
= bswap_32(event
->mmap
.pid
);
447 event
->mmap
.tid
= bswap_32(event
->mmap
.tid
);
448 event
->mmap
.start
= bswap_64(event
->mmap
.start
);
449 event
->mmap
.len
= bswap_64(event
->mmap
.len
);
450 event
->mmap
.pgoff
= bswap_64(event
->mmap
.pgoff
);
453 void *data
= &event
->mmap
.filename
;
455 data
+= PERF_ALIGN(strlen(data
) + 1, sizeof(u64
));
456 swap_sample_id_all(event
, data
);
460 static void perf_event__mmap2_swap(union perf_event
*event
,
463 event
->mmap2
.pid
= bswap_32(event
->mmap2
.pid
);
464 event
->mmap2
.tid
= bswap_32(event
->mmap2
.tid
);
465 event
->mmap2
.start
= bswap_64(event
->mmap2
.start
);
466 event
->mmap2
.len
= bswap_64(event
->mmap2
.len
);
467 event
->mmap2
.pgoff
= bswap_64(event
->mmap2
.pgoff
);
468 event
->mmap2
.maj
= bswap_32(event
->mmap2
.maj
);
469 event
->mmap2
.min
= bswap_32(event
->mmap2
.min
);
470 event
->mmap2
.ino
= bswap_64(event
->mmap2
.ino
);
473 void *data
= &event
->mmap2
.filename
;
475 data
+= PERF_ALIGN(strlen(data
) + 1, sizeof(u64
));
476 swap_sample_id_all(event
, data
);
479 static void perf_event__task_swap(union perf_event
*event
, bool sample_id_all
)
481 event
->fork
.pid
= bswap_32(event
->fork
.pid
);
482 event
->fork
.tid
= bswap_32(event
->fork
.tid
);
483 event
->fork
.ppid
= bswap_32(event
->fork
.ppid
);
484 event
->fork
.ptid
= bswap_32(event
->fork
.ptid
);
485 event
->fork
.time
= bswap_64(event
->fork
.time
);
488 swap_sample_id_all(event
, &event
->fork
+ 1);
491 static void perf_event__read_swap(union perf_event
*event
, bool sample_id_all
)
493 event
->read
.pid
= bswap_32(event
->read
.pid
);
494 event
->read
.tid
= bswap_32(event
->read
.tid
);
495 event
->read
.value
= bswap_64(event
->read
.value
);
496 event
->read
.time_enabled
= bswap_64(event
->read
.time_enabled
);
497 event
->read
.time_running
= bswap_64(event
->read
.time_running
);
498 event
->read
.id
= bswap_64(event
->read
.id
);
501 swap_sample_id_all(event
, &event
->read
+ 1);
504 static void perf_event__aux_swap(union perf_event
*event
, bool sample_id_all
)
506 event
->aux
.aux_offset
= bswap_64(event
->aux
.aux_offset
);
507 event
->aux
.aux_size
= bswap_64(event
->aux
.aux_size
);
508 event
->aux
.flags
= bswap_64(event
->aux
.flags
);
511 swap_sample_id_all(event
, &event
->aux
+ 1);
514 static void perf_event__itrace_start_swap(union perf_event
*event
,
517 event
->itrace_start
.pid
= bswap_32(event
->itrace_start
.pid
);
518 event
->itrace_start
.tid
= bswap_32(event
->itrace_start
.tid
);
521 swap_sample_id_all(event
, &event
->itrace_start
+ 1);
524 static void perf_event__switch_swap(union perf_event
*event
, bool sample_id_all
)
526 if (event
->header
.type
== PERF_RECORD_SWITCH_CPU_WIDE
) {
527 event
->context_switch
.next_prev_pid
=
528 bswap_32(event
->context_switch
.next_prev_pid
);
529 event
->context_switch
.next_prev_tid
=
530 bswap_32(event
->context_switch
.next_prev_tid
);
534 swap_sample_id_all(event
, &event
->context_switch
+ 1);
537 static void perf_event__throttle_swap(union perf_event
*event
,
540 event
->throttle
.time
= bswap_64(event
->throttle
.time
);
541 event
->throttle
.id
= bswap_64(event
->throttle
.id
);
542 event
->throttle
.stream_id
= bswap_64(event
->throttle
.stream_id
);
545 swap_sample_id_all(event
, &event
->throttle
+ 1);
548 static u8
revbyte(u8 b
)
550 int rev
= (b
>> 4) | ((b
& 0xf) << 4);
551 rev
= ((rev
& 0xcc) >> 2) | ((rev
& 0x33) << 2);
552 rev
= ((rev
& 0xaa) >> 1) | ((rev
& 0x55) << 1);
557 * XXX this is hack in attempt to carry flags bitfield
558 * throught endian village. ABI says:
560 * Bit-fields are allocated from right to left (least to most significant)
561 * on little-endian implementations and from left to right (most to least
562 * significant) on big-endian implementations.
564 * The above seems to be byte specific, so we need to reverse each
565 * byte of the bitfield. 'Internet' also says this might be implementation
566 * specific and we probably need proper fix and carry perf_event_attr
567 * bitfield flags in separate data file FEAT_ section. Thought this seems
570 static void swap_bitfield(u8
*p
, unsigned len
)
574 for (i
= 0; i
< len
; i
++) {
580 /* exported for swapping attributes in file header */
581 void perf_event__attr_swap(struct perf_event_attr
*attr
)
583 attr
->type
= bswap_32(attr
->type
);
584 attr
->size
= bswap_32(attr
->size
);
586 #define bswap_safe(f, n) \
587 (attr->size > (offsetof(struct perf_event_attr, f) + \
588 sizeof(attr->f) * (n)))
589 #define bswap_field(f, sz) \
591 if (bswap_safe(f, 0)) \
592 attr->f = bswap_##sz(attr->f); \
594 #define bswap_field_32(f) bswap_field(f, 32)
595 #define bswap_field_64(f) bswap_field(f, 64)
597 bswap_field_64(config
);
598 bswap_field_64(sample_period
);
599 bswap_field_64(sample_type
);
600 bswap_field_64(read_format
);
601 bswap_field_32(wakeup_events
);
602 bswap_field_32(bp_type
);
603 bswap_field_64(bp_addr
);
604 bswap_field_64(bp_len
);
605 bswap_field_64(branch_sample_type
);
606 bswap_field_64(sample_regs_user
);
607 bswap_field_32(sample_stack_user
);
608 bswap_field_32(aux_watermark
);
611 * After read_format are bitfields. Check read_format because
612 * we are unable to use offsetof on bitfield.
