2 * Copyright (C) 2011, Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
4 * Parts came from builtin-{top,stat,record}.c, see those files for further
7 * Released under the GPL v2. (and only v2, not any later version)
11 #include <linux/bitops.h>
14 #include "event-parse.h"
19 #include "thread_map.h"
21 #include "../../../include/linux/hw_breakpoint.h"
22 #include "../../../include/uapi/linux/perf_event.h"
23 #include "perf_regs.h"
25 #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y))
27 static int __perf_evsel__sample_size(u64 sample_type
)
29 u64 mask
= sample_type
& PERF_SAMPLE_MASK
;
33 for (i
= 0; i
< 64; i
++) {
34 if (mask
& (1ULL << i
))
43 void hists__init(struct hists
*hists
)
45 memset(hists
, 0, sizeof(*hists
));
46 hists
->entries_in_array
[0] = hists
->entries_in_array
[1] = RB_ROOT
;
47 hists
->entries_in
= &hists
->entries_in_array
[0];
48 hists
->entries_collapsed
= RB_ROOT
;
49 hists
->entries
= RB_ROOT
;
50 pthread_mutex_init(&hists
->lock
, NULL
);
53 void perf_evsel__init(struct perf_evsel
*evsel
,
54 struct perf_event_attr
*attr
, int idx
)
58 INIT_LIST_HEAD(&evsel
->node
);
59 hists__init(&evsel
->hists
);
60 evsel
->sample_size
= __perf_evsel__sample_size(attr
->sample_type
);
63 struct perf_evsel
*perf_evsel__new(struct perf_event_attr
*attr
, int idx
)
65 struct perf_evsel
*evsel
= zalloc(sizeof(*evsel
));
68 perf_evsel__init(evsel
, attr
, idx
);
73 struct event_format
*event_format__new(const char *sys
, const char *name
)
77 void *bf
= NULL
, *nbf
;
78 size_t size
= 0, alloc_size
= 0;
79 struct event_format
*format
= NULL
;
81 if (asprintf(&filename
, "%s/%s/%s/format", tracing_events_path
, sys
, name
) < 0)
84 fd
= open(filename
, O_RDONLY
);
86 goto out_free_filename
;
89 if (size
== alloc_size
) {
91 nbf
= realloc(bf
, alloc_size
);
97 n
= read(fd
, bf
+ size
, BUFSIZ
);
103 pevent_parse_format(&format
, bf
, size
, sys
);
114 struct perf_evsel
*perf_evsel__newtp(const char *sys
, const char *name
, int idx
)
116 struct perf_evsel
*evsel
= zalloc(sizeof(*evsel
));
119 struct perf_event_attr attr
= {
120 .type
= PERF_TYPE_TRACEPOINT
,
121 .sample_type
= (PERF_SAMPLE_RAW
| PERF_SAMPLE_TIME
|
122 PERF_SAMPLE_CPU
| PERF_SAMPLE_PERIOD
),
125 if (asprintf(&evsel
->name
, "%s:%s", sys
, name
) < 0)
128 evsel
->tp_format
= event_format__new(sys
, name
);
129 if (evsel
->tp_format
== NULL
)
132 event_attr_init(&attr
);
133 attr
.config
= evsel
->tp_format
->id
;
134 attr
.sample_period
= 1;
135 perf_evsel__init(evsel
, &attr
, idx
);
146 const char *perf_evsel__hw_names
[PERF_COUNT_HW_MAX
] = {
154 "stalled-cycles-frontend",
155 "stalled-cycles-backend",
159 static const char *__perf_evsel__hw_name(u64 config
)
161 if (config
< PERF_COUNT_HW_MAX
&& perf_evsel__hw_names
[config
])
162 return perf_evsel__hw_names
[config
];
164 return "unknown-hardware";
167 static int perf_evsel__add_modifiers(struct perf_evsel
*evsel
, char *bf
, size_t size
)
169 int colon
= 0, r
= 0;
170 struct perf_event_attr
*attr
= &evsel
->attr
;
