Loading...
1// SPDX-License-Identifier: GPL-2.0
2#include <errno.h>
3#include <signal.h>
4#include <inttypes.h>
5#include <linux/err.h>
6#include <linux/kernel.h>
7#include <linux/zalloc.h>
8#include <api/fs/fs.h>
9
10#include <byteswap.h>
11#include <unistd.h>
12#include <sys/types.h>
13#include <sys/mman.h>
14#include <perf/cpumap.h>
15
16#include "map_symbol.h"
17#include "branch.h"
18#include "debug.h"
19#include "env.h"
20#include "evlist.h"
21#include "evsel.h"
22#include "memswap.h"
23#include "map.h"
24#include "symbol.h"
25#include "session.h"
26#include "tool.h"
27#include "perf_regs.h"
28#include "asm/bug.h"
29#include "auxtrace.h"
30#include "thread.h"
31#include "thread-stack.h"
32#include "sample-raw.h"
33#include "stat.h"
34#include "tsc.h"
35#include "ui/progress.h"
36#include "../perf.h"
37#include "arch/common.h"
38#include "units.h"
39#include <internal/lib.h>
40
41#ifdef HAVE_ZSTD_SUPPORT
42static int perf_session__process_compressed_event(struct perf_session *session,
43 union perf_event *event, u64 file_offset,
44 const char *file_path)
45{
46 void *src;
47 size_t decomp_size, src_size;
48 u64 decomp_last_rem = 0;
49 size_t mmap_len, decomp_len = session->header.env.comp_mmap_len;
50 struct decomp *decomp, *decomp_last = session->active_decomp->decomp_last;
51
52 if (decomp_last) {
53 decomp_last_rem = decomp_last->size - decomp_last->head;
54 decomp_len += decomp_last_rem;
55 }
56
57 mmap_len = sizeof(struct decomp) + decomp_len;
58 decomp = mmap(NULL, mmap_len, PROT_READ|PROT_WRITE,
59 MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
60 if (decomp == MAP_FAILED) {
61 pr_err("Couldn't allocate memory for decompression\n");
62 return -1;
63 }
64
65 decomp->file_pos = file_offset;
66 decomp->file_path = file_path;
67 decomp->mmap_len = mmap_len;
68 decomp->head = 0;
69
70 if (decomp_last_rem) {
71 memcpy(decomp->data, &(decomp_last->data[decomp_last->head]), decomp_last_rem);
72 decomp->size = decomp_last_rem;
73 }
74
75 src = (void *)event + sizeof(struct perf_record_compressed);
76 src_size = event->pack.header.size - sizeof(struct perf_record_compressed);
77
78 decomp_size = zstd_decompress_stream(session->active_decomp->zstd_decomp, src, src_size,
79 &(decomp->data[decomp_last_rem]), decomp_len - decomp_last_rem);
80 if (!decomp_size) {
81 munmap(decomp, mmap_len);
82 pr_err("Couldn't decompress data\n");
83 return -1;
84 }
85
86 decomp->size += decomp_size;
87
88 if (session->active_decomp->decomp == NULL)
89 session->active_decomp->decomp = decomp;
90 else
91 session->active_decomp->decomp_last->next = decomp;
92
93 session->active_decomp->decomp_last = decomp;
94
95 pr_debug("decomp (B): %zd to %zd\n", src_size, decomp_size);
96
97 return 0;
98}
99#else /* !HAVE_ZSTD_SUPPORT */
100#define perf_session__process_compressed_event perf_session__process_compressed_event_stub
101#endif
102
103static int perf_session__deliver_event(struct perf_session *session,
104 union perf_event *event,
105 struct perf_tool *tool,
106 u64 file_offset,
107 const char *file_path);
108
109static int perf_session__open(struct perf_session *session, int repipe_fd)
110{
111 struct perf_data *data = session->data;
112
113 if (perf_session__read_header(session, repipe_fd) < 0) {
114 pr_err("incompatible file format (rerun with -v to learn more)\n");
115 return -1;
116 }
117
118 if (perf_data__is_pipe(data))
119 return 0;
120
121 if (perf_header__has_feat(&session->header, HEADER_STAT))
122 return 0;
123
124 if (!evlist__valid_sample_type(session->evlist)) {
125 pr_err("non matching sample_type\n");
126 return -1;
127 }
128
129 if (!evlist__valid_sample_id_all(session->evlist)) {
130 pr_err("non matching sample_id_all\n");
131 return -1;
132 }
133
134 if (!evlist__valid_read_format(session->evlist)) {
135 pr_err("non matching read_format\n");
136 return -1;
137 }
138
139 return 0;
140}
141
142void perf_session__set_id_hdr_size(struct perf_session *session)
143{
144 u16 id_hdr_size = evlist__id_hdr_size(session->evlist);
145
146 machines__set_id_hdr_size(&session->machines, id_hdr_size);
147}
148
149int perf_session__create_kernel_maps(struct perf_session *session)
150{
151 int ret = machine__create_kernel_maps(&session->machines.host);
152
153 if (ret >= 0)
154 ret = machines__create_guest_kernel_maps(&session->machines);
155 return ret;
156}
157
158static void perf_session__destroy_kernel_maps(struct perf_session *session)
159{
160 machines__destroy_kernel_maps(&session->machines);
161}
162
163static bool perf_session__has_comm_exec(struct perf_session *session)
164{
165 struct evsel *evsel;
166
167 evlist__for_each_entry(session->evlist, evsel) {
168 if (evsel->core.attr.comm_exec)
169 return true;
170 }
171
172 return false;
173}
174
175static void perf_session__set_comm_exec(struct perf_session *session)
176{
177 bool comm_exec = perf_session__has_comm_exec(session);
178
179 machines__set_comm_exec(&session->machines, comm_exec);
180}
181
182static int ordered_events__deliver_event(struct ordered_events *oe,
183 struct ordered_event *event)
184{
185 struct perf_session *session = container_of(oe, struct perf_session,
186 ordered_events);
187
188 return perf_session__deliver_event(session, event->event,
189 session->tool, event->file_offset,
190 event->file_path);
191}
192
193struct perf_session *__perf_session__new(struct perf_data *data,
194 bool repipe, int repipe_fd,
195 struct perf_tool *tool)
196{
197 int ret = -ENOMEM;
198 struct perf_session *session = zalloc(sizeof(*session));
199
200 if (!session)
201 goto out;
202
203 session->repipe = repipe;
204 session->tool = tool;
205 session->decomp_data.zstd_decomp = &session->zstd_data;
206 session->active_decomp = &session->decomp_data;
207 INIT_LIST_HEAD(&session->auxtrace_index);
208 machines__init(&session->machines);
209 ordered_events__init(&session->ordered_events,
210 ordered_events__deliver_event, NULL);
211
212 perf_env__init(&session->header.env);
213 if (data) {
214 ret = perf_data__open(data);
215 if (ret < 0)
216 goto out_delete;
217
218 session->data = data;
219
220 if (perf_data__is_read(data)) {
221 ret = perf_session__open(session, repipe_fd);
222 if (ret < 0)
223 goto out_delete;
224
225 /*
226 * set session attributes that are present in perf.data
227 * but not in pipe-mode.
228 */
229 if (!data->is_pipe) {
230 perf_session__set_id_hdr_size(session);
231 perf_session__set_comm_exec(session);
232 }
233
234 evlist__init_trace_event_sample_raw(session->evlist);
235
236 /* Open the directory data. */
237 if (data->is_dir) {
238 ret = perf_data__open_dir(data);
239 if (ret)
240 goto out_delete;
241 }
242
243 if (!symbol_conf.kallsyms_name &&
244 !symbol_conf.vmlinux_name)
245 symbol_conf.kallsyms_name = perf_data__kallsyms_name(data);
246 }
247 } else {
248 session->machines.host.env = &perf_env;
249 }
250
251 session->machines.host.single_address_space =
252 perf_env__single_address_space(session->machines.host.env);
253
254 if (!data || perf_data__is_write(data)) {
255 /*
256 * In O_RDONLY mode this will be performed when reading the
257 * kernel MMAP event, in perf_event__process_mmap().
258 */
259 if (perf_session__create_kernel_maps(session) < 0)
260 pr_warning("Cannot read kernel map\n");
261 }
262
263 /*
264 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
265 * processed, so evlist__sample_id_all is not meaningful here.