614 if (bswap_safe(read_format
, 1))
615 swap_bitfield((u8
*) (&attr
->read_format
+ 1),
617 #undef bswap_field_64
618 #undef bswap_field_32
623 static void perf_event__hdr_attr_swap(union perf_event
*event
,
624 bool sample_id_all __maybe_unused
)
628 perf_event__attr_swap(&event
->attr
.attr
);
630 size
= event
->header
.size
;
631 size
-= (void *)&event
->attr
.id
- (void *)event
;
632 mem_bswap_64(event
->attr
.id
, size
);
635 static void perf_event__event_update_swap(union perf_event
*event
,
636 bool sample_id_all __maybe_unused
)
638 event
->event_update
.type
= bswap_64(event
->event_update
.type
);
639 event
->event_update
.id
= bswap_64(event
->event_update
.id
);
642 static void perf_event__event_type_swap(union perf_event
*event
,
643 bool sample_id_all __maybe_unused
)
645 event
->event_type
.event_type
.event_id
=
646 bswap_64(event
->event_type
.event_type
.event_id
);
649 static void perf_event__tracing_data_swap(union perf_event
*event
,
650 bool sample_id_all __maybe_unused
)
652 event
->tracing_data
.size
= bswap_32(event
->tracing_data
.size
);
655 static void perf_event__auxtrace_info_swap(union perf_event
*event
,
656 bool sample_id_all __maybe_unused
)
660 event
->auxtrace_info
.type
= bswap_32(event
->auxtrace_info
.type
);
662 size
= event
->header
.size
;
663 size
-= (void *)&event
->auxtrace_info
.priv
- (void *)event
;
664 mem_bswap_64(event
->auxtrace_info
.priv
, size
);
667 static void perf_event__auxtrace_swap(union perf_event
*event
,
668 bool sample_id_all __maybe_unused
)
670 event
->auxtrace
.size
= bswap_64(event
->auxtrace
.size
);
671 event
->auxtrace
.offset
= bswap_64(event
->auxtrace
.offset
);
672 event
->auxtrace
.reference
= bswap_64(event
->auxtrace
.reference
);
673 event
->auxtrace
.idx
= bswap_32(event
->auxtrace
.idx
);
674 event
->auxtrace
.tid
= bswap_32(event
->auxtrace
.tid
);
675 event
->auxtrace
.cpu
= bswap_32(event
->auxtrace
.cpu
);
678 static void perf_event__auxtrace_error_swap(union perf_event
*event
,
679 bool sample_id_all __maybe_unused
)
681 event
->auxtrace_error
.type
= bswap_32(event
->auxtrace_error
.type
);
682 event
->auxtrace_error
.code
= bswap_32(event
->auxtrace_error
.code
);
683 event
->auxtrace_error
.cpu
= bswap_32(event
->auxtrace_error
.cpu
);
684 event
->auxtrace_error
.pid
= bswap_32(event
->auxtrace_error
.pid
);
685 event
->auxtrace_error
.tid
= bswap_32(event
->auxtrace_error
.tid
);
686 event
->auxtrace_error
.ip
= bswap_64(event
->auxtrace_error
.ip
);
689 static void perf_event__thread_map_swap(union perf_event
*event
,
690 bool sample_id_all __maybe_unused
)
694 event
->thread_map
.nr
= bswap_64(event
->thread_map
.nr
);
696 for (i
= 0; i
< event
->thread_map
.nr
; i
++)
697 event
->thread_map
.entries
[i
].pid
= bswap_64(event
->thread_map
.entries
[i
].pid
);
700 static void perf_event__cpu_map_swap(union perf_event
*event
,
701 bool sample_id_all __maybe_unused
)
703 struct cpu_map_data
*data
= &event
->cpu_map
.data
;
704 struct cpu_map_entries
*cpus
;
705 struct cpu_map_mask
*mask
;
708 data
->type
= bswap_64(data
->type
);
710 switch (data
->type
) {
711 case PERF_CPU_MAP__CPUS
:
712 cpus
= (struct cpu_map_entries
*)data
->data
;
714 cpus
->nr
= bswap_16(cpus
->nr
);
716 for (i
= 0; i
< cpus
->nr
; i
++)
717 cpus
->cpu
[i
] = bswap_16(cpus
->cpu
[i
]);
719 case PERF_CPU_MAP__MASK
:
720 mask
= (struct cpu_map_mask
*) data
->data
;
722 mask
->nr
= bswap_16(mask
->nr
);
723 mask
->long_size
= bswap_16(mask
->long_size
);
725 switch (mask
->long_size
) {
726 case 4: mem_bswap_32(&mask
->mask
, mask
->nr
); break;
727 case 8: mem_bswap_64(&mask
->mask
, mask
->nr
); break;
729 pr_err("cpu_map swap: unsupported long size\n");
736 static void perf_event__stat_config_swap(union perf_event
*event
,
737 bool sample_id_all __maybe_unused
)
741 size
= event
->stat_config
.nr
* sizeof(event
->stat_config
.data
[0]);
742 size
+= 1; /* nr item itself */
743 mem_bswap_64(&event
->stat_config
.nr
, size
);
746 static void perf_event__stat_swap(union perf_event
*event
,
747 bool sample_id_all __maybe_unused
)
749 event
->stat
.id
= bswap_64(event
->stat
.id
);
750 event
->stat
.thread
= bswap_32(event
->stat
.thread
);
751 event
->stat
.cpu
= bswap_32(event
->stat
.cpu
);
752 event
->stat
.val
= bswap_64(event
->stat
.val
);
753 event
->stat
.ena
= bswap_64(event
->stat
.ena
);
754 event
->stat
.run
= bswap_64(event
->stat
.run
);
757 static void perf_event__stat_round_swap(union perf_event
*event
,
758 bool sample_id_all __maybe_unused
)
760 event
->stat_round
.type
= bswap_64(event
->stat_round
.type
);
761 event
->stat_round
.time
= bswap_64(event
->stat_round
.time
);
764 typedef void (*perf_event__swap_op
)(union perf_event
*event
,
767 static perf_event__swap_op perf_event__swap_ops
[] = {
768 [PERF_RECORD_MMAP
] = perf_event__mmap_swap
,
769 [PERF_RECORD_MMAP2
] = perf_event__mmap2_swap
,
770 [PERF_RECORD_COMM
] = perf_event__comm_swap
,
771 [PERF_RECORD_FORK
] = perf_event__task_swap
,
772 [PERF_RECORD_EXIT
] = perf_event__task_swap
,
773 [PERF_RECORD_LOST
] = perf_event__all64_swap
,
774 [PERF_RECORD_READ
] = perf_event__read_swap
,
775 [PERF_RECORD_THROTTLE
] = perf_event__throttle_swap
,
776 [PERF_RECORD_UNTHROTTLE
] = perf_event__throttle_swap
,
777 [PERF_RECORD_SAMPLE
] = perf_event__all64_swap
,
778 [PERF_RECORD_AUX
] = perf_event__aux_swap
,
779 [PERF_RECORD_ITRACE_START
] = perf_event__itrace_start_swap
,
780 [PERF_RECORD_LOST_SAMPLES
] = perf_event__all64_swap
,
781 [PERF_RECORD_SWITCH
] = perf_event__switch_swap
,
782 [PERF_RECORD_SWITCH_CPU_WIDE
] = perf_event__switch_swap
,
783 [PERF_RECORD_HEADER_ATTR
] = perf_event__hdr_attr_swap
,
784 [PERF_RECORD_HEADER_EVENT_TYPE
] = perf_event__event_type_swap
,
785 [PERF_RECORD_HEADER_TRACING_DATA
] = perf_event__tracing_data_swap
,
786 [PERF_RECORD_HEADER_BUILD_ID
] = NULL
,
787 [PERF_RECORD_ID_INDEX
] = perf_event__all64_swap
,
788 [PERF_RECORD_AUXTRACE_INFO
] = perf_event__auxtrace_info_swap
,
789 [PERF_RECORD_AUXTRACE
] = perf_event__auxtrace_swap
,
790 [PERF_RECORD_AUXTRACE_ERROR
] = perf_event__auxtrace_error_swap
,
791 [PERF_RECORD_THREAD_MAP
] = perf_event__thread_map_swap
,
792 [PERF_RECORD_CPU_MAP
] = perf_event__cpu_map_swap
,
793 [PERF_RECORD_STAT_CONFIG
] = perf_event__stat_config_swap
,
794 [PERF_RECORD_STAT
] = perf_event__stat_swap
,
795 [PERF_RECORD_STAT_ROUND
] = perf_event__stat_round_swap
,
796 [PERF_RECORD_EVENT_UPDATE
] = perf_event__event_update_swap
,
797 [PERF_RECORD_HEADER_MAX
] = NULL
,
801 * When perf record finishes a pass on every buffers, it records this pseudo
803 * We record the max timestamp t found in the pass n.