171 bool exclude_guest_default
= false;
173 #define MOD_PRINT(context, mod) do { \
174 if (!attr->exclude_##context) { \
175 if (!colon) colon = ++r; \
176 r += scnprintf(bf + r, size - r, "%c", mod); \
179 if (attr
->exclude_kernel
|| attr
->exclude_user
|| attr
->exclude_hv
) {
180 MOD_PRINT(kernel
, 'k');
181 MOD_PRINT(user
, 'u');
183 exclude_guest_default
= true;
186 if (attr
->precise_ip
) {
189 r
+= scnprintf(bf
+ r
, size
- r
, "%.*s", attr
->precise_ip
, "ppp");
190 exclude_guest_default
= true;
193 if (attr
->exclude_host
|| attr
->exclude_guest
== exclude_guest_default
) {
194 MOD_PRINT(host
, 'H');
195 MOD_PRINT(guest
, 'G');
203 static int perf_evsel__hw_name(struct perf_evsel
*evsel
, char *bf
, size_t size
)
205 int r
= scnprintf(bf
, size
, "%s", __perf_evsel__hw_name(evsel
->attr
.config
));
206 return r
+ perf_evsel__add_modifiers(evsel
, bf
+ r
, size
- r
);
209 const char *perf_evsel__sw_names
[PERF_COUNT_SW_MAX
] = {
221 static const char *__perf_evsel__sw_name(u64 config
)
223 if (config
< PERF_COUNT_SW_MAX
&& perf_evsel__sw_names
[config
])
224 return perf_evsel__sw_names
[config
];
225 return "unknown-software";
228 static int perf_evsel__sw_name(struct perf_evsel
*evsel
, char *bf
, size_t size
)
230 int r
= scnprintf(bf
, size
, "%s", __perf_evsel__sw_name(evsel
->attr
.config
));
231 return r
+ perf_evsel__add_modifiers(evsel
, bf
+ r
, size
- r
);
234 static int __perf_evsel__bp_name(char *bf
, size_t size
, u64 addr
, u64 type
)
238 r
= scnprintf(bf
, size
, "mem:0x%" PRIx64
":", addr
);
240 if (type
& HW_BREAKPOINT_R
)
241 r
+= scnprintf(bf
+ r
, size
- r
, "r");
243 if (type
& HW_BREAKPOINT_W
)
244 r
+= scnprintf(bf
+ r
, size
- r
, "w");
246 if (type
& HW_BREAKPOINT_X
)
247 r
+= scnprintf(bf
+ r
, size
- r
, "x");
252 static int perf_evsel__bp_name(struct perf_evsel
*evsel
, char *bf
, size_t size
)
254 struct perf_event_attr
*attr
= &evsel
->attr
;
255 int r
= __perf_evsel__bp_name(bf
, size
, attr
->bp_addr
, attr
->bp_type
);
256 return r
+ perf_evsel__add_modifiers(evsel
, bf
+ r
, size
- r
);
259 const char *perf_evsel__hw_cache
[PERF_COUNT_HW_CACHE_MAX
]
260 [PERF_EVSEL__MAX_ALIASES
] = {
261 { "L1-dcache", "l1-d", "l1d", "L1-data", },
262 { "L1-icache", "l1-i", "l1i", "L1-instruction", },
264 { "dTLB", "d-tlb", "Data-TLB", },
265 { "iTLB", "i-tlb", "Instruction-TLB", },
266 { "branch", "branches", "bpu", "btb", "bpc", },
270 const char *perf_evsel__hw_cache_op
[PERF_COUNT_HW_CACHE_OP_MAX
]
271 [PERF_EVSEL__MAX_ALIASES
] = {
272 { "load", "loads", "read", },
273 { "store", "stores", "write", },
274 { "prefetch", "prefetches", "speculative-read", "speculative-load", },
277 const char *perf_evsel__hw_cache_result
[PERF_COUNT_HW_CACHE_RESULT_MAX
]
278 [PERF_EVSEL__MAX_ALIASES
] = {
279 { "refs", "Reference", "ops", "access", },
280 { "misses", "miss", },
283 #define C(x) PERF_COUNT_HW_CACHE_##x
284 #define CACHE_READ (1 << C(OP_READ))
285 #define CACHE_WRITE (1 << C(OP_WRITE))
286 #define CACHE_PREFETCH (1 << C(OP_PREFETCH))