266 */
267 if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps &&
268 tool->ordered_events && !evlist__sample_id_all(session->evlist)) {
269 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
270 tool->ordered_events = false;
271 }
272
273 return session;
274
275 out_delete:
276 perf_session__delete(session);
277 out:
278 return ERR_PTR(ret);
279}
280
281static void perf_session__delete_threads(struct perf_session *session)
282{
283 machine__delete_threads(&session->machines.host);
284}
285
286static void perf_decomp__release_events(struct decomp *next)
287{
288 struct decomp *decomp;
289 size_t mmap_len;
290
291 do {
292 decomp = next;
293 if (decomp == NULL)
294 break;
295 next = decomp->next;
296 mmap_len = decomp->mmap_len;
297 munmap(decomp, mmap_len);
298 } while (1);
299}
300
301void perf_session__delete(struct perf_session *session)
302{
303 if (session == NULL)
304 return;
305 auxtrace__free(session);
306 auxtrace_index__free(&session->auxtrace_index);
307 perf_session__destroy_kernel_maps(session);
308 perf_session__delete_threads(session);
309 perf_decomp__release_events(session->decomp_data.decomp);
310 perf_env__exit(&session->header.env);
311 machines__exit(&session->machines);
312 if (session->data) {
313 if (perf_data__is_read(session->data))
314 evlist__delete(session->evlist);
315 perf_data__close(session->data);
316 }
317#ifdef HAVE_LIBTRACEEVENT
318 trace_event__cleanup(&session->tevent);
319#endif
320 free(session);
321}
322
323static int process_event_synth_tracing_data_stub(struct perf_session *session
324 __maybe_unused,
325 union perf_event *event
326 __maybe_unused)
327{
328 dump_printf(": unhandled!\n");
329 return 0;
330}
331
332static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
333 union perf_event *event __maybe_unused,
334 struct evlist **pevlist
335 __maybe_unused)
336{
337 dump_printf(": unhandled!\n");
338 return 0;
339}
340
341static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
342 union perf_event *event __maybe_unused,
343 struct evlist **pevlist
344 __maybe_unused)
345{
346 if (dump_trace)
347 perf_event__fprintf_event_update(event, stdout);
348
349 dump_printf(": unhandled!\n");
350 return 0;
351}
352
353static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
354 union perf_event *event __maybe_unused,
355 struct perf_sample *sample __maybe_unused,
356 struct evsel *evsel __maybe_unused,
357 struct machine *machine __maybe_unused)
358{
359 dump_printf(": unhandled!\n");
360 return 0;
361}
362
363static int process_event_stub(struct perf_tool *tool __maybe_unused,
364 union perf_event *event __maybe_unused,
365 struct perf_sample *sample __maybe_unused,
366 struct machine *machine __maybe_unused)
367{
368 dump_printf(": unhandled!\n");
369 return 0;
370}
371
372static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
373 union perf_event *event __maybe_unused,
374 struct ordered_events *oe __maybe_unused)
375{
376 dump_printf(": unhandled!\n");
377 return 0;
378}
379
380static int skipn(int fd, off_t n)
381{
382 char buf[4096];
383 ssize_t ret;
384
385 while (n > 0) {
386 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
387 if (ret <= 0)
388 return ret;
389 n -= ret;
390 }
391
392 return 0;
393}
394
395static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused,
396 union perf_event *event)
397{
398 dump_printf(": unhandled!\n");
399 if (perf_data__is_pipe(session->data))
400 skipn(perf_data__fd(session->data), event->auxtrace.size);
401 return event->auxtrace.size;
402}
403
404static int process_event_op2_stub(struct perf_session *session __maybe_unused,
405 union perf_event *event __maybe_unused)
406{
407 dump_printf(": unhandled!\n");
408 return 0;
409}
410
411
412static
413int process_event_thread_map_stub(struct perf_session *session __maybe_unused,
414 union perf_event *event __maybe_unused)
415{
416 if (dump_trace)
417 perf_event__fprintf_thread_map(event, stdout);
418
419 dump_printf(": unhandled!\n");
420 return 0;
421}
422
423static
424int process_event_cpu_map_stub(struct perf_session *session __maybe_unused,
425 union perf_event *event __maybe_unused)
426{
427 if (dump_trace)
428 perf_event__fprintf_cpu_map(event, stdout);
429
430 dump_printf(": unhandled!\n");
431 return 0;
432}
433
434static
435int process_event_stat_config_stub(struct perf_session *session __maybe_unused,
436 union perf_event *event __maybe_unused)
437{
438 if (dump_trace)
439 perf_event__fprintf_stat_config(event, stdout);
440
441 dump_printf(": unhandled!\n");
442 return 0;
443}
444
445static int process_stat_stub(struct perf_session *perf_session __maybe_unused,
446 union perf_event *event)
447{
448 if (dump_trace)
449 perf_event__fprintf_stat(event, stdout);
450
451 dump_printf(": unhandled!\n");
452 return 0;
453}
454
455static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused,
456 union perf_event *event)
457{
458 if (dump_trace)
459 perf_event__fprintf_stat_round(event, stdout);
460
461 dump_printf(": unhandled!\n");
462 return 0;
463}
464
465static int process_event_time_conv_stub(struct perf_session *perf_session __maybe_unused,
466 union perf_event *event)
467{
468 if (dump_trace)
469 perf_event__fprintf_time_conv(event, stdout);
470
471 dump_printf(": unhandled!\n");
472 return 0;
473}
474
475static int perf_session__process_compressed_event_stub(struct perf_session *session __maybe_unused,
476 union perf_event *event __maybe_unused,
477 u64 file_offset __maybe_unused,
478 const char *file_path __maybe_unused)
479{
480 dump_printf(": unhandled!\n");
481 return 0;
482}
483
484void perf_tool__fill_defaults(struct perf_tool *tool)
485{
486 if (tool->sample == NULL)
487 tool->sample = process_event_sample_stub;
488 if (tool->mmap == NULL)
489 tool->mmap = process_event_stub;
490 if (tool->mmap2 == NULL)
491 tool->mmap2 = process_event_stub;
492 if (tool->comm == NULL)
493 tool->comm = process_event_stub;
494 if (tool->namespaces == NULL)
495 tool->namespaces = process_event_stub;
496 if (tool->cgroup == NULL)
497 tool->cgroup = process_event_stub;
498 if (tool->fork == NULL)
499 tool->fork = process_event_stub;
500 if (tool->exit == NULL)
501 tool->exit = process_event_stub;
502 if (tool->lost == NULL)
503 tool->lost = perf_event__process_lost;
504 if (tool->lost_samples == NULL)
505 tool->lost_samples = perf_event__process_lost_samples;
506 if (tool->aux == NULL)
507 tool->aux = perf_event__process_aux;
508 if (tool->itrace_start == NULL)
509 tool->itrace_start = perf_event__process_itrace_start;
510 if (tool->context_switch == NULL)
511 tool->context_switch = perf_event__process_switch;
512 if (tool->ksymbol == NULL)
513 tool->ksymbol = perf_event__process_ksymbol;
514 if (tool->bpf == NULL)
515 tool->bpf = perf_event__process_bpf;
516 if (tool->text_poke == NULL)
517 tool->text_poke = perf_event__process_text_poke;
518 if (tool->aux_output_hw_id == NULL)
519 tool->aux_output_hw_id = perf_event__process_aux_output_hw_id;
520 if (tool->read == NULL)
521 tool->read = process_event_sample_stub;
522 if (tool->throttle == NULL)
523 tool->throttle = process_event_stub;
524 if (tool->unthrottle == NULL)
525 tool->unthrottle = process_event_stub;
526 if (tool->attr == NULL)
527 tool->attr = process_event_synth_attr_stub;
528 if (tool->event_update == NULL)
529 tool->event_update = process_event_synth_event_update_stub;
530 if (tool->tracing_data == NULL)
531 tool->tracing_data = process_event_synth_tracing_data_stub;
532 if (tool->build_id == NULL)
533 tool->build_id = process_event_op2_stub;
534 if (tool->finished_round == NULL) {
535 if (tool->ordered_events)
536 tool->finished_round = perf_event__process_finished_round;
537 else
538 tool->finished_round = process_finished_round_stub;
539 }
540 if (tool->id_index == NULL)
541 tool->id_index = process_event_op2_stub;
542 if (tool->auxtrace_info == NULL)
543 tool->auxtrace_info = process_event_op2_stub;
544 if (tool->auxtrace == NULL)
545 tool->auxtrace = process_event_auxtrace_stub;
546 if (tool->auxtrace_error == NULL)
547 tool->auxtrace_error = process_event_op2_stub;
548 if (tool->thread_map == NULL)
549 tool->thread_map = process_event_thread_map_stub;
550 if (tool->cpu_map == NULL)
551 tool->cpu_map = process_event_cpu_map_stub;
552 if (tool->stat_config == NULL)
553 tool->stat_config = process_event_stat_config_stub;
554 if (tool->stat == NULL)
555 tool->stat = process_stat_stub;
556 if (tool->stat_round == NULL)
557 tool->stat_round = process_stat_round_stub;
558 if (tool->time_conv == NULL)
559 tool->time_conv = process_event_time_conv_stub;
560 if (tool->feature == NULL)
561 tool->feature = process_event_op2_stub;
562 if (tool->compressed == NULL)
563 tool->compressed = perf_session__process_compressed_event;
564 if (tool->finished_init == NULL)
565 tool->finished_init = process_event_op2_stub;
566}
567
568static void swap_sample_id_all(union perf_event *event, void *data)
569{
570 void *end = (void *) event + event->header.size;
571 int size = end - data;
572
573 BUG_ON(size % sizeof(u64));
574 mem_bswap_64(data, size);
575}
576
577static void perf_event__all64_swap(union perf_event *event,
578 bool sample_id_all __maybe_unused)
579{
580 struct perf_event_header *hdr = &event->header;
581 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
582}
583
584static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
585{
586 event->comm.pid = bswap_32(event->comm.pid);
587 event->comm.tid = bswap_32(event->comm.tid);
588
589 if (sample_id_all) {
590 void *data = &event->comm.comm;
591
592 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
593 swap_sample_id_all(event, data);
594 }
595}
596
597static void perf_event__mmap_swap(union perf_event *event,
598 bool sample_id_all)
599{
600 event->mmap.pid = bswap_32(event->mmap.pid);
601 event->mmap.tid = bswap_32(event->mmap.tid);
602 event->mmap.start = bswap_64(event->mmap.start);
603 event->mmap.len = bswap_64(event->mmap.len);
604 event->mmap.pgoff = bswap_64(event->mmap.pgoff);
605
606 if (sample_id_all) {
607 void *data = &event->mmap.filename;
608
609 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
610 swap_sample_id_all(event, data);
611 }
612}
613
614static void perf_event__mmap2_swap(union perf_event *event,
615 bool sample_id_all)
616{
617 event->mmap2.pid = bswap_32(event->mmap2.pid);
618 event->mmap2.tid = bswap_32(event->mmap2.tid);
619 event->mmap2.start = bswap_64(event->mmap2.start);
620 event->mmap2.len = bswap_64(event->mmap2.len);
621 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
622
623 if (!(event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID)) {
624 event->mmap2.maj = bswap_32(event->mmap2.maj);
625 event->mmap2.min = bswap_32(event->mmap2.min);
626 event->mmap2.ino = bswap_64(event->mmap2.ino);
627 event->mmap2.ino_generation = bswap_64(event->mmap2.ino_generation);
628 }
629
630 if (sample_id_all) {
631 void *data = &event->mmap2.filename;
632
633 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
634 swap_sample_id_all(event, data);
635 }
636}
637static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
638{
639 event->fork.pid = bswap_32(event->fork.pid);
640 event->fork.tid = bswap_32(event->fork.tid);
641 event->fork.ppid = bswap_32(event->fork.ppid);
642 event->fork.ptid = bswap_32(event->fork.ptid);
643 event->fork.time = bswap_64(event->fork.time);
644
645 if (sample_id_all)
646 swap_sample_id_all(event, &event->fork + 1);
647}
648
649static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
650{
651 event->read.pid = bswap_32(event->read.pid);
652 event->read.tid = bswap_32(event->read.tid);
653 event->read.value = bswap_64(event->read.value);
654 event->read.time_enabled = bswap_64(event->read.time_enabled);
655 event->read.time_running = bswap_64(event->read.time_running);
656 event->read.id = bswap_64(event->read.id);
657
658 if (sample_id_all)
659 swap_sample_id_all(event, &event->read + 1);
660}
661
662static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
663{
664 event->aux.aux_offset = bswap_64(event->aux.aux_offset);
665 event->aux.aux_size = bswap_64(event->aux.aux_size);
666 event->aux.flags = bswap_64(event->aux.flags);
667
668 if (sample_id_all)
669 swap_sample_id_all(event, &event->aux + 1);
670}
671
672static void perf_event__itrace_start_swap(union perf_event *event,
673 bool sample_id_all)
674{
675 event->itrace_start.pid = bswap_32(event->itrace_start.pid);
676 event->itrace_start.tid = bswap_32(event->itrace_start.tid);
677
678 if (sample_id_all)
679 swap_sample_id_all(event, &event->itrace_start + 1);
680}
681
682static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
683{
684 if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
685 event->context_switch.next_prev_pid =
686 bswap_32(event->context_switch.next_prev_pid);
687 event->context_switch.next_prev_tid =
688 bswap_32(event->context_switch.next_prev_tid);
689 }
690
691 if (sample_id_all)
692 swap_sample_id_all(event, &event->context_switch + 1);
693}
694
695static void perf_event__text_poke_swap(union perf_event *event, bool sample_id_all)
696{
697 event->text_poke.addr = bswap_64(event->text_poke.addr);
698 event->text_poke.old_len = bswap_16(event->text_poke.old_len);
699 event->text_poke.new_len = bswap_16(event->text_poke.new_len);
700
701 if (sample_id_all) {
702 size_t len = sizeof(event->text_poke.old_len) +
703 sizeof(event->text_poke.new_len) +
704 event->text_poke.old_len +
705 event->text_poke.new_len;
706 void *data = &event->text_poke.old_len;
707
708 data += PERF_ALIGN(len, sizeof(u64));
709 swap_sample_id_all(event, data);
710 }
711}
712
713static void perf_event__throttle_swap(union perf_event *event,
714 bool sample_id_all)
715{
716 event->throttle.time = bswap_64(event->throttle.time);
717 event->throttle.id = bswap_64(event->throttle.id);
718 event->throttle.stream_id = bswap_64(event->throttle.stream_id);
719
720 if (sample_id_all)
721 swap_sample_id_all(event, &event->throttle + 1);
722}
723
724static void perf_event__namespaces_swap(union perf_event *event,
725 bool sample_id_all)
726{
727 u64 i;
728
729 event->namespaces.pid = bswap_32(event->namespaces.pid);
730 event->namespaces.tid = bswap_32(event->namespaces.tid);
731 event->namespaces.nr_namespaces = bswap_64(event->namespaces.nr_namespaces);
732
733 for (i = 0; i < event->namespaces.nr_namespaces; i++) {
734 struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
735
736 ns->dev = bswap_64(ns->dev);
737 ns->ino = bswap_64(ns->ino);
738 }
739
740 if (sample_id_all)
741 swap_sample_id_all(event, &event->namespaces.link_info[i]);
742}
743
744static void perf_event__cgroup_swap(union perf_event *event, bool sample_id_all)
745{
746 event->cgroup.id = bswap_64(event->cgroup.id);
747
748 if (sample_id_all) {
749 void *data = &event->cgroup.path;
750
751 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
752 swap_sample_id_all(event, data);
753 }
754}
755
756static u8 revbyte(u8 b)
757{
758 int rev = (b >> 4) | ((b & 0xf) << 4);
759 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
760 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
761 return (u8) rev;
762}
763
764/*
765 * XXX this is hack in attempt to carry flags bitfield
766 * through endian village. ABI says:
767 *
768 * Bit-fields are allocated from right to left (least to most significant)
769 * on little-endian implementations and from left to right (most to least
770 * significant) on big-endian implementations.