804 * Assuming these timestamps are monotonic across cpus, we know that if
805 * a buffer still has events with timestamps below t, they will be all
806 * available and then read in the pass n + 1.
807 * Hence when we start to read the pass n + 2, we can safely flush every
808 * events with timestamps below t.
810 * ============ PASS n =================
813 * cnt1 timestamps | cnt2 timestamps
816 * - | 4 <--- max recorded
818 * ============ PASS n + 1 ==============
821 * cnt1 timestamps | cnt2 timestamps
824 * 5 | 7 <---- max recorded
826 * Flush every events below timestamp 4
828 * ============ PASS n + 2 ==============
831 * cnt1 timestamps | cnt2 timestamps
836 * Flush every events below timestamp 7
839 static int process_finished_round(struct perf_tool
*tool __maybe_unused
,
840 union perf_event
*event __maybe_unused
,
841 struct ordered_events
*oe
)
844 fprintf(stdout
, "\n");
845 return ordered_events__flush(oe
, OE_FLUSH__ROUND
);
848 int perf_session__queue_event(struct perf_session
*s
, union perf_event
*event
,
849 struct perf_sample
*sample
, u64 file_offset
)
851 return ordered_events__queue(&s
->ordered_events
, event
, sample
, file_offset
);
854 static void callchain__lbr_callstack_printf(struct perf_sample
*sample
)
856 struct ip_callchain
*callchain
= sample
->callchain
;
857 struct branch_stack
*lbr_stack
= sample
->branch_stack
;
858 u64 kernel_callchain_nr
= callchain
->nr
;
861 for (i
= 0; i
< kernel_callchain_nr
; i
++) {
862 if (callchain
->ips
[i
] == PERF_CONTEXT_USER
)
866 if ((i
!= kernel_callchain_nr
) && lbr_stack
->nr
) {
869 * LBR callstack can only get user call chain,
870 * i is kernel call chain number,
871 * 1 is PERF_CONTEXT_USER.
873 * The user call chain is stored in LBR registers.
874 * LBR are pair registers. The caller is stored
875 * in "from" register, while the callee is stored
877 * For example, there is a call stack
878 * "A"->"B"->"C"->"D".
879 * The LBR registers will recorde like
880 * "C"->"D", "B"->"C", "A"->"B".
881 * So only the first "to" register and all "from"
882 * registers are needed to construct the whole stack.
884 total_nr
= i
+ 1 + lbr_stack
->nr
+ 1;
885 kernel_callchain_nr
= i
+ 1;
887 printf("... LBR call chain: nr:%" PRIu64
"\n", total_nr
);
889 for (i
= 0; i
< kernel_callchain_nr
; i
++)
890 printf("..... %2d: %016" PRIx64
"\n",
891 i
, callchain
->ips
[i
]);
893 printf("..... %2d: %016" PRIx64
"\n",
894 (int)(kernel_callchain_nr
), lbr_stack
->entries
[0].to
);
895 for (i
= 0; i
< lbr_stack
->nr
; i
++)
896 printf("..... %2d: %016" PRIx64
"\n",
897 (int)(i
+ kernel_callchain_nr
+ 1), lbr_stack
->entries
[i
].from
);
901 static void callchain__printf(struct perf_evsel
*evsel
,
902 struct perf_sample
*sample
)
905 struct ip_callchain
*callchain
= sample
->callchain
;
907 if (has_branch_callstack(evsel
))
908 callchain__lbr_callstack_printf(sample
);
910 printf("... FP chain: nr:%" PRIu64
"\n", callchain
->nr
);
912 for (i
= 0; i
< callchain
->nr
; i
++)
913 printf("..... %2d: %016" PRIx64
"\n",
914 i
, callchain
->ips
[i
]);
917 static void branch_stack__printf(struct perf_sample
*sample
)
921 printf("... branch stack: nr:%" PRIu64
"\n", sample
->branch_stack
->nr
);
923 for (i
= 0; i
< sample
->branch_stack
->nr
; i
++) {
924 struct branch_entry
*e
= &sample
->branch_stack
->entries
[i
];
926 printf("..... %2"PRIu64
": %016" PRIx64
" -> %016" PRIx64
" %hu cycles %s%s%s%s %x\n",
929 e
->flags
.mispred
? "M" : " ",
930 e
->flags
.predicted
? "P" : " ",
931 e
->flags
.abort
? "A" : " ",
932 e
->flags
.in_tx
? "T" : " ",
933 (unsigned)e
->flags
.reserved
);
937 static void regs_dump__printf(u64 mask
, u64
*regs
)
941 for_each_set_bit(rid
, (unsigned long *) &mask
, sizeof(mask
) * 8) {
944 printf(".... %-5s 0x%" PRIx64
"\n",
945 perf_reg_name(rid
), val
);
949 static const char *regs_abi
[] = {
950 [PERF_SAMPLE_REGS_ABI_NONE
] = "none",
951 [PERF_SAMPLE_REGS_ABI_32
] = "32-bit",
952 [PERF_SAMPLE_REGS_ABI_64
] = "64-bit",
955 static inline const char *regs_dump_abi(struct regs_dump
*d
)
957 if (d
->abi
> PERF_SAMPLE_REGS_ABI_64
)
960 return regs_abi
[d
->abi
];
963 static void regs__printf(const char *type
, struct regs_dump
*regs
)
965 u64 mask
= regs
->mask
;
967 printf("... %s regs: mask 0x%" PRIx64
" ABI %s\n",
970 regs_dump_abi(regs
));
972 regs_dump__printf(mask
, regs
->regs
);
975 static void regs_user__printf(struct perf_sample
*sample
)
977 struct regs_dump
*user_regs
= &sample
->user_regs
;
980 regs__printf("user", user_regs
);
983 static void regs_intr__printf(struct perf_sample
*sample
)
985 struct regs_dump
*intr_regs
= &sample
->intr_regs
;
988 regs__printf("intr", intr_regs
);
991 static void stack_user__printf(struct stack_dump
*dump
)
993 printf("... ustack: size %" PRIu64
", offset 0x%x\n",
994 dump
->size
, dump
->offset
);
997 static void perf_evlist__print_tstamp(struct perf_evlist
*evlist
,
998 union perf_event
*event
,
999 struct perf_sample
*sample
)
1001 u64 sample_type
= __perf_evlist__combined_sample_type(evlist
);
1003 if (event
->header
.type
!= PERF_RECORD_SAMPLE
&&
1004 !perf_evlist__sample_id_all(evlist
)) {
1005 fputs("-1 -1 ", stdout
);
1009 if ((sample_type
& PERF_SAMPLE_CPU
))
1010 printf("%u ", sample
->cpu
);
1012 if (sample_type
& PERF_SAMPLE_TIME
)
1013 printf("%" PRIu64
" ", sample
->time
);
1016 static void sample_read__printf(struct perf_sample
*sample
, u64 read_format
)
1018 printf("... sample_read:\n");
1020 if (read_format
& PERF_FORMAT_TOTAL_TIME_ENABLED
)
1021 printf("...... time enabled %016" PRIx64
"\n",
1022 sample
->read
.time_enabled
);
1024 if (read_format
& PERF_FORMAT_TOTAL_TIME_RUNNING
)
1025 printf("...... time running %016" PRIx64
"\n",
1026 sample
->read
.time_running
);
1028 if (read_format
& PERF_FORMAT_GROUP
) {
1031 printf(".... group nr %" PRIu64
"\n", sample
->read
.group
.nr
);
1033 for (i
= 0; i
< sample
->read
.group
.nr
; i
++) {
1034 struct sample_read_value
*value
;
1036 value
= &sample
->read
.group
.values
[i
];
1037 printf("..... id %016" PRIx64
1038 ", value %016" PRIx64