287 #define COP(x) (1 << x)
290 * cache operartion stat
291 * L1I : Read and prefetch only
292 * ITLB and BPU : Read-only
294 static unsigned long perf_evsel__hw_cache_stat
[C(MAX
)] = {
295 [C(L1D
)] = (CACHE_READ
| CACHE_WRITE
| CACHE_PREFETCH
),
296 [C(L1I
)] = (CACHE_READ
| CACHE_PREFETCH
),
297 [C(LL
)] = (CACHE_READ
| CACHE_WRITE
| CACHE_PREFETCH
),
298 [C(DTLB
)] = (CACHE_READ
| CACHE_WRITE
| CACHE_PREFETCH
),
299 [C(ITLB
)] = (CACHE_READ
),
300 [C(BPU
)] = (CACHE_READ
),
301 [C(NODE
)] = (CACHE_READ
| CACHE_WRITE
| CACHE_PREFETCH
),
304 bool perf_evsel__is_cache_op_valid(u8 type
, u8 op
)
306 if (perf_evsel__hw_cache_stat
[type
] & COP(op
))
307 return true; /* valid */
309 return false; /* invalid */
312 int __perf_evsel__hw_cache_type_op_res_name(u8 type
, u8 op
, u8 result
,
313 char *bf
, size_t size
)
316 return scnprintf(bf
, size
, "%s-%s-%s", perf_evsel__hw_cache
[type
][0],
317 perf_evsel__hw_cache_op
[op
][0],
318 perf_evsel__hw_cache_result
[result
][0]);
321 return scnprintf(bf
, size
, "%s-%s", perf_evsel__hw_cache
[type
][0],
322 perf_evsel__hw_cache_op
[op
][1]);
325 static int __perf_evsel__hw_cache_name(u64 config
, char *bf
, size_t size
)
327 u8 op
, result
, type
= (config
>> 0) & 0xff;
328 const char *err
= "unknown-ext-hardware-cache-type";
330 if (type
> PERF_COUNT_HW_CACHE_MAX
)
333 op
= (config
>> 8) & 0xff;
334 err
= "unknown-ext-hardware-cache-op";
335 if (op
> PERF_COUNT_HW_CACHE_OP_MAX
)
338 result
= (config
>> 16) & 0xff;
339 err
= "unknown-ext-hardware-cache-result";
340 if (result
> PERF_COUNT_HW_CACHE_RESULT_MAX
)
343 err
= "invalid-cache";
344 if (!perf_evsel__is_cache_op_valid(type
, op
))
347 return __perf_evsel__hw_cache_type_op_res_name(type
, op
, result
, bf
, size
);
349 return scnprintf(bf
, size
, "%s", err
);
352 static int perf_evsel__hw_cache_name(struct perf_evsel
*evsel
, char *bf
, size_t size
)
354 int ret
= __perf_evsel__hw_cache_name(evsel
->attr
.config
, bf
, size
);
355 return ret
+ perf_evsel__add_modifiers(evsel
, bf
+ ret
, size
- ret
);
358 static int perf_evsel__raw_name(struct perf_evsel
*evsel
, char *bf
, size_t size
)
360 int ret
= scnprintf(bf
, size
, "raw 0x%" PRIx64
, evsel
->attr
.config
);
361 return ret
+ perf_evsel__add_modifiers(evsel
, bf
+ ret
, size
- ret
);
364 const char *perf_evsel__name(struct perf_evsel
*evsel
)
371 switch (evsel
->attr
.type
) {
373 perf_evsel__raw_name(evsel
, bf
, sizeof(bf
));
376 case PERF_TYPE_HARDWARE
:
377 perf_evsel__hw_name(evsel
, bf
, sizeof(bf
));
380 case PERF_TYPE_HW_CACHE
:
381 perf_evsel__hw_cache_name(evsel
, bf
, sizeof(bf
));
384 case PERF_TYPE_SOFTWARE
:
385 perf_evsel__sw_name(evsel
, bf
, sizeof(bf
));
388 case PERF_TYPE_TRACEPOINT
:
389 scnprintf(bf
, sizeof(bf
), "%s", "unknown tracepoint");
392 case PERF_TYPE_BREAKPOINT
:
393 perf_evsel__bp_name(evsel
, bf
, sizeof(bf
));
397 scnprintf(bf
, sizeof(bf
), "unknown attr type: %d",
402 evsel
->name
= strdup(bf
);
404 return evsel
->name
?: "unknown";