771 *
772 * The above seems to be byte specific, so we need to reverse each
773 * byte of the bitfield. 'Internet' also says this might be implementation
774 * specific and we probably need proper fix and carry perf_event_attr
775 * bitfield flags in separate data file FEAT_ section. Thought this seems
776 * to work for now.
777 */
778static void swap_bitfield(u8 *p, unsigned len)
779{
780 unsigned i;
781
782 for (i = 0; i < len; i++) {
783 *p = revbyte(*p);
784 p++;
785 }
786}
787
788/* exported for swapping attributes in file header */
789void perf_event__attr_swap(struct perf_event_attr *attr)
790{
791 attr->type = bswap_32(attr->type);
792 attr->size = bswap_32(attr->size);
793
794#define bswap_safe(f, n) \
795 (attr->size > (offsetof(struct perf_event_attr, f) + \
796 sizeof(attr->f) * (n)))
797#define bswap_field(f, sz) \
798do { \
799 if (bswap_safe(f, 0)) \
800 attr->f = bswap_##sz(attr->f); \
801} while(0)
802#define bswap_field_16(f) bswap_field(f, 16)
803#define bswap_field_32(f) bswap_field(f, 32)
804#define bswap_field_64(f) bswap_field(f, 64)
805
806 bswap_field_64(config);
807 bswap_field_64(sample_period);
808 bswap_field_64(sample_type);
809 bswap_field_64(read_format);
810 bswap_field_32(wakeup_events);
811 bswap_field_32(bp_type);
812 bswap_field_64(bp_addr);
813 bswap_field_64(bp_len);
814 bswap_field_64(branch_sample_type);
815 bswap_field_64(sample_regs_user);
816 bswap_field_32(sample_stack_user);
817 bswap_field_32(aux_watermark);
818 bswap_field_16(sample_max_stack);
819 bswap_field_32(aux_sample_size);
820
821 /*
822 * After read_format are bitfields. Check read_format because
823 * we are unable to use offsetof on bitfield.
824 */
825 if (bswap_safe(read_format, 1))
826 swap_bitfield((u8 *) (&attr->read_format + 1),
827 sizeof(u64));
828#undef bswap_field_64
829#undef bswap_field_32
830#undef bswap_field
831#undef bswap_safe
832}
833
834static void perf_event__hdr_attr_swap(union perf_event *event,
835 bool sample_id_all __maybe_unused)
836{
837 size_t size;
838
839 perf_event__attr_swap(&event->attr.attr);
840
841 size = event->header.size;
842 size -= (void *)&event->attr.id - (void *)event;
843 mem_bswap_64(event->attr.id, size);
844}
845
846static void perf_event__event_update_swap(union perf_event *event,
847 bool sample_id_all __maybe_unused)
848{
849 event->event_update.type = bswap_64(event->event_update.type);
850 event->event_update.id = bswap_64(event->event_update.id);
851}
852
853static void perf_event__event_type_swap(union perf_event *event,
854 bool sample_id_all __maybe_unused)
855{
856 event->event_type.event_type.event_id =
857 bswap_64(event->event_type.event_type.event_id);
858}
859
860static void perf_event__tracing_data_swap(union perf_event *event,
861 bool sample_id_all __maybe_unused)
862{
863 event->tracing_data.size = bswap_32(event->tracing_data.size);
864}
865
866static void perf_event__auxtrace_info_swap(union perf_event *event,
867 bool sample_id_all __maybe_unused)
868{
869 size_t size;
870
871 event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
872
873 size = event->header.size;
874 size -= (void *)&event->auxtrace_info.priv - (void *)event;
875 mem_bswap_64(event->auxtrace_info.priv, size);
876}
877
878static void perf_event__auxtrace_swap(union perf_event *event,
879 bool sample_id_all __maybe_unused)
880{
881 event->auxtrace.size = bswap_64(event->auxtrace.size);
882 event->auxtrace.offset = bswap_64(event->auxtrace.offset);
883 event->auxtrace.reference = bswap_64(event->auxtrace.reference);
884 event->auxtrace.idx = bswap_32(event->auxtrace.idx);
885 event->auxtrace.tid = bswap_32(event->auxtrace.tid);
886 event->auxtrace.cpu = bswap_32(event->auxtrace.cpu);
887}
888
889static void perf_event__auxtrace_error_swap(union perf_event *event,
890 bool sample_id_all __maybe_unused)
891{
892 event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
893 event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
894 event->auxtrace_error.cpu = bswap_32(event->auxtrace_error.cpu);
895 event->auxtrace_error.pid = bswap_32(event->auxtrace_error.pid);
896 event->auxtrace_error.tid = bswap_32(event->auxtrace_error.tid);
897 event->auxtrace_error.fmt = bswap_32(event->auxtrace_error.fmt);
898 event->auxtrace_error.ip = bswap_64(event->auxtrace_error.ip);
899 if (event->auxtrace_error.fmt)
900 event->auxtrace_error.time = bswap_64(event->auxtrace_error.time);
901 if (event->auxtrace_error.fmt >= 2) {
902 event->auxtrace_error.machine_pid = bswap_32(event->auxtrace_error.machine_pid);
903 event->auxtrace_error.vcpu = bswap_32(event->auxtrace_error.vcpu);
904 }
905}
906
907static void perf_event__thread_map_swap(union perf_event *event,
908 bool sample_id_all __maybe_unused)
909{
910 unsigned i;
911
912 event->thread_map.nr = bswap_64(event->thread_map.nr);
913
914 for (i = 0; i < event->thread_map.nr; i++)
915 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
916}
917
918static void perf_event__cpu_map_swap(union perf_event *event,
919 bool sample_id_all __maybe_unused)
920{
921 struct perf_record_cpu_map_data *data = &event->cpu_map.data;
922
923 data->type = bswap_16(data->type);
924
925 switch (data->type) {
926 case PERF_CPU_MAP__CPUS:
927 data->cpus_data.nr = bswap_16(data->cpus_data.nr);
928
929 for (unsigned i = 0; i < data->cpus_data.nr; i++)
930 data->cpus_data.cpu[i] = bswap_16(data->cpus_data.cpu[i]);
931 break;
932 case PERF_CPU_MAP__MASK:
933 data->mask32_data.long_size = bswap_16(data->mask32_data.long_size);
934
935 switch (data->mask32_data.long_size) {
936 case 4:
937 data->mask32_data.nr = bswap_16(data->mask32_data.nr);
938 for (unsigned i = 0; i < data->mask32_data.nr; i++)
939 data->mask32_data.mask[i] = bswap_32(data->mask32_data.mask[i]);
940 break;
941 case 8:
942 data->mask64_data.nr = bswap_16(data->mask64_data.nr);
943 for (unsigned i = 0; i < data->mask64_data.nr; i++)
944 data->mask64_data.mask[i] = bswap_64(data->mask64_data.mask[i]);
945 break;
946 default:
947 pr_err("cpu_map swap: unsupported long size\n");
948 }
949 break;
950 case PERF_CPU_MAP__RANGE_CPUS:
951 data->range_cpu_data.start_cpu = bswap_16(data->range_cpu_data.start_cpu);
952 data->range_cpu_data.end_cpu = bswap_16(data->range_cpu_data.end_cpu);
953 break;
954 default:
955 break;
956 }
957}
958
959static void perf_event__stat_config_swap(union perf_event *event,
960 bool sample_id_all __maybe_unused)
961{
962 u64 size;
963
964 size = bswap_64(event->stat_config.nr) * sizeof(event->stat_config.data[0]);
965 size += 1; /* nr item itself */
966 mem_bswap_64(&event->stat_config.nr, size);
967}
968
969static void perf_event__stat_swap(union perf_event *event,
970 bool sample_id_all __maybe_unused)
971{
972 event->stat.id = bswap_64(event->stat.id);
973 event->stat.thread = bswap_32(event->stat.thread);
974 event->stat.cpu = bswap_32(event->stat.cpu);
975 event->stat.val = bswap_64(event->stat.val);
976 event->stat.ena = bswap_64(event->stat.ena);
977 event->stat.run = bswap_64(event->stat.run);
978}
979
980static void perf_event__stat_round_swap(union perf_event *event,
981 bool sample_id_all __maybe_unused)
982{
983 event->stat_round.type = bswap_64(event->stat_round.type);
984 event->stat_round.time = bswap_64(event->stat_round.time);
985}
986
987static void perf_event__time_conv_swap(union perf_event *event,
988 bool sample_id_all __maybe_unused)
989{
990 event->time_conv.time_shift = bswap_64(event->time_conv.time_shift);
991 event->time_conv.time_mult = bswap_64(event->time_conv.time_mult);
992 event->time_conv.time_zero = bswap_64(event->time_conv.time_zero);
993
994 if (event_contains(event->time_conv, time_cycles)) {
995 event->time_conv.time_cycles = bswap_64(event->time_conv.time_cycles);
996 event->time_conv.time_mask = bswap_64(event->time_conv.time_mask);
997 }
998}
999
1000typedef void (*perf_event__swap_op)(union perf_event *event,
1001 bool sample_id_all);
1002
1003static perf_event__swap_op perf_event__swap_ops[] = {
1004 [PERF_RECORD_MMAP] = perf_event__mmap_swap,
1005 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap,
1006 [PERF_RECORD_COMM] = perf_event__comm_swap,
1007 [PERF_RECORD_FORK] = perf_event__task_swap,
1008 [PERF_RECORD_EXIT] = perf_event__task_swap,
1009 [PERF_RECORD_LOST] = perf_event__all64_swap,
1010 [PERF_RECORD_READ] = perf_event__read_swap,
1011 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap,
1012 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap,
1013 [PERF_RECORD_SAMPLE] = perf_event__all64_swap,
1014 [PERF_RECORD_AUX] = perf_event__aux_swap,
1015 [PERF_RECORD_ITRACE_START] = perf_event__itrace_start_swap,
1016 [PERF_RECORD_LOST_SAMPLES] = perf_event__all64_swap,
1017 [PERF_RECORD_SWITCH] = perf_event__switch_swap,
1018 [PERF_RECORD_SWITCH_CPU_WIDE] = perf_event__switch_swap,
1019 [PERF_RECORD_NAMESPACES] = perf_event__namespaces_swap,
1020 [PERF_RECORD_CGROUP] = perf_event__cgroup_swap,
1021 [PERF_RECORD_TEXT_POKE] = perf_event__text_poke_swap,
1022 [PERF_RECORD_AUX_OUTPUT_HW_ID] = perf_event__all64_swap,
1023 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap,
1024 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap,
1025 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
1026 [PERF_RECORD_HEADER_BUILD_ID] = NULL,
1027 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap,
1028 [PERF_RECORD_AUXTRACE_INFO] = perf_event__auxtrace_info_swap,
1029 [PERF_RECORD_AUXTRACE] = perf_event__auxtrace_swap,
1030 [PERF_RECORD_AUXTRACE_ERROR] = perf_event__auxtrace_error_swap,
1031 [PERF_RECORD_THREAD_MAP] = perf_event__thread_map_swap,
1032 [PERF_RECORD_CPU_MAP] = perf_event__cpu_map_swap,
1033 [PERF_RECORD_STAT_CONFIG] = perf_event__stat_config_swap,
1034 [PERF_RECORD_STAT] = perf_event__stat_swap,
1035 [PERF_RECORD_STAT_ROUND] = perf_event__stat_round_swap,
1036 [PERF_RECORD_EVENT_UPDATE] = perf_event__event_update_swap,
1037 [PERF_RECORD_TIME_CONV] = perf_event__time_conv_swap,
1038 [PERF_RECORD_HEADER_MAX] = NULL,
1039};
1040
1041/*
1042 * When perf record finishes a pass on every buffers, it records this pseudo
1043 * event.