"\n",
1039 value
->id
, value
->value
);
1042 printf("..... id %016" PRIx64
", value %016" PRIx64
"\n",
1043 sample
->read
.one
.id
, sample
->read
.one
.value
);
1046 static void dump_event(struct perf_evlist
*evlist
, union perf_event
*event
,
1047 u64 file_offset
, struct perf_sample
*sample
)
1052 printf("\n%#" PRIx64
" [%#x]: event: %d\n",
1053 file_offset
, event
->header
.size
, event
->header
.type
);
1058 perf_evlist__print_tstamp(evlist
, event
, sample
);
1060 printf("%#" PRIx64
" [%#x]: PERF_RECORD_%s", file_offset
,
1061 event
->header
.size
, perf_event__name(event
->header
.type
));
1064 static void dump_sample(struct perf_evsel
*evsel
, union perf_event
*event
,
1065 struct perf_sample
*sample
)
1072 printf("(IP, 0x%x): %d/%d: %#" PRIx64
" period: %" PRIu64
" addr: %#" PRIx64
"\n",
1073 event
->header
.misc
, sample
->pid
, sample
->tid
, sample
->ip
,
1074 sample
->period
, sample
->addr
);
1076 sample_type
= evsel
->attr
.sample_type
;
1078 if (sample_type
& PERF_SAMPLE_CALLCHAIN
)
1079 callchain__printf(evsel
, sample
);
1081 if ((sample_type
& PERF_SAMPLE_BRANCH_STACK
) && !has_branch_callstack(evsel
))
1082 branch_stack__printf(sample
);
1084 if (sample_type
& PERF_SAMPLE_REGS_USER
)
1085 regs_user__printf(sample
);
1087 if (sample_type
& PERF_SAMPLE_REGS_INTR
)
1088 regs_intr__printf(sample
);
1090 if (sample_type
& PERF_SAMPLE_STACK_USER
)
1091 stack_user__printf(&sample
->user_stack
);
1093 if (sample_type
& PERF_SAMPLE_WEIGHT
)
1094 printf("... weight: %" PRIu64
"\n", sample
->weight
);
1096 if (sample_type
& PERF_SAMPLE_DATA_SRC
)
1097 printf(" . data_src: 0x%"PRIx64
"\n", sample
->data_src
);
1099 if (sample_type
& PERF_SAMPLE_TRANSACTION
)
1100 printf("... transaction: %" PRIx64
"\n", sample
->transaction
);
1102 if (sample_type
& PERF_SAMPLE_READ
)
1103 sample_read__printf(sample
, evsel
->attr
.read_format
);
1106 static struct machine
*machines__find_for_cpumode(struct machines
*machines
,
1107 union perf_event
*event
,
1108 struct perf_sample
*sample
)
1110 struct machine
*machine
;
1113 ((sample
->cpumode
== PERF_RECORD_MISC_GUEST_KERNEL
) ||
1114 (sample
->cpumode
== PERF_RECORD_MISC_GUEST_USER
))) {
1117 if (event
->header
.type
== PERF_RECORD_MMAP
1118 || event
->header
.type
== PERF_RECORD_MMAP2
)
1119 pid
= event
->mmap
.pid
;
1123 machine
= machines__find(machines
, pid
);
1125 machine
= machines__findnew(machines
, DEFAULT_GUEST_KERNEL_ID
);
1129 return &machines
->host
;
1132 static int deliver_sample_value(struct perf_evlist
*evlist
,
1133 struct perf_tool
*tool
,
1134 union perf_event
*event
,
1135 struct perf_sample
*sample
,
1136 struct sample_read_value
*v
,
1137 struct machine
*machine
)
1139 struct perf_sample_id
*sid
= perf_evlist__id2sid(evlist
, v
->id
);
1143 sample
->period
= v
->value
- sid
->period
;
1144 sid
->period
= v
->value
;
1147 if (!sid
|| sid
->evsel
== NULL
) {
1148 ++evlist
->stats
.nr_unknown_id
;
1152 return tool
->sample(tool
, event
, sample
, sid
->evsel
, machine
);
1155 static int deliver_sample_group(struct perf_evlist
*evlist
,
1156 struct perf_tool
*tool
,
1157 union perf_event
*event
,
1158 struct perf_sample
*sample
,
1159 struct machine
*machine
)
1164 for (i
= 0; i
< sample
->read
.group
.nr
; i
++) {
1165 ret
= deliver_sample_value(evlist
, tool
, event
, sample
,
1166 &sample
->read
.group
.values
[i
],
1176 perf_evlist__deliver_sample(struct perf_evlist
*evlist
,
1177 struct perf_tool
*tool
,
1178 union perf_event
*event
,
1179 struct perf_sample
*sample
,
1180 struct perf_evsel
*evsel
,
1181 struct machine
*machine
)
1183 /* We know evsel != NULL. */
1184 u64 sample_type
= evsel
->attr
.sample_type
;
1185 u64 read_format
= evsel
->attr
.read_format
;
1187 /* Standard sample delievery. */
1188 if (!(sample_type
& PERF_SAMPLE_READ
))
1189 return tool
->sample(tool
, event
, sample
, evsel
, machine
);
1191 /* For PERF_SAMPLE_READ we have either single or group mode. */
1192 if (read_format
& PERF_FORMAT_GROUP
)
1193 return deliver_sample_group(evlist
, tool
, event
, sample
,
1196 return deliver_sample_value(evlist
, tool
, event
, sample
,
1197 &sample
->read
.one
, machine
);
1200 static int machines__deliver_event(struct machines
*machines
,
1201 struct perf_evlist
*evlist
,
1202 union perf_event
*event
,
1203 struct perf_sample
*sample
,
1204 struct perf_tool
*tool
, u64 file_offset
)
1206 struct perf_evsel
*evsel
;
1207 struct machine
*machine
;
1209 dump_event(evlist
, event
, file_offset
, sample
);
1211 evsel
= perf_evlist__id2evsel(evlist
, sample
->id
);
1213 machine
= machines__find_for_cpumode(machines
, event
, sample
);
1215 switch (event
->header
.type
) {
1216 case PERF_RECORD_SAMPLE
:
1217 if (evsel
== NULL
) {
1218 ++evlist
->stats
.nr_unknown_id
;
1221 dump_sample(evsel
, event
, sample
);
1222 if (machine
== NULL
) {
1223 ++evlist
->stats
.nr_unprocessable_samples
;
1226 return perf_evlist__deliver_sample(evlist
, tool
, event
, sample
, evsel
, machine
);
1227 case PERF_RECORD_MMAP
:
1228 return tool
->mmap(tool
, event
, sample
, machine
);
1229 case PERF_RECORD_MMAP2
:
1230 if (event
->header
.misc
& PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT
)
1231 ++evlist
->stats
.nr_proc_map_timeout
;
1232 return tool
->mmap2(tool
, event
, sample
, machine
);
1233 case PERF_RECORD_COMM
:
1234 return tool
->comm(tool
, event
, sample
, machine
);
1235 case PERF_RECORD_FORK
:
1236 return tool
->fork(tool
, event
, sample
, machine
);
1237 case PERF_RECORD_EXIT
:
1238 return tool
->exit(tool
, event
, sample
, machine
);
1239 case PERF_RECORD_LOST
:
1240 if (tool
->lost
== perf_event__process_lost
)
1241 evlist
->stats
.total_lost
+= event
->lost
.lost
;
1242 return tool
->lost(tool
, event
, sample
, machine
);
1243 case PERF_RECORD_LOST_SAMPLES
:
1244 if (tool
->lost_samples
== perf_event__process_lost_samples
)
1245 evlist
->stats
.total_lost_samples
+= event
->lost_samples
.lost
;
1246 return tool
->lost_samples(tool
, event
, sample
, machine
);
1247 case PERF_RECORD_READ
:
1248 return tool
->read(tool