407 void perf_evsel__config(struct perf_evsel
*evsel
,
408 struct perf_record_opts
*opts
)
410 struct perf_event_attr
*attr
= &evsel
->attr
;
411 int track
= !evsel
->idx
; /* only the first counter needs these */
414 attr
->sample_id_all
= opts
->sample_id_all_missing
? 0 : 1;
415 attr
->inherit
= !opts
->no_inherit
;
416 attr
->read_format
= PERF_FORMAT_TOTAL_TIME_ENABLED
|
417 PERF_FORMAT_TOTAL_TIME_RUNNING
|
420 attr
->sample_type
|= PERF_SAMPLE_IP
| PERF_SAMPLE_TID
;
423 * We default some events to a 1 default interval. But keep
424 * it a weak assumption overridable by the user.
426 if (!attr
->sample_period
|| (opts
->user_freq
!= UINT_MAX
&&
427 opts
->user_interval
!= ULLONG_MAX
)) {
429 attr
->sample_type
|= PERF_SAMPLE_PERIOD
;
431 attr
->sample_freq
= opts
->freq
;
433 attr
->sample_period
= opts
->default_interval
;
437 if (opts
->no_samples
)
438 attr
->sample_freq
= 0;
440 if (opts
->inherit_stat
)
441 attr
->inherit_stat
= 1;
443 if (opts
->sample_address
) {
444 attr
->sample_type
|= PERF_SAMPLE_ADDR
;
445 attr
->mmap_data
= track
;
448 if (opts
->call_graph
) {
449 attr
->sample_type
|= PERF_SAMPLE_CALLCHAIN
;
451 if (opts
->call_graph
== CALLCHAIN_DWARF
) {
452 attr
->sample_type
|= PERF_SAMPLE_REGS_USER
|
453 PERF_SAMPLE_STACK_USER
;
454 attr
->sample_regs_user
= PERF_REGS_MASK
;
455 attr
->sample_stack_user
= opts
->stack_dump_size
;
456 attr
->exclude_callchain_user
= 1;
460 if (perf_target__has_cpu(&opts
->target
))
461 attr
->sample_type
|= PERF_SAMPLE_CPU
;
464 attr
->sample_type
|= PERF_SAMPLE_PERIOD
;
466 if (!opts
->sample_id_all_missing
&&
467 (opts
->sample_time
|| !opts
->no_inherit
||
468 perf_target__has_cpu(&opts
->target
)))
469 attr
->sample_type
|= PERF_SAMPLE_TIME
;
471 if (opts
->raw_samples
) {
472 attr
->sample_type
|= PERF_SAMPLE_TIME
;
473 attr
->sample_type
|= PERF_SAMPLE_RAW
;
474 attr
->sample_type
|= PERF_SAMPLE_CPU
;
477 if (opts
->no_delay
) {
479 attr
->wakeup_events
= 1;
481 if (opts
->branch_stack
) {
482 attr
->sample_type
|= PERF_SAMPLE_BRANCH_STACK
;
483 attr
->branch_sample_type
= opts
->branch_stack
;
489 if (perf_target__none(&opts
->target
) && (!evsel
->leader
))
490 attr
->enable_on_exec
= 1;
493 int perf_evsel__alloc_fd(struct perf_evsel
*evsel
, int ncpus
, int nthreads
)
496 evsel
->fd
= xyarray__new(ncpus
, nthreads
, sizeof(int));
499 for (cpu
= 0; cpu
< ncpus
; cpu
++) {
500 for (thread
= 0; thread
< nthreads
; thread
++) {
501 FD(evsel
, cpu
, thread
) = -1;
506 return evsel
->fd
!= NULL
? 0 : -ENOMEM
;
509 int perf_evsel__set_filter(struct perf_evsel
*evsel
, int ncpus
, int nthreads
,
514 for (cpu
= 0; cpu
< ncpus
; cpu
++) {
515 for (thread
= 0; thread
< nthreads
; thread
++) {
516 int fd
= FD(evsel
, cpu
, thread
),
517 err
= ioctl(fd
, PERF_EVENT_IOC_SET_FILTER
, filter
);
527 int perf_evsel__alloc_id(struct perf_evsel
*evsel
, int ncpus
, int nthreads
)
529 evsel
->sample_id
= xyarray__new(ncpus
, nthreads
, sizeof(struct perf_sample_id
));
530 if (evsel
->sample_id
== NULL
)
533 evsel
->id
= zalloc(ncpus