1044 * We record the max timestamp t found in the pass n.
1045 * Assuming these timestamps are monotonic across cpus, we know that if
1046 * a buffer still has events with timestamps below t, they will be all
1047 * available and then read in the pass n + 1.
1048 * Hence when we start to read the pass n + 2, we can safely flush every
1049 * events with timestamps below t.
1050 *
1051 * ============ PASS n =================
1052 * CPU 0 | CPU 1
1053 * |
1054 * cnt1 timestamps | cnt2 timestamps
1055 * 1 | 2
1056 * 2 | 3
1057 * - | 4 <--- max recorded
1058 *
1059 * ============ PASS n + 1 ==============
1060 * CPU 0 | CPU 1
1061 * |
1062 * cnt1 timestamps | cnt2 timestamps
1063 * 3 | 5
1064 * 4 | 6
1065 * 5 | 7 <---- max recorded
1066 *
1067 * Flush every events below timestamp 4
1068 *
1069 * ============ PASS n + 2 ==============
1070 * CPU 0 | CPU 1
1071 * |
1072 * cnt1 timestamps | cnt2 timestamps
1073 * 6 | 8
1074 * 7 | 9
1075 * - | 10
1076 *
1077 * Flush every events below timestamp 7
1078 * etc...
1079 */
1080int perf_event__process_finished_round(struct perf_tool *tool __maybe_unused,
1081 union perf_event *event __maybe_unused,
1082 struct ordered_events *oe)
1083{
1084 if (dump_trace)
1085 fprintf(stdout, "\n");
1086 return ordered_events__flush(oe, OE_FLUSH__ROUND);
1087}
1088
1089int perf_session__queue_event(struct perf_session *s, union perf_event *event,
1090 u64 timestamp, u64 file_offset, const char *file_path)
1091{
1092 return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset, file_path);
1093}
1094
1095static void callchain__lbr_callstack_printf(struct perf_sample *sample)
1096{
1097 struct ip_callchain *callchain = sample->callchain;
1098 struct branch_stack *lbr_stack = sample->branch_stack;
1099 struct branch_entry *entries = perf_sample__branch_entries(sample);
1100 u64 kernel_callchain_nr = callchain->nr;
1101 unsigned int i;
1102
1103 for (i = 0; i < kernel_callchain_nr; i++) {
1104 if (callchain->ips[i] == PERF_CONTEXT_USER)
1105 break;
1106 }
1107
1108 if ((i != kernel_callchain_nr) && lbr_stack->nr) {
1109 u64 total_nr;
1110 /*
1111 * LBR callstack can only get user call chain,
1112 * i is kernel call chain number,
1113 * 1 is PERF_CONTEXT_USER.
1114 *
1115 * The user call chain is stored in LBR registers.
1116 * LBR are pair registers. The caller is stored
1117 * in "from" register, while the callee is stored
1118 * in "to" register.
1119 * For example, there is a call stack
1120 * "A"->"B"->"C"->"D".
1121 * The LBR registers will be recorded like
1122 * "C"->"D", "B"->"C", "A"->"B".
1123 * So only the first "to" register and all "from"
1124 * registers are needed to construct the whole stack.
1125 */
1126 total_nr = i + 1 + lbr_stack->nr + 1;
1127 kernel_callchain_nr = i + 1;
1128
1129 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
1130
1131 for (i = 0; i < kernel_callchain_nr; i++)
1132 printf("..... %2d: %016" PRIx64 "\n",
1133 i, callchain->ips[i]);
1134
1135 printf("..... %2d: %016" PRIx64 "\n",
1136 (int)(kernel_callchain_nr), entries[0].to);
1137 for (i = 0; i < lbr_stack->nr; i++)
1138 printf("..... %2d: %016" PRIx64 "\n",
1139 (int)(i + kernel_callchain_nr + 1), entries[i].from);
1140 }
1141}
1142
1143static void callchain__printf(struct evsel *evsel,
1144 struct perf_sample *sample)
1145{
1146 unsigned int i;
1147 struct ip_callchain *callchain = sample->callchain;
1148
1149 if (evsel__has_branch_callstack(evsel))
1150 callchain__lbr_callstack_printf(sample);
1151
1152 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
1153
1154 for (i = 0; i < callchain->nr; i++)
1155 printf("..... %2d: %016" PRIx64 "\n",
1156 i, callchain->ips[i]);
1157}
1158
1159static void branch_stack__printf(struct perf_sample *sample, bool callstack)
1160{
1161 struct branch_entry *entries = perf_sample__branch_entries(sample);
1162 uint64_t i;
1163
1164 if (!callstack) {
1165 printf("%s: nr:%" PRIu64 "\n", "... branch stack", sample->branch_stack->nr);
1166 } else {
1167 /* the reason of adding 1 to nr is because after expanding
1168 * branch stack it generates nr + 1 callstack records. e.g.,
1169 * B()->C()
1170 * A()->B()
1171 * the final callstack should be:
1172 * C()
1173 * B()
1174 * A()
1175 */
1176 printf("%s: nr:%" PRIu64 "\n", "... branch callstack", sample->branch_stack->nr+1);
1177 }
1178
1179 for (i = 0; i < sample->branch_stack->nr; i++) {
1180 struct branch_entry *e = &entries[i];
1181
1182 if (!callstack) {
1183 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x %s\n",
1184 i, e->from, e->to,
1185 (unsigned short)e->flags.cycles,
1186 e->flags.mispred ? "M" : " ",
1187 e->flags.predicted ? "P" : " ",
1188 e->flags.abort ? "A" : " ",
1189 e->flags.in_tx ? "T" : " ",
1190 (unsigned)e->flags.reserved,
1191 get_branch_type(e));
1192 } else {
1193 if (i == 0) {
1194 printf("..... %2"PRIu64": %016" PRIx64 "\n"
1195 "..... %2"PRIu64": %016" PRIx64 "\n",
1196 i, e->to, i+1, e->from);
1197 } else {
1198 printf("..... %2"PRIu64": %016" PRIx64 "\n", i+1, e->from);
1199 }
1200 }
1201 }
1202}
1203
1204static void regs_dump__printf(u64 mask, u64 *regs, const char *arch)
1205{
1206 unsigned rid, i = 0;
1207
1208 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
1209 u64 val = regs[i++];
1210
1211 printf(".... %-5s 0x%016" PRIx64 "\n",
1212 perf_reg_name(rid, arch), val);
1213 }
1214}
1215
1216static const char *regs_abi[] = {
1217 [PERF_SAMPLE_REGS_ABI_NONE] = "none",
1218 [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
1219 [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
1220};
1221
1222static inline const char *regs_dump_abi(struct regs_dump *d)
1223{
1224 if (d->abi > PERF_SAMPLE_REGS_ABI_64)
1225 return "unknown";
1226
1227 return regs_abi[d->abi];
1228}
1229
1230static void regs__printf(const char *type, struct regs_dump *regs, const char *arch)
1231{
1232 u64 mask = regs->mask;
1233
1234 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
1235 type,
1236 mask,
1237 regs_dump_abi(regs));
1238
1239 regs_dump__printf(mask, regs->regs, arch);
1240}
1241
1242static void regs_user__printf(struct perf_sample *sample, const char *arch)
1243{
1244 struct regs_dump *user_regs = &sample->user_regs;
1245
1246 if (user_regs->regs)
1247 regs__printf("user", user_regs, arch);
1248}
1249
1250static void regs_intr__printf(struct perf_sample *sample, const char *arch)
1251{
1252 struct regs_dump *intr_regs = &sample->intr_regs;
1253
1254 if (intr_regs->regs)
1255 regs__printf("intr", intr_regs, arch);
1256}
1257
1258static void stack_user__printf(struct stack_dump *dump)
1259{
1260 printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1261 dump->size, dump->offset);
1262}
1263
1264static void evlist__print_tstamp(struct evlist *evlist, union perf_event *event, struct perf_sample *sample)
1265{
1266 u64 sample_type = __evlist__combined_sample_type(evlist);
1267
1268 if (event->header.type != PERF_RECORD_SAMPLE &&
1269 !evlist__sample_id_all(evlist)) {
1270 fputs("-1 -1 ", stdout);
1271 return;
1272 }
1273
1274 if ((sample_type & PERF_SAMPLE_CPU))
1275 printf("%u ", sample->cpu);
1276
1277 if (sample_type & PERF_SAMPLE_TIME)
1278 printf("%" PRIu64 " ", sample->time);
1279}
1280
1281static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1282{
1283 printf("... sample_read:\n");
1284
1285 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1286 printf("...... time enabled %016" PRIx64 "\n",
1287 sample->read.time_enabled);
1288
1289 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1290 printf("...... time running %016" PRIx64 "\n",
1291 sample->read.time_running);
1292
1293 if (read_format & PERF_FORMAT_GROUP) {
1294 struct sample_read_value *value = sample->read.group.values;
1295
1296 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1297
1298 sample_read_group__for_each(value, sample->read.group.nr, read_format) {
1299 printf("..... id %016" PRIx64
1300 ", value %016" PRIx64,
1301 value->id, value->value);
1302 if (read_format & PERF_FORMAT_LOST)
1303 printf(", lost %" PRIu64, value->lost);
1304 printf("\n");
1305 }
1306 } else {
1307 printf("..... id %016" PRIx64 ", value %016" PRIx64,
1308 sample->read.one.id, sample->read.one.value);
1309 if (read_format & PERF_FORMAT_LOST)
1310 printf(", lost %" PRIu64, sample->read.one.lost);
1311 printf("\n");
1312 }
1313}
1314
1315static void dump_event(struct evlist *evlist, union perf_event *event,
1316 u64 file_offset, struct perf_sample *sample,
1317 const char *file_path)
1318{
1319 if (!dump_trace)
1320 return;
1321
1322 printf("\n%#" PRIx64 "@%s [%#x]: event: %d\n",
1323 file_offset, file_path, event->header.size, event->header.type);
1324
1325 trace_event(event);
1326 if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw)
1327 evlist->trace_event_sample_raw(evlist, event, sample);
1328
1329 if (sample)
1330 evlist__print_tstamp(evlist, event, sample);
1331
1332 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1333 event->header.size, perf_event__name(event->header.type));
1334}
1335
1336char *get_page_size_name(u64 size, char *str)
1337{
1338 if (!size || !unit_number__scnprintf(str, PAGE_SIZE_NAME_LEN, size))
1339 snprintf(str, PAGE_SIZE_NAME_LEN, "%s", "N/A");
1340
1341 return str;
1342}
1343
1344static void dump_sample(struct evsel *evsel, union perf_event *event,
1345 struct perf_sample *sample, const char *arch)
1346{
1347 u64 sample_type;
1348 char str[PAGE_SIZE_NAME_LEN];
1349
1350 if (!dump_trace)
1351 return;
1352
1353 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1354 event->header.misc, sample->pid, sample->tid, sample->ip,
1355 sample->period, sample->addr);
1356
1357 sample_type = evsel->core.attr.sample_type;
1358
1359 if (evsel__has_callchain(evsel))
1360 callchain__printf(evsel, sample);
1361
1362 if (evsel__has_br_stack(evsel))
1363 branch_stack__printf(sample, evsel__has_branch_callstack(evsel));
1364
1365 if (sample_type & PERF_SAMPLE_REGS_USER)
1366 regs_user__printf(sample, arch);
1367
1368 if (sample_type & PERF_SAMPLE_REGS_INTR)
1369 regs_intr__printf(sample, arch);
1370
1371 if (sample_type & PERF_SAMPLE_STACK_USER)
1372 stack_user__printf(&sample->user_stack);
1373
1374 if (sample_type & PERF_SAMPLE_WEIGHT_TYPE) {
1375 printf("... weight: %" PRIu64 "", sample->weight);