, event
, sample
, evsel
, machine
);
1249 case PERF_RECORD_THROTTLE
:
1250 return tool
->throttle(tool
, event
, sample
, machine
);
1251 case PERF_RECORD_UNTHROTTLE
:
1252 return tool
->unthrottle(tool
, event
, sample
, machine
);
1253 case PERF_RECORD_AUX
:
1254 if (tool
->aux
== perf_event__process_aux
&&
1255 (event
->aux
.flags
& PERF_AUX_FLAG_TRUNCATED
))
1256 evlist
->stats
.total_aux_lost
+= 1;
1257 return tool
->aux(tool
, event
, sample
, machine
);
1258 case PERF_RECORD_ITRACE_START
:
1259 return tool
->itrace_start(tool
, event
, sample
, machine
);
1260 case PERF_RECORD_SWITCH
:
1261 case PERF_RECORD_SWITCH_CPU_WIDE
:
1262 return tool
->context_switch(tool
, event
, sample
, machine
);
1264 ++evlist
->stats
.nr_unknown_events
;
1269 static int perf_session__deliver_event(struct perf_session
*session
,
1270 union perf_event
*event
,
1271 struct perf_sample
*sample
,
1272 struct perf_tool
*tool
,
1277 ret
= auxtrace__process_event(session
, event
, sample
, tool
);
1283 return machines__deliver_event(&session
->machines
, session
->evlist
,
1284 event
, sample
, tool
, file_offset
);
1287 static s64
perf_session__process_user_event(struct perf_session
*session
,
1288 union perf_event
*event
,
1291 struct ordered_events
*oe
= &session
->ordered_events
;
1292 struct perf_tool
*tool
= session
->tool
;
1293 int fd
= perf_data_file__fd(session
->file
);
1296 dump_event(session
->evlist
, event
, file_offset
, NULL
);
1298 /* These events are processed right away */
1299 switch (event
->header
.type
) {
1300 case PERF_RECORD_HEADER_ATTR
:
1301 err
= tool
->attr(tool
, event
, &session
->evlist
);
1303 perf_session__set_id_hdr_size(session
);
1304 perf_session__set_comm_exec(session
);
1307 case PERF_RECORD_EVENT_UPDATE
:
1308 return tool
->event_update(tool
, event
, &session
->evlist
);
1309 case PERF_RECORD_HEADER_EVENT_TYPE
:
1311 * Depreceated, but we need to handle it for sake
1312 * of old data files create in pipe mode.
1315 case PERF_RECORD_HEADER_TRACING_DATA
:
1316 /* setup for reading amidst mmap */
1317 lseek(fd
, file_offset
, SEEK_SET
);
1318 return tool
->tracing_data(tool
, event
, session
);
1319 case PERF_RECORD_HEADER_BUILD_ID
:
1320 return tool
->build_id(tool
, event
, session
);
1321 case PERF_RECORD_FINISHED_ROUND
:
1322 return tool
->finished_round(tool
, event
, oe
);
1323 case PERF_RECORD_ID_INDEX
:
1324 return tool
->id_index(tool
, event
, session
);
1325 case PERF_RECORD_AUXTRACE_INFO
:
1326 return tool
->auxtrace_info(tool
, event
, session
);
1327 case PERF_RECORD_AUXTRACE
:
1328 /* setup for reading amidst mmap */
1329 lseek(fd
, file_offset
+ event
->header
.size
, SEEK_SET
);
1330 return tool
->auxtrace(tool
, event
, session
);
1331 case PERF_RECORD_AUXTRACE_ERROR
:
1332 perf_session__auxtrace_error_inc(session
, event
);
1333 return tool
->auxtrace_error(tool
, event
, session
);
1334 case PERF_RECORD_THREAD_MAP
:
1335 return tool
->thread_map(tool
, event
, session
);
1336 case PERF_RECORD_CPU_MAP
:
1337 return tool
->cpu_map(tool
, event
, session
);
1338 case PERF_RECORD_STAT_CONFIG
:
1339 return tool
->stat_config(tool
, event
, session
);
1340 case PERF_RECORD_STAT
:
1341 return tool
->stat(tool
, event
, session
);
1342 case PERF_RECORD_STAT_ROUND
:
1343 return tool
->stat_round(tool
, event
, session
);
1349 int perf_session__deliver_synth_event(struct perf_session
*session
,
1350 union perf_event
*event
,
1351 struct perf_sample
*sample
)
1353 struct perf_evlist
*evlist
= session
->evlist
;
1354 struct perf_tool
*tool
= session
->tool
;
1356 events_stats__inc(&evlist
->stats
, event
->header
.type
);
1358 if (event
->header
.type
>= PERF_RECORD_USER_TYPE_START
)
1359 return perf_session__process_user_event(session
, event
, 0);
1361 return machines__deliver_event(&session
->machines
, evlist
, event
, sample
, tool
, 0);
1364 static void event_swap(union perf_event
*event
, bool sample_id_all
)
1366 perf_event__swap_op swap
;
1368 swap
= perf_event__swap_ops
[event
->header
.type
];
1370 swap(event
, sample_id_all
);
1373 int perf_session__peek_event(struct perf_session
*session
, off_t file_offset
,
1374 void *buf
, size_t buf_sz
,
1375 union perf_event
**event_ptr
,
1376 struct perf_sample
*sample
)
1378 union perf_event
*event
;
1379 size_t hdr_sz
, rest
;
1382 if (session
->one_mmap
&& !session
->header
.needs_swap
) {
1383 event
= file_offset
- session
->one_mmap_offset
+
1384 session
->one_mmap_addr
;
1385 goto out_parse_sample
;
1388 if (perf_data_file__is_pipe(session
->file
))
1391 fd
= perf_data_file__fd(session
->file
);
1392 hdr_sz
= sizeof(struct perf_event_header
);
1394 if (buf_sz
< hdr_sz
)
1397 if (lseek(fd
, file_offset
, SEEK_SET
) == (off_t
)-1 ||
1398 readn(fd
, buf
, hdr_sz
) != (ssize_t
)hdr_sz
)
1401 event
= (union perf_event
*)buf
;
1403 if (session
->header
.needs_swap
)
1404 perf_event_header__bswap(&event
->header
);
1406 if (event
->header
.size
< hdr_sz
|| event
->header
.size
> buf_sz
)
1409 rest
= event
->header
.size
- hdr_sz
;
1411 if (readn(fd
, buf
, rest
) != (ssize_t
)rest
)
1414 if (session
->header
.needs_swap
)
1415 event_swap(event
, perf_evlist__sample_id_all(session
->evlist
));
1419 if (sample
&& event
->header
.type
< PERF_RECORD_USER_TYPE_START
&&
1420 perf_evlist__parse_sample(session
->evlist
, event
, sample
))
1428 static s64
perf_session__process_event(struct perf_session
*session
,
1429 union perf_event
*event
, u64 file_offset
)
1431 struct perf_evlist
*evlist
= session
->evlist
;
1432 struct perf_tool
*tool
= session
->tool
;
1433 struct perf_sample sample
;
1436 if (session
->header
.needs_swap
)
1437 event_swap(event
, perf_evlist__sample_id_all(evlist
));
1439 if (event
->header
.type
>= PERF_RECORD_HEADER_MAX
)
1442 events_stats__inc(&evlist
->stats
, event
->header
.type
);
1444 if (event
->header
.type
>= PERF_RECORD_USER_TYPE_START
)
1445 return perf_session__process_user_event(session
, event
, file_offset
);
1448 * For all kernel events we get the sample data
1450 ret
= perf_evlist__parse_sample(evlist