* nthreads
* sizeof(u64
));
534 if (evsel
->id
== NULL
) {
535 xyarray__delete(evsel
->sample_id
);
536 evsel
->sample_id
= NULL
;
543 int perf_evsel__alloc_counts(struct perf_evsel
*evsel
, int ncpus
)
545 evsel
->counts
= zalloc((sizeof(*evsel
->counts
) +
546 (ncpus
* sizeof(struct perf_counts_values
))));
547 return evsel
->counts
!= NULL
? 0 : -ENOMEM
;
550 void perf_evsel__free_fd(struct perf_evsel
*evsel
)
552 xyarray__delete(evsel
->fd
);
556 void perf_evsel__free_id(struct perf_evsel
*evsel
)
558 xyarray__delete(evsel
->sample_id
);
559 evsel
->sample_id
= NULL
;
564 void perf_evsel__close_fd(struct perf_evsel
*evsel
, int ncpus
, int nthreads
)
568 for (cpu
= 0; cpu
< ncpus
; cpu
++)
569 for (thread
= 0; thread
< nthreads
; ++thread
) {
570 close(FD(evsel
, cpu
, thread
));
571 FD(evsel
, cpu
, thread
) = -1;
575 void perf_evsel__exit(struct perf_evsel
*evsel
)
577 assert(list_empty(&evsel
->node
));
578 xyarray__delete(evsel
->fd
);
579 xyarray__delete(evsel
->sample_id
);
583 void perf_evsel__delete(struct perf_evsel
*evsel
)
585 perf_evsel__exit(evsel
);
586 close_cgroup(evsel
->cgrp
);
587 free(evsel
->group_name
);
588 if (evsel
->tp_format
)
589 pevent_free_format(evsel
->tp_format
);
594 int __perf_evsel__read_on_cpu(struct perf_evsel
*evsel
,
595 int cpu
, int thread
, bool scale
)
597 struct perf_counts_values count
;
598 size_t nv
= scale
? 3 : 1;
600 if (FD(evsel
, cpu
, thread
) < 0)
603 if (evsel
->counts
== NULL
&& perf_evsel__alloc_counts(evsel
, cpu
+ 1) < 0)
606 if (readn(FD(evsel
, cpu
, thread
), &count
, nv
* sizeof(u64
)) < 0)
612 else if (count
.run
< count
.ena
)
613 count
.val
= (u64
)((double)count
.val
* count
.ena
/ count
.run
+ 0.5);
615 count
.ena
= count
.run
= 0;
617 evsel
->counts
->cpu
[cpu
] = count
;
621 int __perf_evsel__read(struct perf_evsel
*evsel
,
622 int ncpus
, int nthreads
, bool scale
)
624 size_t nv
= scale
? 3 : 1;
626 struct perf_counts_values
*aggr
= &evsel
->counts
->aggr
, count
;
628 aggr
->val
= aggr
->ena
= aggr
->run
= 0;
630 for (cpu
= 0; cpu
< ncpus
; cpu
++) {
631 for (thread
= 0; thread
< nthreads
; thread
++) {
632 if (FD(evsel
, cpu
, thread
) < 0)
635 if (readn(FD(evsel
, cpu
, thread
),
636 &count
, nv
* sizeof(u64
)) < 0)
639 aggr
->val
+= count
.val
;
641 aggr
->ena
+= count
.ena
;
642 aggr
->run
+= count
.run
;
647 evsel
->counts
->scaled
= 0;
649 if (aggr
->run
== 0) {
650 evsel
->counts
->scaled
= -1;
655 if (aggr
->run
< aggr
->ena
) {
656 evsel
->counts
->scaled
= 1;
657 aggr
->val
= (u64
)((double)aggr
->val
* aggr
->ena
/ aggr
->run
+ 0.5);
660 aggr
->ena
= aggr
->run
= 0;
665 static int get_group_fd(struct perf_evsel
*evsel
, int cpu
, int thread
)
667 struct perf_evsel
*leader
= evsel
->leader
;
674 * Leader must be already processed/open,
679 fd
= FD(leader
, cpu
, thread
);
685 static int __perf_evsel__open(struct perf_evsel
*evsel
, struct cpu_map
*cpus
,
686 struct thread_map
*threads
)
689 unsigned long flags
= 0;
692 if (evsel
->fd
== NULL
&&
693 perf_evsel__alloc_fd(evsel