1376 if (sample_type & PERF_SAMPLE_WEIGHT_STRUCT) {
1377 printf(",0x%"PRIx16"", sample->ins_lat);
1378 printf(",0x%"PRIx16"", sample->p_stage_cyc);
1379 }
1380 printf("\n");
1381 }
1382
1383 if (sample_type & PERF_SAMPLE_DATA_SRC)
1384 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1385
1386 if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1387 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1388
1389 if (sample_type & PERF_SAMPLE_DATA_PAGE_SIZE)
1390 printf(" .. data page size: %s\n", get_page_size_name(sample->data_page_size, str));
1391
1392 if (sample_type & PERF_SAMPLE_CODE_PAGE_SIZE)
1393 printf(" .. code page size: %s\n", get_page_size_name(sample->code_page_size, str));
1394
1395 if (sample_type & PERF_SAMPLE_TRANSACTION)
1396 printf("... transaction: %" PRIx64 "\n", sample->transaction);
1397
1398 if (sample_type & PERF_SAMPLE_READ)
1399 sample_read__printf(sample, evsel->core.attr.read_format);
1400}
1401
1402static void dump_read(struct evsel *evsel, union perf_event *event)
1403{
1404 struct perf_record_read *read_event = &event->read;
1405 u64 read_format;
1406
1407 if (!dump_trace)
1408 return;
1409
1410 printf(": %d %d %s %" PRI_lu64 "\n", event->read.pid, event->read.tid,
1411 evsel__name(evsel), event->read.value);
1412
1413 if (!evsel)
1414 return;
1415
1416 read_format = evsel->core.attr.read_format;
1417
1418 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1419 printf("... time enabled : %" PRI_lu64 "\n", read_event->time_enabled);
1420
1421 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1422 printf("... time running : %" PRI_lu64 "\n", read_event->time_running);
1423
1424 if (read_format & PERF_FORMAT_ID)
1425 printf("... id : %" PRI_lu64 "\n", read_event->id);
1426
1427 if (read_format & PERF_FORMAT_LOST)
1428 printf("... lost : %" PRI_lu64 "\n", read_event->lost);
1429}
1430
1431static struct machine *machines__find_for_cpumode(struct machines *machines,
1432 union perf_event *event,
1433 struct perf_sample *sample)
1434{
1435 if (perf_guest &&
1436 ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1437 (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1438 u32 pid;
1439
1440 if (sample->machine_pid)
1441 pid = sample->machine_pid;
1442 else if (event->header.type == PERF_RECORD_MMAP
1443 || event->header.type == PERF_RECORD_MMAP2)
1444 pid = event->mmap.pid;
1445 else
1446 pid = sample->pid;
1447
1448 /*
1449 * Guest code machine is created as needed and does not use
1450 * DEFAULT_GUEST_KERNEL_ID.
1451 */
1452 if (symbol_conf.guest_code)
1453 return machines__findnew(machines, pid);
1454
1455 return machines__find_guest(machines, pid);
1456 }
1457
1458 return &machines->host;
1459}
1460
1461static int deliver_sample_value(struct evlist *evlist,
1462 struct perf_tool *tool,
1463 union perf_event *event,
1464 struct perf_sample *sample,
1465 struct sample_read_value *v,
1466 struct machine *machine)
1467{
1468 struct perf_sample_id *sid = evlist__id2sid(evlist, v->id);
1469 struct evsel *evsel;
1470
1471 if (sid) {
1472 sample->id = v->id;
1473 sample->period = v->value - sid->period;
1474 sid->period = v->value;
1475 }
1476
1477 if (!sid || sid->evsel == NULL) {
1478 ++evlist->stats.nr_unknown_id;
1479 return 0;
1480 }
1481
1482 /*
1483 * There's no reason to deliver sample
1484 * for zero period, bail out.
1485 */
1486 if (!sample->period)
1487 return 0;
1488
1489 evsel = container_of(sid->evsel, struct evsel, core);
1490 return tool->sample(tool, event, sample, evsel, machine);
1491}
1492
1493static int deliver_sample_group(struct evlist *evlist,
1494 struct perf_tool *tool,
1495 union perf_event *event,
1496 struct perf_sample *sample,
1497 struct machine *machine,
1498 u64 read_format)
1499{
1500 int ret = -EINVAL;
1501 struct sample_read_value *v = sample->read.group.values;
1502
1503 sample_read_group__for_each(v, sample->read.group.nr, read_format) {
1504 ret = deliver_sample_value(evlist, tool, event, sample, v,
1505 machine);
1506 if (ret)
1507 break;
1508 }
1509
1510 return ret;
1511}
1512
1513static int evlist__deliver_sample(struct evlist *evlist, struct perf_tool *tool,
1514 union perf_event *event, struct perf_sample *sample,
1515 struct evsel *evsel, struct machine *machine)
1516{
1517 /* We know evsel != NULL. */
1518 u64 sample_type = evsel->core.attr.sample_type;
1519 u64 read_format = evsel->core.attr.read_format;
1520
1521 /* Standard sample delivery. */
1522 if (!(sample_type & PERF_SAMPLE_READ))
1523 return tool->sample(tool, event, sample, evsel, machine);
1524
1525 /* For PERF_SAMPLE_READ we have either single or group mode. */
1526 if (read_format & PERF_FORMAT_GROUP)
1527 return deliver_sample_group(evlist, tool, event, sample,
1528 machine, read_format);
1529 else
1530 return deliver_sample_value(evlist, tool, event, sample,
1531 &sample->read.one, machine);
1532}
1533
1534static int machines__deliver_event(struct machines *machines,
1535 struct evlist *evlist,
1536 union perf_event *event,
1537 struct perf_sample *sample,
1538 struct perf_tool *tool, u64 file_offset,
1539 const char *file_path)
1540{
1541 struct evsel *evsel;
1542 struct machine *machine;
1543
1544 dump_event(evlist, event, file_offset, sample, file_path);
1545
1546 evsel = evlist__id2evsel(evlist, sample->id);
1547
1548 machine = machines__find_for_cpumode(machines, event, sample);
1549
1550 switch (event->header.type) {
1551 case PERF_RECORD_SAMPLE:
1552 if (evsel == NULL) {
1553 ++evlist->stats.nr_unknown_id;
1554 return 0;
1555 }
1556 if (machine == NULL) {
1557 ++evlist->stats.nr_unprocessable_samples;
1558 dump_sample(evsel, event, sample, perf_env__arch(NULL));
1559 return 0;
1560 }
1561 dump_sample(evsel, event, sample, perf_env__arch(machine->env));
1562 return evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1563 case PERF_RECORD_MMAP:
1564 return tool->mmap(tool, event, sample, machine);
1565 case PERF_RECORD_MMAP2:
1566 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1567 ++evlist->stats.nr_proc_map_timeout;
1568 return tool->mmap2(tool, event, sample, machine);
1569 case PERF_RECORD_COMM:
1570 return tool->comm(tool, event, sample, machine);
1571 case PERF_RECORD_NAMESPACES:
1572 return tool->namespaces(tool, event, sample, machine);
1573 case PERF_RECORD_CGROUP:
1574 return tool->cgroup(tool, event, sample, machine);
1575 case PERF_RECORD_FORK:
1576 return tool->fork(tool, event, sample, machine);
1577 case PERF_RECORD_EXIT:
1578 return tool->exit(tool, event, sample, machine);
1579 case PERF_RECORD_LOST:
1580 if (tool->lost == perf_event__process_lost)
1581 evlist->stats.total_lost += event->lost.lost;
1582 return tool->lost(tool, event, sample, machine);
1583 case PERF_RECORD_LOST_SAMPLES:
1584 if (tool->lost_samples == perf_event__process_lost_samples)
1585 evlist->stats.total_lost_samples += event->lost_samples.lost;
1586 return tool->lost_samples(tool, event, sample, machine);
1587 case PERF_RECORD_READ:
1588 dump_read(evsel, event);
1589 return tool->read(tool, event, sample, evsel, machine);
1590 case PERF_RECORD_THROTTLE:
1591 return tool->throttle(tool, event, sample, machine);
1592 case PERF_RECORD_UNTHROTTLE:
1593 return tool->unthrottle(tool, event, sample, machine);
1594 case PERF_RECORD_AUX:
1595 if (tool->aux == perf_event__process_aux) {
1596 if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1597 evlist->stats.total_aux_lost += 1;
1598 if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1599 evlist->stats.total_aux_partial += 1;
1600 if (event->aux.flags & PERF_AUX_FLAG_COLLISION)
1601 evlist->stats.total_aux_collision += 1;
1602 }
1603 return tool->aux(tool, event, sample, machine);
1604 case PERF_RECORD_ITRACE_START:
1605 return tool->itrace_start(tool, event, sample, machine);
1606 case PERF_RECORD_SWITCH:
1607 case PERF_RECORD_SWITCH_CPU_WIDE:
1608 return tool->context_switch(tool, event, sample, machine);
1609 case PERF_RECORD_KSYMBOL:
1610 return tool->ksymbol(tool, event, sample, machine);
1611 case PERF_RECORD_BPF_EVENT:
1612 return tool->bpf(tool, event, sample, machine);
1613 case PERF_RECORD_TEXT_POKE:
1614 return tool->text_poke(tool, event, sample, machine);
1615 case PERF_RECORD_AUX_OUTPUT_HW_ID:
1616 return tool->aux_output_hw_id(tool, event, sample, machine);
1617 default:
1618 ++evlist->stats.nr_unknown_events;
1619 return -1;
1620 }
1621}
1622
1623static int perf_session__deliver_event(struct perf_session *session,
1624 union perf_event *event,
1625 struct perf_tool *tool,
1626 u64 file_offset,
1627 const char *file_path)
1628{
1629 struct perf_sample sample;
1630 int ret = evlist__parse_sample(session->evlist, event, &sample);
1631
1632 if (ret) {
1633 pr_err("Can't parse sample, err = %d\n", ret);
1634 return ret;
1635 }
1636
1637 ret = auxtrace__process_event(session, event, &sample, tool);
1638 if (ret < 0)
1639 return ret;
1640 if (ret > 0)
1641 return 0;
1642
1643 ret = machines__deliver_event(&session->machines, session->evlist,
1644 event, &sample, tool, file_offset, file_path);
1645
1646 if (dump_trace && sample.aux_sample.size)
1647 auxtrace__dump_auxtrace_sample(session, &sample);
1648
1649 return ret;
1650}
1651
1652static s64 perf_session__process_user_event(struct perf_session *session,
1653 union perf_event *event,
1654 u64 file_offset,
1655 const char *file_path)
1656{
1657 struct ordered_events *oe = &session->ordered_events;
1658 struct perf_tool *tool = session->tool;
1659 struct perf_sample sample = { .time = 0, };
1660 int fd = perf_data__fd(session->data);
1661 int err;
1662
1663 if (event->header.type != PERF_RECORD_COMPRESSED ||
1664 tool->compressed == perf_session__process_compressed_event_stub)
1665 dump_event(session->evlist, event, file_offset, &sample, file_path);
1666
1667 /* These events are processed right away */
1668 switch (event->header.type) {
1669 case PERF_RECORD_HEADER_ATTR:
1670 err = tool->attr(tool, event, &session->evlist);
1671 if (err == 0) {
1672 perf_session__set_id_hdr_size(session);
1673 perf_session__set_comm_exec(session);
1674 }
1675 return err;
1676 case PERF_RECORD_EVENT_UPDATE:
1677 return tool->event_update(tool, event, &session->evlist);
1678 case PERF_RECORD_HEADER_EVENT_TYPE:
1679 /*
1680 * Deprecated, but we need to handle it for sake
1681 * of old data files create in pipe mode.