, event
, &sample
);
1454 if (tool
->ordered_events
) {
1455 ret
= perf_session__queue_event(session
, event
, &sample
, file_offset
);
1460 return perf_session__deliver_event(session
, event
, &sample
, tool
,
1464 void perf_event_header__bswap(struct perf_event_header
*hdr
)
1466 hdr
->type
= bswap_32(hdr
->type
);
1467 hdr
->misc
= bswap_16(hdr
->misc
);
1468 hdr
->size
= bswap_16(hdr
->size
);
1471 struct thread
*perf_session__findnew(struct perf_session
*session
, pid_t pid
)
1473 return machine__findnew_thread(&session
->machines
.host
, -1, pid
);
1476 int perf_session__register_idle_thread(struct perf_session
*session
)
1478 struct thread
*thread
;
1481 thread
= machine__findnew_thread(&session
->machines
.host
, 0, 0);
1482 if (thread
== NULL
|| thread__set_comm(thread
, "swapper", 0)) {
1483 pr_err("problem inserting idle task.\n");
1487 /* machine__findnew_thread() got the thread, so put it */
1488 thread__put(thread
);
1492 static void perf_session__warn_about_errors(const struct perf_session
*session
)
1494 const struct events_stats
*stats
= &session
->evlist
->stats
;
1495 const struct ordered_events
*oe
= &session
->ordered_events
;
1497 if (session
->tool
->lost
== perf_event__process_lost
&&
1498 stats
->nr_events
[PERF_RECORD_LOST
] != 0) {
1499 ui__warning("Processed %d events and lost %d chunks!\n\n"
1500 "Check IO/CPU overload!\n\n",
1501 stats
->nr_events
[0],
1502 stats
->nr_events
[PERF_RECORD_LOST
]);
1505 if (session
->tool
->lost_samples
== perf_event__process_lost_samples
) {
1508 drop_rate
= (double)stats
->total_lost_samples
/
1509 (double) (stats
->nr_events
[PERF_RECORD_SAMPLE
] + stats
->total_lost_samples
);
1510 if (drop_rate
> 0.05) {
1511 ui__warning("Processed %" PRIu64
" samples and lost %3.2f%% samples!\n\n",
1512 stats
->nr_events
[PERF_RECORD_SAMPLE
] + stats
->total_lost_samples
,
1517 if (session
->tool
->aux
== perf_event__process_aux
&&
1518 stats
->total_aux_lost
!= 0) {
1519 ui__warning("AUX data lost %" PRIu64
" times out of %u!\n\n",
1520 stats
->total_aux_lost
,
1521 stats
->nr_events
[PERF_RECORD_AUX
]);
1524 if (stats
->nr_unknown_events
!= 0) {
1525 ui__warning("Found %u unknown events!\n\n"
1526 "Is this an older tool processing a perf.data "
1527 "file generated by a more recent tool?\n\n"
1528 "If that is not the case, consider "
1529 "reporting to linux-kernel@vger.kernel.org.\n\n",
1530 stats
->nr_unknown_events
);
1533 if (stats
->nr_unknown_id
!= 0) {
1534 ui__warning("%u samples with id not present in the header\n",
1535 stats
->nr_unknown_id
);
1538 if (stats
->nr_invalid_chains
!= 0) {
1539 ui__warning("Found invalid callchains!\n\n"
1540 "%u out of %u events were discarded for this reason.\n\n"
1541 "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1542 stats
->nr_invalid_chains
,
1543 stats
->nr_events
[PERF_RECORD_SAMPLE
]);
1546 if (stats
->nr_unprocessable_samples
!= 0) {
1547 ui__warning("%u unprocessable samples recorded.\n"
1548 "Do you have a KVM guest running and not using 'perf kvm'?\n",
1549 stats
->nr_unprocessable_samples
);
1552 if (oe
->nr_unordered_events
!= 0)
1553 ui__warning("%u out of order events recorded.\n", oe
->nr_unordered_events
);
1555 events_stats__auxtrace_error_warn(stats
);
1557 if (stats
->nr_proc_map_timeout
!= 0) {
1558 ui__warning("%d map information files for pre-existing threads were\n"
1559 "not processed, if there are samples for addresses they\n"
1560 "will not be resolved, you may find out which are these\n"
1561 "threads by running with -v and redirecting the output\n"
1563 "The time limit to process proc map is too short?\n"
1564 "Increase it by --proc-map-timeout\n",
1565 stats
->nr_proc_map_timeout
);
1569 static int perf_session__flush_thread_stack(struct thread
*thread
,
1570 void *p __maybe_unused
)
1572 return thread_stack__flush(thread
);
1575 static int perf_session__flush_thread_stacks(struct perf_session
*session
)
1577 return machines__for_each_thread(&session
->machines
,
1578 perf_session__flush_thread_stack
,
1582 volatile int session_done
;
1584 static int __perf_session__process_pipe_events(struct perf_session
*session
)
1586 struct ordered_events
*oe
= &session
->ordered_events
;
1587 struct perf_tool
*tool
= session
->tool
;
1588 int fd
= perf_data_file__fd(session
->file
);
1589 union perf_event
*event
;
1590 uint32_t size
, cur_size
= 0;
1597 perf_tool__fill_defaults(tool
);
1600 cur_size
= sizeof(union perf_event
);
1602 buf
= malloc(cur_size
);
1607 err
= readn(fd
, event
, sizeof(struct perf_event_header
));
1612 pr_err("failed to read event header\n");
1616 if (session
->header
.needs_swap
)
1617 perf_event_header__bswap(&event
->header
);
1619 size
= event
->header
.size
;
1620 if (size
< sizeof(struct perf_event_header
)) {
1621 pr_err("bad event header size\n");
1625 if (size
> cur_size
) {
1626 void *new = realloc(buf
, size
);
1628 pr_err("failed to allocate memory to read event\n");
1636 p
+= sizeof(struct perf_event_header
);
1638 if (size
- sizeof(struct perf_event_header
)) {
1639 err
= readn(fd
, p
, size
- sizeof(struct perf_event_header
));
1642 pr_err("unexpected end of event stream\n");
1646 pr_err("failed to read event data\n");
1651 if ((skip
= perf_session__process_event(session
, event
, head
)) < 0) {
1652 pr_err("%#" PRIx64
" [%#x]: failed to process type: %d\n",
1653 head
, event
->header
.size
, event
->header
.type
);
1663 if (!session_done())
1666 /* do the final flush for ordered samples */
1667 err
= ordered_events__flush(oe
, OE_FLUSH__FINAL
);
1670 err
= auxtrace__flush_events(session
, tool
);
1673 err
= perf_session__flush_thread_stacks(session
);
1676 perf_session__warn_about_errors(session
);
1677 ordered_events__free(&session
->ordered_events
);
1678 auxtrace__free_events(session
);
1682 static union perf_event
*
1683 fetch_mmaped_event(struct perf_session
*session
,
1684 u64 head
, size_t mmap_size
, char *buf
)
1686 union perf_event
*event
;
1689 * Ensure we have enough space remaining to read
1690 * the size of the event in the headers.