, cpus
->nr
, threads
->nr
) < 0)
697 flags
= PERF_FLAG_PID_CGROUP
;
698 pid
= evsel
->cgrp
->fd
;
701 for (cpu
= 0; cpu
< cpus
->nr
; cpu
++) {
703 for (thread
= 0; thread
< threads
->nr
; thread
++) {
707 pid
= threads
->map
[thread
];
709 group_fd
= get_group_fd(evsel
, cpu
, thread
);
711 FD(evsel
, cpu
, thread
) = sys_perf_event_open(&evsel
->attr
,
715 if (FD(evsel
, cpu
, thread
) < 0) {
726 while (--thread
>= 0) {
727 close(FD(evsel
, cpu
, thread
));
728 FD(evsel
, cpu
, thread
) = -1;
730 thread
= threads
->nr
;
731 } while (--cpu
>= 0);
735 void perf_evsel__close(struct perf_evsel
*evsel
, int ncpus
, int nthreads
)
737 if (evsel
->fd
== NULL
)
740 perf_evsel__close_fd(evsel
, ncpus
, nthreads
);
741 perf_evsel__free_fd(evsel
);
754 struct thread_map map
;
756 } empty_thread_map
= {
761 int perf_evsel__open(struct perf_evsel
*evsel
, struct cpu_map
*cpus
,
762 struct thread_map
*threads
)
765 /* Work around old compiler warnings about strict aliasing */
766 cpus
= &empty_cpu_map
.map
;
770 threads
= &empty_thread_map
.map
;
772 return __perf_evsel__open(evsel
, cpus
, threads
);
775 int perf_evsel__open_per_cpu(struct perf_evsel
*evsel
,
776 struct cpu_map
*cpus
)
778 return __perf_evsel__open(evsel
, cpus
, &empty_thread_map
.map
);
781 int perf_evsel__open_per_thread(struct perf_evsel
*evsel
,
782 struct thread_map
*threads
)
784 return __perf_evsel__open(evsel
, &empty_cpu_map
.map
, threads
);
787 static int perf_evsel__parse_id_sample(const struct perf_evsel
*evsel
,
788 const union perf_event
*event
,
789 struct perf_sample
*sample
)
791 u64 type
= evsel
->attr
.sample_type
;
792 const u64
*array
= event
->sample
.array
;
793 bool swapped
= evsel
->needs_swap
;
796 array
+= ((event
->header
.size
-
797 sizeof(event
->header
)) / sizeof(u64
)) - 1;
799 if (type
& PERF_SAMPLE_CPU
) {
802 /* undo swap of u64, then swap on individual u32s */
803 u
.val64
= bswap_64(u
.val64
);
804 u
.val32
[0] = bswap_32(u
.val32
[0]);
807 sample
->cpu
= u
.val32
[0];
811 if (type
& PERF_SAMPLE_STREAM_ID
) {
812 sample
->stream_id
= *array
;
816 if (type
& PERF_SAMPLE_ID
) {
821 if (type
& PERF_SAMPLE_TIME
) {
822 sample
->time
= *array
;
826 if (type
& PERF_SAMPLE_TID
) {
829 /* undo swap of u64, then swap on individual u32s */
830 u
.val64
= bswap_64(u
.val64
);
831 u
.val32
[0] = bswap_32(u
.val32
[0]);
832 u
.val32
[1] = bswap_32(u
.val32
[1]);
835 sample
->pid
= u
.val32
[0];
836 sample
->tid
= u
.val32
[1];
842 static bool sample_overlap(const union perf_event
*event
,
843 const void *offset
, u64 size
)
845 const void *base
= event
;
847 if (offset
+ size
> base
+ event
->header
.size
)
853 int perf_evsel__parse_sample(struct perf_evsel
*evsel
, union perf_event
*event
,
854 struct perf_sample
*data
)
856 u64 type
= evsel
->attr
.sample_type
;
857 u64 regs_user
= evsel
->attr
.sample_regs_user
;
858 bool swapped
= evsel
->needs_swap
;
862 * used for cross-endian analysis. See git commit 65014ab3
863 * for why this goofiness is needed.