1682 */
1683 return 0;
1684 case PERF_RECORD_HEADER_TRACING_DATA:
1685 /*
1686 * Setup for reading amidst mmap, but only when we
1687 * are in 'file' mode. The 'pipe' fd is in proper
1688 * place already.
1689 */
1690 if (!perf_data__is_pipe(session->data))
1691 lseek(fd, file_offset, SEEK_SET);
1692 return tool->tracing_data(session, event);
1693 case PERF_RECORD_HEADER_BUILD_ID:
1694 return tool->build_id(session, event);
1695 case PERF_RECORD_FINISHED_ROUND:
1696 return tool->finished_round(tool, event, oe);
1697 case PERF_RECORD_ID_INDEX:
1698 return tool->id_index(session, event);
1699 case PERF_RECORD_AUXTRACE_INFO:
1700 return tool->auxtrace_info(session, event);
1701 case PERF_RECORD_AUXTRACE:
1702 /* setup for reading amidst mmap */
1703 lseek(fd, file_offset + event->header.size, SEEK_SET);
1704 return tool->auxtrace(session, event);
1705 case PERF_RECORD_AUXTRACE_ERROR:
1706 perf_session__auxtrace_error_inc(session, event);
1707 return tool->auxtrace_error(session, event);
1708 case PERF_RECORD_THREAD_MAP:
1709 return tool->thread_map(session, event);
1710 case PERF_RECORD_CPU_MAP:
1711 return tool->cpu_map(session, event);
1712 case PERF_RECORD_STAT_CONFIG:
1713 return tool->stat_config(session, event);
1714 case PERF_RECORD_STAT:
1715 return tool->stat(session, event);
1716 case PERF_RECORD_STAT_ROUND:
1717 return tool->stat_round(session, event);
1718 case PERF_RECORD_TIME_CONV:
1719 session->time_conv = event->time_conv;
1720 return tool->time_conv(session, event);
1721 case PERF_RECORD_HEADER_FEATURE:
1722 return tool->feature(session, event);
1723 case PERF_RECORD_COMPRESSED:
1724 err = tool->compressed(session, event, file_offset, file_path);
1725 if (err)
1726 dump_event(session->evlist, event, file_offset, &sample, file_path);
1727 return err;
1728 case PERF_RECORD_FINISHED_INIT:
1729 return tool->finished_init(session, event);
1730 default:
1731 return -EINVAL;
1732 }
1733}
1734
1735int perf_session__deliver_synth_event(struct perf_session *session,
1736 union perf_event *event,
1737 struct perf_sample *sample)
1738{
1739 struct evlist *evlist = session->evlist;
1740 struct perf_tool *tool = session->tool;
1741
1742 events_stats__inc(&evlist->stats, event->header.type);
1743
1744 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1745 return perf_session__process_user_event(session, event, 0, NULL);
1746
1747 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0, NULL);
1748}
1749
1750static void event_swap(union perf_event *event, bool sample_id_all)
1751{
1752 perf_event__swap_op swap;
1753
1754 swap = perf_event__swap_ops[event->header.type];
1755 if (swap)
1756 swap(event, sample_id_all);
1757}
1758
1759int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1760 void *buf, size_t buf_sz,
1761 union perf_event **event_ptr,
1762 struct perf_sample *sample)
1763{
1764 union perf_event *event;
1765 size_t hdr_sz, rest;
1766 int fd;
1767
1768 if (session->one_mmap && !session->header.needs_swap) {
1769 event = file_offset - session->one_mmap_offset +
1770 session->one_mmap_addr;
1771 goto out_parse_sample;
1772 }
1773
1774 if (perf_data__is_pipe(session->data))
1775 return -1;
1776
1777 fd = perf_data__fd(session->data);
1778 hdr_sz = sizeof(struct perf_event_header);
1779
1780 if (buf_sz < hdr_sz)
1781 return -1;
1782
1783 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1784 readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1785 return -1;
1786
1787 event = (union perf_event *)buf;
1788
1789 if (session->header.needs_swap)
1790 perf_event_header__bswap(&event->header);
1791
1792 if (event->header.size < hdr_sz || event->header.size > buf_sz)
1793 return -1;
1794
1795 buf += hdr_sz;
1796 rest = event->header.size - hdr_sz;
1797
1798 if (readn(fd, buf, rest) != (ssize_t)rest)
1799 return -1;
1800
1801 if (session->header.needs_swap)
1802 event_swap(event, evlist__sample_id_all(session->evlist));
1803
1804out_parse_sample:
1805
1806 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1807 evlist__parse_sample(session->evlist, event, sample))
1808 return -1;
1809
1810 *event_ptr = event;
1811
1812 return 0;
1813}
1814
1815int perf_session__peek_events(struct perf_session *session, u64 offset,
1816 u64 size, peek_events_cb_t cb, void *data)
1817{
1818 u64 max_offset = offset + size;
1819 char buf[PERF_SAMPLE_MAX_SIZE];
1820 union perf_event *event;
1821 int err;
1822
1823 do {
1824 err = perf_session__peek_event(session, offset, buf,
1825 PERF_SAMPLE_MAX_SIZE, &event,
1826 NULL);
1827 if (err)
1828 return err;
1829
1830 err = cb(session, event, offset, data);
1831 if (err)
1832 return err;
1833
1834 offset += event->header.size;
1835 if (event->header.type == PERF_RECORD_AUXTRACE)
1836 offset += event->auxtrace.size;
1837
1838 } while (offset < max_offset);
1839
1840 return err;
1841}
1842
1843static s64 perf_session__process_event(struct perf_session *session,
1844 union perf_event *event, u64 file_offset,
1845 const char *file_path)
1846{
1847 struct evlist *evlist = session->evlist;
1848 struct perf_tool *tool = session->tool;
1849 int ret;
1850
1851 if (session->header.needs_swap)
1852 event_swap(event, evlist__sample_id_all(evlist));
1853
1854 if (event->header.type >= PERF_RECORD_HEADER_MAX)
1855 return -EINVAL;
1856
1857 events_stats__inc(&evlist->stats, event->header.type);
1858
1859 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1860 return perf_session__process_user_event(session, event, file_offset, file_path);
1861
1862 if (tool->ordered_events) {
1863 u64 timestamp = -1ULL;
1864
1865 ret = evlist__parse_sample_timestamp(evlist, event, ×tamp);
1866 if (ret && ret != -1)
1867 return ret;
1868
1869 ret = perf_session__queue_event(session, event, timestamp, file_offset, file_path);
1870 if (ret != -ETIME)
1871 return ret;
1872 }
1873
1874 return perf_session__deliver_event(session, event, tool, file_offset, file_path);
1875}
1876
1877void perf_event_header__bswap(struct perf_event_header *hdr)
1878{
1879 hdr->type = bswap_32(hdr->type);
1880 hdr->misc = bswap_16(hdr->misc);
1881 hdr->size = bswap_16(hdr->size);
1882}
1883
1884struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1885{
1886 return machine__findnew_thread(&session->machines.host, -1, pid);
1887}
1888
1889int perf_session__register_idle_thread(struct perf_session *session)
1890{
1891 struct thread *thread = machine__idle_thread(&session->machines.host);
1892
1893 /* machine__idle_thread() got the thread, so put it */
1894 thread__put(thread);
1895 return thread ? 0 : -1;
1896}
1897
1898static void
1899perf_session__warn_order(const struct perf_session *session)
1900{
1901 const struct ordered_events *oe = &session->ordered_events;
1902 struct evsel *evsel;
1903 bool should_warn = true;
1904
1905 evlist__for_each_entry(session->evlist, evsel) {
1906 if (evsel->core.attr.write_backward)
1907 should_warn = false;
1908 }
1909
1910 if (!should_warn)
1911 return;
1912 if (oe->nr_unordered_events != 0)
1913 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1914}
1915
1916static void perf_session__warn_about_errors(const struct perf_session *session)
1917{
1918 const struct events_stats *stats = &session->evlist->stats;
1919
1920 if (session->tool->lost == perf_event__process_lost &&
1921 stats->nr_events[PERF_RECORD_LOST] != 0) {
1922 ui__warning("Processed %d events and lost %d chunks!\n\n"
1923 "Check IO/CPU overload!\n\n",
1924 stats->nr_events[0],
1925 stats->nr_events[PERF_RECORD_LOST]);
1926 }
1927
1928 if (session->tool->lost_samples == perf_event__process_lost_samples) {
1929 double drop_rate;
1930
1931 drop_rate = (double)stats->total_lost_samples /
1932 (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1933 if (drop_rate > 0.05) {
1934 ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n",
1935 stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1936 drop_rate * 100.0);
1937 }
1938 }
1939
1940 if (session->tool->aux == perf_event__process_aux &&
1941 stats->total_aux_lost != 0) {
1942 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1943 stats->total_aux_lost,
1944 stats->nr_events[PERF_RECORD_AUX]);
1945 }
1946
1947 if (session->tool->aux == perf_event__process_aux &&
1948 stats->total_aux_partial != 0) {
1949 bool vmm_exclusive = false;
1950
1951 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1952 &vmm_exclusive);
1953
1954 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1955 "Are you running a KVM guest in the background?%s\n\n",
1956 stats->total_aux_partial,
1957 stats->nr_events[PERF_RECORD_AUX],
1958 vmm_exclusive ?