1692 if (head
+ sizeof(event
->header
) > mmap_size
)
1695 event
= (union perf_event
*)(buf
+ head
);
1697 if (session
->header
.needs_swap
)
1698 perf_event_header__bswap(&event
->header
);
1700 if (head
+ event
->header
.size
> mmap_size
) {
1701 /* We're not fetching the event so swap back again */
1702 if (session
->header
.needs_swap
)
1703 perf_event_header__bswap(&event
->header
);
1711 * On 64bit we can mmap the data file in one go. No need for tiny mmap
1712 * slices. On 32bit we use 32MB.
1714 #if BITS_PER_LONG == 64
1715 #define MMAP_SIZE ULLONG_MAX
1718 #define MMAP_SIZE (32 * 1024 * 1024ULL)
1719 #define NUM_MMAPS 128
1722 static int __perf_session__process_events(struct perf_session
*session
,
1723 u64 data_offset
, u64 data_size
,
1726 struct ordered_events
*oe
= &session
->ordered_events
;
1727 struct perf_tool
*tool
= session
->tool
;
1728 int fd
= perf_data_file__fd(session
->file
);
1729 u64 head
, page_offset
, file_offset
, file_pos
, size
;
1730 int err
, mmap_prot
, mmap_flags
, map_idx
= 0;
1732 char *buf
, *mmaps
[NUM_MMAPS
];
1733 union perf_event
*event
;
1734 struct ui_progress prog
;
1737 perf_tool__fill_defaults(tool
);
1739 page_offset
= page_size
* (data_offset
/ page_size
);
1740 file_offset
= page_offset
;
1741 head
= data_offset
- page_offset
;
1746 if (data_offset
+ data_size
< file_size
)
1747 file_size
= data_offset
+ data_size
;
1749 ui_progress__init(&prog
, file_size
, "Processing events...");
1751 mmap_size
= MMAP_SIZE
;
1752 if (mmap_size
> file_size
) {
1753 mmap_size
= file_size
;
1754 session
->one_mmap
= true;
1757 memset(mmaps
, 0, sizeof(mmaps
));
1759 mmap_prot
= PROT_READ
;
1760 mmap_flags
= MAP_SHARED
;
1762 if (session
->header
.needs_swap
) {
1763 mmap_prot
|= PROT_WRITE
;
1764 mmap_flags
= MAP_PRIVATE
;
1767 buf
= mmap(NULL
, mmap_size
, mmap_prot
, mmap_flags
, fd
,
1769 if (buf
== MAP_FAILED
) {
1770 pr_err("failed to mmap file\n");
1774 mmaps
[map_idx
] = buf
;
1775 map_idx
= (map_idx
+ 1) & (ARRAY_SIZE(mmaps
) - 1);
1776 file_pos
= file_offset
+ head
;
1777 if (session
->one_mmap
) {
1778 session
->one_mmap_addr
= buf
;
1779 session
->one_mmap_offset
= file_offset
;
1783 event
= fetch_mmaped_event(session
, head
, mmap_size
, buf
);
1785 if (mmaps
[map_idx
]) {
1786 munmap(mmaps
[map_idx
], mmap_size
);
1787 mmaps
[map_idx
] = NULL
;
1790 page_offset
= page_size
* (head
/ page_size
);
1791 file_offset
+= page_offset
;
1792 head
-= page_offset
;
1796 size
= event
->header
.size
;
1798 if (size
< sizeof(struct perf_event_header
) ||
1799 (skip
= perf_session__process_event(session
, event
, file_pos
)) < 0) {
1800 pr_err("%#" PRIx64
" [%#x]: failed to process type: %d\n",
1801 file_offset
+ head
, event
->header
.size
,
1802 event
->header
.type
);
1813 ui_progress__update(&prog
, size
);
1818 if (file_pos
< file_size
)
1822 /* do the final flush for ordered samples */
1823 err
= ordered_events__flush(oe
, OE_FLUSH__FINAL
);
1826 err
= auxtrace__flush_events(session
, tool
);
1829 err
= perf_session__flush_thread_stacks(session
);
1831 ui_progress__finish();
1832 perf_session__warn_about_errors(session
);
1833 ordered_events__free(&session
->ordered_events
);
1834 auxtrace__free_events(session
);
1835 session
->one_mmap
= false;
1839 int perf_session__process_events(struct perf_session
*session
)
1841 u64 size
= perf_data_file__size(session
->file
);
1844 if (perf_session__register_idle_thread(session
) < 0)
1847 if (!perf_data_file__is_pipe(session
->file
))
1848 err
= __perf_session__process_events(session
,
1849 session
->header
.data_offset
,
1850 session
->header
.data_size
, size
);
1852 err
= __perf_session__process_pipe_events(session
);
1857 bool perf_session__has_traces(struct perf_session
*session
, const char *msg
)
1859 struct perf_evsel
*evsel
;
1861 evlist__for_each(session
->evlist
, evsel
) {
1862 if (evsel
->attr
.type
== PERF_TYPE_TRACEPOINT
)
1866 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg
);
1870 int maps__set_kallsyms_ref_reloc_sym(struct map
**maps
,
1871 const char *symbol_name
, u64 addr
)
1875 struct ref_reloc_sym
*ref
;
1877 ref
= zalloc(sizeof(struct ref_reloc_sym
));
1881 ref
->name
= strdup(symbol_name
);
1882 if (ref
->name
== NULL
) {
1887 bracket
= strchr(ref
->name
, ']');
1893 for (i
= 0; i
< MAP__NR_TYPES
; ++i
) {
1894 struct kmap
*kmap
= map__kmap(maps
[i
]);
1898 kmap
->ref_reloc_sym
= ref
;
1904 size_t perf_session__fprintf_dsos(struct perf_session
*session
, FILE *fp
)
1906 return machines__fprintf_dsos(&session
->machines
, fp
);
1909 size_t perf_session__fprintf_dsos_buildid(struct perf_session
*session
, FILE *fp
,
1910 bool (skip
)(struct dso
*dso
, int parm
), int parm
)
1912 return machines__fprintf_dsos_buildid(&session
->machines
, fp
, skip
, parm
);
1915 size_t perf_session__fprintf_nr_events(struct perf_session
*session
, FILE *fp
)
1918 const char *msg
= "";
1920 if (perf_header__has_feat(&session
->header
, HEADER_AUXTRACE
))
1921 msg
= " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
1923 ret
= fprintf(fp
, "\nAggregated stats:%s\n", msg
);
1925 ret
+= events_stats__fprintf(&session
->evlist
->stats
, fp
);
1929 size_t perf_session__fprintf(struct perf_session
*session
, FILE *fp
)
1932 * FIXME: Here we have to actually print all the machines in this
1933 * session, not just the host...