867 memset(data
, 0, sizeof(*data
));
868 data
->cpu
= data
->pid
= data
->tid
= -1;
869 data
->stream_id
= data
->id
= data
->time
= -1ULL;
872 if (event
->header
.type
!= PERF_RECORD_SAMPLE
) {
873 if (!evsel
->attr
.sample_id_all
)
875 return perf_evsel__parse_id_sample(evsel
, event
, data
);
878 array
= event
->sample
.array
;
880 if (evsel
->sample_size
+ sizeof(event
->header
) > event
->header
.size
)
883 if (type
& PERF_SAMPLE_IP
) {
884 data
->ip
= event
->ip
.ip
;
888 if (type
& PERF_SAMPLE_TID
) {
891 /* undo swap of u64, then swap on individual u32s */
892 u
.val64
= bswap_64(u
.val64
);
893 u
.val32
[0] = bswap_32(u
.val32
[0]);
894 u
.val32
[1] = bswap_32(u
.val32
[1]);
897 data
->pid
= u
.val32
[0];
898 data
->tid
= u
.val32
[1];
902 if (type
& PERF_SAMPLE_TIME
) {
908 if (type
& PERF_SAMPLE_ADDR
) {
914 if (type
& PERF_SAMPLE_ID
) {
919 if (type
& PERF_SAMPLE_STREAM_ID
) {
920 data
->stream_id
= *array
;
924 if (type
& PERF_SAMPLE_CPU
) {
928 /* undo swap of u64, then swap on individual u32s */
929 u
.val64
= bswap_64(u
.val64
);
930 u
.val32
[0] = bswap_32(u
.val32
[0]);
933 data
->cpu
= u
.val32
[0];
937 if (type
& PERF_SAMPLE_PERIOD
) {
938 data
->period
= *array
;
942 if (type
& PERF_SAMPLE_READ
) {
943 fprintf(stderr
, "PERF_SAMPLE_READ is unsupported for now\n");
947 if (type
& PERF_SAMPLE_CALLCHAIN
) {
948 if (sample_overlap(event
, array
, sizeof(data
->callchain
->nr
)))
951 data
->callchain
= (struct ip_callchain
*)array
;
953 if (sample_overlap(event
, array
, data
->callchain
->nr
))
956 array
+= 1 + data
->callchain
->nr
;
959 if (type
& PERF_SAMPLE_RAW
) {
963 if (WARN_ONCE(swapped
,
964 "Endianness of raw data not corrected!\n")) {
965 /* undo swap of u64, then swap on individual u32s */
966 u
.val64
= bswap_64(u
.val64
);
967 u
.val32
[0] = bswap_32(u
.val32
[0]);
968 u
.val32
[1] = bswap_32(u
.val32
[1]);
971 if (sample_overlap(event
, array
, sizeof(u32
)))
974 data
->raw_size
= u
.val32
[0];
975 pdata
= (void *) array
+ sizeof(u32
);
977 if (sample_overlap(event
, pdata
, data
->raw_size
))
980 data
->raw_data
= (void *) pdata
;
982 array
= (void *)array
+ data
->raw_size
+ sizeof(u32
);
985 if (type
& PERF_SAMPLE_BRANCH_STACK
) {
988 data
->branch_stack
= (struct branch_stack
*)array
;
991 sz
= data
->branch_stack
->nr
* sizeof(struct branch_entry
);
996 if (type
& PERF_SAMPLE_REGS_USER
) {
997 /* First u64 tells us if we have any regs in sample. */
998 u64 avail
= *array
++;
1001 data