1959 "\nReloading kvm_intel module with vmm_exclusive=0\n"
1960 "will reduce the gaps to only guest's timeslices." :
1961 "");
1962 }
1963
1964 if (session->tool->aux == perf_event__process_aux &&
1965 stats->total_aux_collision != 0) {
1966 ui__warning("AUX data detected collision %" PRIu64 " times out of %u!\n\n",
1967 stats->total_aux_collision,
1968 stats->nr_events[PERF_RECORD_AUX]);
1969 }
1970
1971 if (stats->nr_unknown_events != 0) {
1972 ui__warning("Found %u unknown events!\n\n"
1973 "Is this an older tool processing a perf.data "
1974 "file generated by a more recent tool?\n\n"
1975 "If that is not the case, consider "
1976 "reporting to linux-kernel@vger.kernel.org.\n\n",
1977 stats->nr_unknown_events);
1978 }
1979
1980 if (stats->nr_unknown_id != 0) {
1981 ui__warning("%u samples with id not present in the header\n",
1982 stats->nr_unknown_id);
1983 }
1984
1985 if (stats->nr_invalid_chains != 0) {
1986 ui__warning("Found invalid callchains!\n\n"
1987 "%u out of %u events were discarded for this reason.\n\n"
1988 "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1989 stats->nr_invalid_chains,
1990 stats->nr_events[PERF_RECORD_SAMPLE]);
1991 }
1992
1993 if (stats->nr_unprocessable_samples != 0) {
1994 ui__warning("%u unprocessable samples recorded.\n"
1995 "Do you have a KVM guest running and not using 'perf kvm'?\n",
1996 stats->nr_unprocessable_samples);
1997 }
1998
1999 perf_session__warn_order(session);
2000
2001 events_stats__auxtrace_error_warn(stats);
2002
2003 if (stats->nr_proc_map_timeout != 0) {
2004 ui__warning("%d map information files for pre-existing threads were\n"
2005 "not processed, if there are samples for addresses they\n"
2006 "will not be resolved, you may find out which are these\n"
2007 "threads by running with -v and redirecting the output\n"
2008 "to a file.\n"
2009 "The time limit to process proc map is too short?\n"
2010 "Increase it by --proc-map-timeout\n",
2011 stats->nr_proc_map_timeout);
2012 }
2013}
2014
2015static int perf_session__flush_thread_stack(struct thread *thread,
2016 void *p __maybe_unused)
2017{
2018 return thread_stack__flush(thread);
2019}
2020
2021static int perf_session__flush_thread_stacks(struct perf_session *session)
2022{
2023 return machines__for_each_thread(&session->machines,
2024 perf_session__flush_thread_stack,
2025 NULL);
2026}
2027
2028volatile sig_atomic_t session_done;
2029
2030static int __perf_session__process_decomp_events(struct perf_session *session);
2031
2032static int __perf_session__process_pipe_events(struct perf_session *session)
2033{
2034 struct ordered_events *oe = &session->ordered_events;
2035 struct perf_tool *tool = session->tool;
2036 union perf_event *event;
2037 uint32_t size, cur_size = 0;
2038 void *buf = NULL;
2039 s64 skip = 0;
2040 u64 head;
2041 ssize_t err;
2042 void *p;
2043
2044 perf_tool__fill_defaults(tool);
2045
2046 head = 0;
2047 cur_size = sizeof(union perf_event);
2048
2049 buf = malloc(cur_size);
2050 if (!buf)
2051 return -errno;
2052 ordered_events__set_copy_on_queue(oe, true);
2053more:
2054 event = buf;
2055 err = perf_data__read(session->data, event,
2056 sizeof(struct perf_event_header));
2057 if (err <= 0) {
2058 if (err == 0)
2059 goto done;
2060
2061 pr_err("failed to read event header\n");
2062 goto out_err;
2063 }
2064
2065 if (session->header.needs_swap)
2066 perf_event_header__bswap(&event->header);
2067
2068 size = event->header.size;
2069 if (size < sizeof(struct perf_event_header)) {
2070 pr_err("bad event header size\n");
2071 goto out_err;
2072 }
2073
2074 if (size > cur_size) {
2075 void *new = realloc(buf, size);
2076 if (!new) {
2077 pr_err("failed to allocate memory to read event\n");
2078 goto out_err;
2079 }
2080 buf = new;
2081 cur_size = size;
2082 event = buf;
2083 }
2084 p = event;
2085 p += sizeof(struct perf_event_header);
2086
2087 if (size - sizeof(struct perf_event_header)) {
2088 err = perf_data__read(session->data, p,
2089 size - sizeof(struct perf_event_header));
2090 if (err <= 0) {
2091 if (err == 0) {
2092 pr_err("unexpected end of event stream\n");
2093 goto done;
2094 }
2095
2096 pr_err("failed to read event data\n");
2097 goto out_err;
2098 }
2099 }
2100
2101 if ((skip = perf_session__process_event(session, event, head, "pipe")) < 0) {
2102 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2103 head, event->header.size, event->header.type);
2104 err = -EINVAL;
2105 goto out_err;
2106 }
2107
2108 head += size;
2109
2110 if (skip > 0)
2111 head += skip;
2112
2113 err = __perf_session__process_decomp_events(session);
2114 if (err)
2115 goto out_err;
2116
2117 if (!session_done())
2118 goto more;
2119done:
2120 /* do the final flush for ordered samples */
2121 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2122 if (err)
2123 goto out_err;
2124 err = auxtrace__flush_events(session, tool);
2125 if (err)
2126 goto out_err;
2127 err = perf_session__flush_thread_stacks(session);
2128out_err:
2129 free(buf);
2130 if (!tool->no_warn)
2131 perf_session__warn_about_errors(session);
2132 ordered_events__free(&session->ordered_events);
2133 auxtrace__free_events(session);
2134 return err;
2135}
2136
2137static union perf_event *
2138prefetch_event(char *buf, u64 head, size_t mmap_size,
2139 bool needs_swap, union perf_event *error)
2140{
2141 union perf_event *event;
2142 u16 event_size;
2143
2144 /*
2145 * Ensure we have enough space remaining to read
2146 * the size of the event in the headers.
2147 */
2148 if (head + sizeof(event->header) > mmap_size)
2149 return NULL;
2150
2151 event = (union perf_event *)(buf + head);
2152 if (needs_swap)
2153 perf_event_header__bswap(&event->header);
2154
2155 event_size = event->header.size;
2156 if (head + event_size <= mmap_size)
2157 return event;
2158
2159 /* We're not fetching the event so swap back again */
2160 if (needs_swap)
2161 perf_event_header__bswap(&event->header);
2162
2163 /* Check if the event fits into the next mmapped buf. */
2164 if (event_size <= mmap_size - head % page_size) {
2165 /* Remap buf and fetch again. */
2166 return NULL;
2167 }
2168
2169 /* Invalid input. Event size should never exceed mmap_size. */
2170 pr_debug("%s: head=%#" PRIx64 " event->header.size=%#x, mmap_size=%#zx:"
2171 " fuzzed or compressed perf.data?\n", __func__, head, event_size, mmap_size);
2172
2173 return error;
2174}
2175
2176static union perf_event *
2177fetch_mmaped_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2178{
2179 return prefetch_event(buf, head, mmap_size, needs_swap, ERR_PTR(-EINVAL));
2180}
2181
2182static union perf_event *
2183fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2184{
2185 return prefetch_event(buf, head, mmap_size, needs_swap, NULL);
2186}
2187
2188static int __perf_session__process_decomp_events(struct perf_session *session)
2189{
2190 s64 skip;
2191 u64 size;
2192 struct decomp *decomp = session->active_decomp->decomp_last;
2193
2194 if (!decomp)
2195 return 0;
2196
2197 while (decomp->head < decomp->size && !session_done()) {
2198 union perf_event *event = fetch_decomp_event(decomp->head, decomp->size, decomp->data,
2199 session->header.needs_swap);
2200
2201 if (!event)
2202 break;
2203
2204 size = event->header.size;
2205
2206 if (size < sizeof(struct perf_event_header) ||
2207 (skip = perf_session__process_event(session, event, decomp->file_pos,
2208 decomp->file_path)) < 0) {
2209 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2210 decomp->file_pos + decomp->head, event->header.size, event->header.type);
2211 return -EINVAL;
2212 }
2213
2214 if (skip)
2215 size += skip;
2216
2217 decomp->head += size;
2218 }
2219
2220 return 0;
2221}
2222
2223/*
2224 * On 64bit we can mmap the data file in one go. No need for tiny mmap
2225 * slices. On 32bit we use 32MB.
2226 */
2227#if BITS_PER_LONG == 64
2228#define MMAP_SIZE ULLONG_MAX
2229#define NUM_MMAPS 1
2230#else
2231#define MMAP_SIZE (32 * 1024 * 1024ULL)
2232#define NUM_MMAPS 128
2233#endif
2234
2235struct reader;
2236
2237typedef s64 (*reader_cb_t)(struct perf_session *session,
2238 union perf_event *event,
2239 u64 file_offset,
2240 const char *file_path);
2241
2242struct reader {
2243 int fd;
2244 const char *path;
2245 u64 data_size;
2246 u64 data_offset;
2247 reader_cb_t process;
2248 bool in_place_update;
2249 char *mmaps[NUM_MMAPS];
2250 size_t mmap_size;
2251 int mmap_idx;
2252 char *mmap_cur;
2253 u64 file_pos;
2254 u64 file_offset;
2255 u64 head;
2256 u64 size;
2257 bool done;
2258 struct zstd_data zstd_data;
2259 struct decomp_data decomp_data;
2260};
2261
2262static int
2263reader__init(struct reader *rd, bool *one_mmap)
2264{
2265 u64 data_size = rd->data_size;
2266 char **mmaps = rd->mmaps;
2267
2268 rd->head = rd->data_offset;
2269 data_size += rd->data_offset;
2270
2271 rd->mmap_size = MMAP_SIZE;
2272 if (rd->mmap_size > data_size) {
2273 rd->mmap_size = data_size;
2274 if (one_mmap)
2275 *one_mmap = true;
2276 }
2277
2278 memset(mmaps, 0, sizeof(rd->mmaps));
2279
2280 if (zstd_init(&rd->zstd_data, 0))
2281 return -1;
2282 rd->decomp_data.zstd_decomp = &rd->zstd_data;
2283
2284 return 0;
2285}
2286
2287static void
2288reader__release_decomp(struct reader *rd)
2289{
2290 perf_decomp__release_events(rd->decomp_data.decomp);
2291 zstd_fini(&rd->zstd_data);
2292}
2293
2294static int
2295reader__mmap(struct reader *rd, struct perf_session *session)
2296{
2297 int mmap_prot, mmap_flags;
2298 char *buf, **mmaps = rd->mmaps;
2299 u64 page_offset;
2300
2301 mmap_prot = PROT_READ;
2302 mmap_flags = MAP_SHARED;
2303
2304 if (rd->in_place_update) {
2305 mmap_prot |= PROT_WRITE;
2306 } else if (session->header.needs_swap) {
2307 mmap_prot |= PROT_WRITE;
2308 mmap_flags = MAP_PRIVATE;
2309 }
2310
2311 if (mmaps[rd->mmap_idx]) {
2312 munmap(mmaps[rd->mmap_idx], rd->mmap_size);
2313 mmaps[rd->mmap_idx] = NULL;
2314 }
2315
2316 page_offset = page_size * (rd->head / page_size);
2317 rd->file_offset += page_offset;
2318 rd->head -= page_offset;
2319
2320 buf = mmap(NULL, rd->mmap_size, mmap_prot, mmap_flags, rd->fd,
2321 rd->file_offset);
2322 if (buf == MAP_FAILED) {
2323 pr_err("failed to mmap file\n");
2324 return -errno;
2325 }
2326 mmaps[rd->mmap_idx] = rd->mmap_cur = buf;
2327 rd->mmap_idx = (rd->mmap_idx + 1) & (ARRAY_SIZE(rd->mmaps) - 1);
2328 rd->file_pos = rd->file_offset + rd->head;
2329 if (session->one_mmap) {
2330 session->one_mmap_addr = buf;
2331 session->one_mmap_offset = rd->file_offset;
2332 }
2333
2334 return 0;
2335}
2336
2337enum {