1935 return machine__fprintf(&session
->machines
.host
, fp
);
1938 struct perf_evsel
*perf_session__find_first_evtype(struct perf_session
*session
,
1941 struct perf_evsel
*pos
;
1943 evlist__for_each(session
->evlist
, pos
) {
1944 if (pos
->attr
.type
== type
)
1950 void perf_evsel__print_ip(struct perf_evsel
*evsel
, struct perf_sample
*sample
,
1951 struct addr_location
*al
,
1952 unsigned int print_opts
, unsigned int stack_depth
)
1954 struct callchain_cursor_node
*node
;
1955 int print_ip
= print_opts
& PRINT_IP_OPT_IP
;
1956 int print_sym
= print_opts
& PRINT_IP_OPT_SYM
;
1957 int print_dso
= print_opts
& PRINT_IP_OPT_DSO
;
1958 int print_symoffset
= print_opts
& PRINT_IP_OPT_SYMOFFSET
;
1959 int print_oneline
= print_opts
& PRINT_IP_OPT_ONELINE
;
1960 int print_srcline
= print_opts
& PRINT_IP_OPT_SRCLINE
;
1961 char s
= print_oneline
? ' ' : '\t';
1963 if (symbol_conf
.use_callchain
&& sample
->callchain
) {
1964 struct addr_location node_al
;
1966 if (thread__resolve_callchain(al
->thread
, evsel
,
1968 stack_depth
) != 0) {
1970 error("Failed to resolve callchain. Skipping\n");
1973 callchain_cursor_commit(&callchain_cursor
);
1975 if (print_symoffset
)
1978 while (stack_depth
) {
1981 node
= callchain_cursor_current(&callchain_cursor
);
1985 if (node
->sym
&& node
->sym
->ignore
)
1989 printf("%c%16" PRIx64
, s
, node
->ip
);
1992 addr
= node
->map
->map_ip(node
->map
, node
->ip
);
1996 if (print_symoffset
) {
1997 node_al
.addr
= addr
;
1998 node_al
.map
= node
->map
;
1999 symbol__fprintf_symname_offs(node
->sym
, &node_al
, stdout
);
2001 symbol__fprintf_symname(node
->sym
, stdout
);
2006 map__fprintf_dsoname(node
->map
, stdout
);
2011 map__fprintf_srcline(node
->map
, addr
, "\n ",
2019 callchain_cursor_advance(&callchain_cursor
);
2023 if (al
->sym
&& al
->sym
->ignore
)
2027 printf("%16" PRIx64
, sample
->ip
);
2031 if (print_symoffset
)
2032 symbol__fprintf_symname_offs(al
->sym
, al
,
2035 symbol__fprintf_symname(al
->sym
, stdout
);
2040 map__fprintf_dsoname(al
->map
, stdout
);
2045 map__fprintf_srcline(al
->map
, al
->addr
, "\n ", stdout
);
2049 int perf_session__cpu_bitmap(struct perf_session
*session
,
2050 const char *cpu_list
, unsigned long *cpu_bitmap
)
2053 struct cpu_map
*map
;
2055 for (i
= 0; i
< PERF_TYPE_MAX
; ++i
) {
2056 struct perf_evsel
*evsel
;
2058 evsel
= perf_session__find_first_evtype(session
, i
);
2062 if (!(evsel
->attr
.sample_type
& PERF_SAMPLE_CPU
)) {
2063 pr_err("File does not contain CPU events. "
2064 "Remove -c option to proceed.\n");
2069 map
= cpu_map__new(cpu_list
);
2071 pr_err("Invalid cpu_list\n");
2075 for (i
= 0; i
< map
->nr
; i
++) {
2076 int cpu
= map
->map
[i
];
2078 if (cpu
>= MAX_NR_CPUS
) {
2079 pr_err("Requested CPU %d too large. "
2080 "Consider raising MAX_NR_CPUS\n", cpu
);
2081 goto out_delete_map
;
2084 set_bit(cpu
, cpu_bitmap
);
2094 void perf_session__fprintf_info(struct perf_session
*session
, FILE *fp
,
2100 if (session
== NULL
|| fp
== NULL
)
2103 fd
= perf_data_file__fd(session
->file
);
2105 ret
= fstat(fd
, &st
);
2109 fprintf(fp
, "# ========\n");
2110 fprintf(fp
, "# captured on: %s", ctime(&st
.st_ctime
));
2111 perf_header__fprintf_info(session
, fp
, full
);
2112 fprintf(fp
, "# ========\n#\n");
2116 int __perf_session__set_tracepoints_handlers(struct perf_session
*session
,
2117 const struct perf_evsel_str_handler
*assocs
,
2120 struct perf_evsel
*evsel
;
2124 for (i
= 0; i
< nr_assocs
; i
++) {
2126 * Adding a handler for an event not in the session,
2129 evsel
= perf_evlist__find_tracepoint_by_name(session
->evlist
, assocs
[i
].name
);
2134 if (evsel
->handler
!= NULL
)
2136 evsel
->handler
= assocs
[i
].handler
;
2144 int perf_event__process_id_index(struct perf_tool
*tool __maybe_unused
,
2145 union perf_event
*event
,
2146 struct perf_session
*session
)
2148 struct perf_evlist
*evlist
= session
->evlist
;
2149 struct id_index_event
*ie
= &event
->id_index
;
2150 size_t i
, nr
, max_nr
;
2152 max_nr
= (ie
->header
.size
- sizeof(struct id_index_event
)) /
2153 sizeof(struct id_index_entry
);
2159 fprintf(stdout
, " nr: %zu\n", nr
);
2161 for (i
= 0; i
< nr
; i
++) {
2162 struct id_index_entry
*e
= &ie
->entries
[i
];
2163 struct perf_sample_id
*sid
;
2166 fprintf(stdout
, " ... id: %"PRIu64
, e
->id
);
2167 fprintf(stdout
, " idx: %"PRIu64
, e
->idx
);
2168 fprintf(stdout
, " cpu: %"PRId64
, e
->cpu
);
2169 fprintf(stdout
, " tid: %"PRId64
"\n", e
->tid
);
2172 sid
= perf_evlist__id2sid(evlist
, e
->id
);
2182 int perf_event__synthesize_id_index(struct perf_tool
*tool
,
2183 perf_event__handler_t process
,
2184 struct perf_evlist
*evlist
,
2185 struct machine
*machine
)
2187 union perf_event
*ev
;
2188 struct perf_evsel
*evsel
;
2189 size_t nr
= 0, i
= 0, sz
, max_nr
, n
;
2192 pr_debug2("Synthesizing id index\n");
2194 max_nr
= (UINT16_MAX
- sizeof(struct id_index_event
)) /
2195 sizeof(struct id_index_entry
);
2197 evlist__for_each(evlist
, evsel
)
2200 n
= nr
> max_nr
? max_nr
: nr
;
2201 sz
= sizeof(struct id_index_event
) + n
* sizeof(struct id_index_entry
);
2206 ev
->id_index
.header
.type
= PERF_RECORD_ID_INDEX
;
2207 ev
->id_index
.header
.size
= sz
;
2208 ev
->id_index
.nr
= n
;
2210 evlist__for_each(evlist
, evsel
) {
2213 for (j
= 0; j
< evsel
->ids
; j
++) {
2214 struct id_index_entry
*e
;
2215 struct perf_sample_id
*sid
;
2218 err
= process(tool
, ev
, NULL
, machine
);
2225 e
= &ev
->id_index
.entries
[i
++];
2227 e
->id
= evsel
->id
[j
];
2229 sid
= perf_evlist__id2sid(evlist
, e
->id
);
2241 sz
= sizeof(struct id_index_event
) + nr
* sizeof(struct id_index_entry
);
2242 ev
->id_index
.header
.size
= sz
;
2243 ev
->id_index
.nr
= nr
;
2245 err
= process(tool
, ev
, NULL
, machine
);