->user_regs
.regs
= (u64
*)array
;
1002 array
+= hweight_long(regs_user
);
1006 if (type
& PERF_SAMPLE_STACK_USER
) {
1007 u64 size
= *array
++;
1009 data
->user_stack
.offset
= ((char *)(array
- 1)
1013 data
->user_stack
.size
= 0;
1015 data
->user_stack
.data
= (char *)array
;
1016 array
+= size
/ sizeof(*array
);
1017 data
->user_stack
.size
= *array
;
1024 int perf_event__synthesize_sample(union perf_event
*event
, u64 type
,
1025 const struct perf_sample
*sample
,
1031 * used for cross-endian analysis. See git commit 65014ab3
1032 * for why this goofiness is needed.
1036 array
= event
->sample
.array
;
1038 if (type
& PERF_SAMPLE_IP
) {
1039 event
->ip
.ip
= sample
->ip
;
1043 if (type
& PERF_SAMPLE_TID
) {
1044 u
.val32
[0] = sample
->pid
;
1045 u
.val32
[1] = sample
->tid
;
1048 * Inverse of what is done in perf_evsel__parse_sample
1050 u
.val32
[0] = bswap_32(u
.val32
[0]);
1051 u
.val32
[1] = bswap_32(u
.val32
[1]);
1052 u
.val64
= bswap_64(u
.val64
);
1059 if (type
& PERF_SAMPLE_TIME
) {
1060 *array
= sample
->time
;
1064 if (type
& PERF_SAMPLE_ADDR
) {
1065 *array
= sample
->addr
;
1069 if (type
& PERF_SAMPLE_ID
) {
1070 *array
= sample
->id
;
1074 if (type
& PERF_SAMPLE_STREAM_ID
) {
1075 *array
= sample
->stream_id
;
1079 if (type
& PERF_SAMPLE_CPU
) {
1080 u
.val32
[0] = sample
->cpu
;
1083 * Inverse of what is done in perf_evsel__parse_sample
1085 u
.val32
[0] = bswap_32(u
.val32
[0]);
1086 u
.val64
= bswap_64(u
.val64
);
1092 if (type
& PERF_SAMPLE_PERIOD
) {
1093 *array
= sample
->period
;
1100 struct format_field
*perf_evsel__field(struct perf_evsel
*evsel
, const char *name
)
1102 return pevent_find_field(evsel
->tp_format
, name
);
1105 void *perf_evsel__rawptr(struct perf_evsel
*evsel
, struct perf_sample
*sample
,
1108 struct format_field
*field
= perf_evsel__field(evsel
, name
);
1114 offset
= field
->offset
;
1116 if (field
->flags
& FIELD_IS_DYNAMIC
) {
1117 offset
= *(int *)(sample
->raw_data
+ field
->offset
);
1121 return sample
->raw_data
+ offset
;
1124 u64
perf_evsel__intval(struct perf_evsel
*evsel
, struct perf_sample
*sample
,
1127 struct format_field
*field
= perf_evsel__field(evsel
, name
);
1134 ptr
= sample
->raw_data
+ field
->offset
;
1136 switch (field
->size
) {
1140 value
= *(u16
*)ptr
;
1143 value
= *(u32
*)ptr
;
1146 value
= *(u64
*)ptr
;
1152 if (!evsel
->needs_swap
)
1155 switch (field
->size
) {
1157 return bswap_16(value
);
1159 return bswap_32(value
);
1161 return bswap_64(value
);