2338 READER_OK,
2339 READER_NODATA,
2340};
2341
2342static int
2343reader__read_event(struct reader *rd, struct perf_session *session,
2344 struct ui_progress *prog)
2345{
2346 u64 size;
2347 int err = READER_OK;
2348 union perf_event *event;
2349 s64 skip;
2350
2351 event = fetch_mmaped_event(rd->head, rd->mmap_size, rd->mmap_cur,
2352 session->header.needs_swap);
2353 if (IS_ERR(event))
2354 return PTR_ERR(event);
2355
2356 if (!event)
2357 return READER_NODATA;
2358
2359 size = event->header.size;
2360
2361 skip = -EINVAL;
2362
2363 if (size < sizeof(struct perf_event_header) ||
2364 (skip = rd->process(session, event, rd->file_pos, rd->path)) < 0) {
2365 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n",
2366 rd->file_offset + rd->head, event->header.size,
2367 event->header.type, strerror(-skip));
2368 err = skip;
2369 goto out;
2370 }
2371
2372 if (skip)
2373 size += skip;
2374
2375 rd->size += size;
2376 rd->head += size;
2377 rd->file_pos += size;
2378
2379 err = __perf_session__process_decomp_events(session);
2380 if (err)
2381 goto out;
2382
2383 ui_progress__update(prog, size);
2384
2385out:
2386 return err;
2387}
2388
2389static inline bool
2390reader__eof(struct reader *rd)
2391{
2392 return (rd->file_pos >= rd->data_size + rd->data_offset);
2393}
2394
2395static int
2396reader__process_events(struct reader *rd, struct perf_session *session,
2397 struct ui_progress *prog)
2398{
2399 int err;
2400
2401 err = reader__init(rd, &session->one_mmap);
2402 if (err)
2403 goto out;
2404
2405 session->active_decomp = &rd->decomp_data;
2406
2407remap:
2408 err = reader__mmap(rd, session);
2409 if (err)
2410 goto out;
2411
2412more:
2413 err = reader__read_event(rd, session, prog);
2414 if (err < 0)
2415 goto out;
2416 else if (err == READER_NODATA)
2417 goto remap;
2418
2419 if (session_done())
2420 goto out;
2421
2422 if (!reader__eof(rd))
2423 goto more;
2424
2425out:
2426 session->active_decomp = &session->decomp_data;
2427 return err;
2428}
2429
2430static s64 process_simple(struct perf_session *session,
2431 union perf_event *event,
2432 u64 file_offset,
2433 const char *file_path)
2434{
2435 return perf_session__process_event(session, event, file_offset, file_path);
2436}
2437
2438static int __perf_session__process_events(struct perf_session *session)
2439{
2440 struct reader rd = {
2441 .fd = perf_data__fd(session->data),
2442 .path = session->data->file.path,
2443 .data_size = session->header.data_size,
2444 .data_offset = session->header.data_offset,
2445 .process = process_simple,
2446 .in_place_update = session->data->in_place_update,
2447 };
2448 struct ordered_events *oe = &session->ordered_events;
2449 struct perf_tool *tool = session->tool;
2450 struct ui_progress prog;
2451 int err;
2452
2453 perf_tool__fill_defaults(tool);
2454
2455 if (rd.data_size == 0)
2456 return -1;
2457
2458 ui_progress__init_size(&prog, rd.data_size, "Processing events...");
2459
2460 err = reader__process_events(&rd, session, &prog);
2461 if (err)
2462 goto out_err;
2463 /* do the final flush for ordered samples */
2464 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2465 if (err)
2466 goto out_err;
2467 err = auxtrace__flush_events(session, tool);
2468 if (err)
2469 goto out_err;
2470 err = perf_session__flush_thread_stacks(session);
2471out_err:
2472 ui_progress__finish();
2473 if (!tool->no_warn)
2474 perf_session__warn_about_errors(session);
2475 /*
2476 * We may switching perf.data output, make ordered_events
2477 * reusable.
2478 */
2479 ordered_events__reinit(&session->ordered_events);
2480 auxtrace__free_events(session);
2481 reader__release_decomp(&rd);
2482 session->one_mmap = false;
2483 return err;
2484}
2485
2486/*
2487 * Processing 2 MB of data from each reader in sequence,
2488 * because that's the way the ordered events sorting works
2489 * most efficiently.
2490 */
2491#define READER_MAX_SIZE (2 * 1024 * 1024)
2492
2493/*
2494 * This function reads, merge and process directory data.
2495 * It assumens the version 1 of directory data, where each
2496 * data file holds per-cpu data, already sorted by kernel.
2497 */
2498static int __perf_session__process_dir_events(struct perf_session *session)
2499{
2500 struct perf_data *data = session->data;
2501 struct perf_tool *tool = session->tool;
2502 int i, ret, readers, nr_readers;
2503 struct ui_progress prog;
2504 u64 total_size = perf_data__size(session->data);
2505 struct reader *rd;
2506
2507 perf_tool__fill_defaults(tool);
2508
2509 ui_progress__init_size(&prog, total_size, "Sorting events...");
2510
2511 nr_readers = 1;
2512 for (i = 0; i < data->dir.nr; i++) {
2513 if (data->dir.files[i].size)
2514 nr_readers++;
2515 }
2516
2517 rd = zalloc(nr_readers * sizeof(struct reader));
2518 if (!rd)
2519 return -ENOMEM;
2520
2521 rd[0] = (struct reader) {
2522 .fd = perf_data__fd(session->data),
2523 .path = session->data->file.path,
2524 .data_size = session->header.data_size,
2525 .data_offset = session->header.data_offset,
2526 .process = process_simple,
2527 .in_place_update = session->data->in_place_update,
2528 };
2529 ret = reader__init(&rd[0], NULL);
2530 if (ret)
2531 goto out_err;
2532 ret = reader__mmap(&rd[0], session);
2533 if (ret)
2534 goto out_err;
2535 readers = 1;
2536
2537 for (i = 0; i < data->dir.nr; i++) {
2538 if (!data->dir.files[i].size)
2539 continue;
2540 rd[readers] = (struct reader) {
2541 .fd = data->dir.files[i].fd,
2542 .path = data->dir.files[i].path,
2543 .data_size = data->dir.files[i].size,
2544 .data_offset = 0,
2545 .process = process_simple,
2546 .in_place_update = session->data->in_place_update,
2547 };
2548 ret = reader__init(&rd[readers], NULL);
2549 if (ret)
2550 goto out_err;
2551 ret = reader__mmap(&rd[readers], session);
2552 if (ret)
2553 goto out_err;
2554 readers++;
2555 }
2556
2557 i = 0;
2558 while (readers) {
2559 if (session_done())
2560 break;
2561
2562 if (rd[i].done) {
2563 i = (i + 1) % nr_readers;
2564 continue;
2565 }
2566 if (reader__eof(&rd[i])) {
2567 rd[i].done = true;
2568 readers--;
2569 continue;
2570 }
2571
2572 session->active_decomp = &rd[i].decomp_data;
2573 ret = reader__read_event(&rd[i], session, &prog);
2574 if (ret < 0) {
2575 goto out_err;
2576 } else if (ret == READER_NODATA) {
2577 ret = reader__mmap(&rd[i], session);
2578 if (ret)
2579 goto out_err;
2580 }
2581
2582 if (rd[i].size >= READER_MAX_SIZE) {
2583 rd[i].size = 0;
2584 i = (i + 1) % nr_readers;
2585 }
2586 }
2587
2588 ret = ordered_events__flush(&session->ordered_events, OE_FLUSH__FINAL);
2589 if (ret)
2590 goto out_err;
2591
2592 ret = perf_session__flush_thread_stacks(session);
2593out_err:
2594 ui_progress__finish();
2595
2596 if (!tool->no_warn)
2597 perf_session__warn_about_errors(session);
2598
2599 /*
2600 * We may switching perf.data output, make ordered_events
2601 * reusable.
2602 */
2603 ordered_events__reinit(&session->ordered_events);
2604
2605 session->one_mmap = false;
2606
2607 session->active_decomp = &session->decomp_data;
2608 for (i = 0; i < nr_readers; i++)
2609 reader__release_decomp(&rd[i]);
2610 zfree(&rd);
2611
2612 return ret;
2613}
2614
2615int perf_session__process_events(struct perf_session *session)
2616{
2617 if (perf_session__register_idle_thread(session) < 0)
2618 return -ENOMEM;
2619
2620 if (perf_data__is_pipe(session->data))
2621 return __perf_session__process_pipe_events(session);
2622
2623 if (perf_data__is_dir(session->data) && session->data->dir.nr)
2624 return __perf_session__process_dir_events(session);
2625
2626 return __perf_session__process_events(session);
2627}
2628
2629bool perf_session__has_traces(struct perf_session *session, const char *msg)
2630{
2631 struct evsel *evsel;
2632
2633 evlist__for_each_entry(session->evlist, evsel) {
2634 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT)
2635 return true;
2636 }
2637
2638 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
2639 return false;
2640}
2641
2642int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr)
2643{
2644 char *bracket;
2645 struct ref_reloc_sym *ref;
2646 struct kmap *kmap;
2647
2648 ref = zalloc(sizeof(struct ref_reloc_sym));
2649 if (ref == NULL)
2650 return -ENOMEM;
2651
2652 ref->name = strdup(symbol_name);
2653 if (ref->name == NULL) {
2654 free(ref);
2655 return -ENOMEM;
2656 }
2657
2658 bracket = strchr(ref->name, ']');
2659 if (bracket)
2660 *bracket = '\0';
2661
2662 ref->addr = addr;
2663
2664 kmap = map__kmap(map);
2665 if (kmap)
2666 kmap->ref_reloc_sym = ref;
2667
2668 return 0;
2669}
2670
2671size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2672{
2673 return machines__fprintf_dsos(&session->machines, fp);
2674}
2675
2676size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2677 bool (skip)(struct dso *dso, int parm), int parm)
2678{
2679 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2680}
2681
2682size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp,
2683 bool skip_empty)
2684{
2685 size_t ret;
2686 const char *msg = "";
2687
2688 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2689 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2690
2691 ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2692
2693 ret += events_stats__fprintf(&session->evlist->stats, fp, skip_empty);
2694 return ret;
2695}
2696
2697size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2698{
2699 /*
2700 * FIXME: Here we have to actually print all the machines in this
2701 * session, not just the host...
2702 */
2703 return machine__fprintf(&session->machines.host, fp);
2704}
2705
2706struct evsel *perf_session__find_first_evtype(struct perf_session *session,
2707 unsigned int type)
2708{
2709 struct evsel *pos;
2710
2711 evlist__for_each_entry(session->evlist, pos) {
2712 if (pos->core.attr.type == type)
2713 return pos;
2714 }
2715 return NULL;
2716}
2717
2718int perf_session__cpu_bitmap(struct perf_session *session,
2719 const char *cpu_list, unsigned long *cpu_bitmap)
2720{
2721 int i, err = -1;
2722 struct perf_cpu_map *map;
2723 int nr_cpus = min(session->header.env.nr_cpus_avail, MAX_NR_CPUS);
2724
2725 for (i = 0; i < PERF_TYPE_MAX; ++i) {
2726 struct evsel *evsel;
2727
2728 evsel = perf_session__find_first_evtype(session, i);
2729 if (!evsel)
2730 continue;
2731
2732 if (!(evsel->core.attr.sample_type & PERF_SAMPLE_CPU)) {
2733 pr_err("File does not contain CPU events. "
2734 "Remove -C option to proceed.\n");
2735 return -1;
2736 }
2737 }
2738
2739 map = perf_cpu_map__new(cpu_list);
2740 if (map == NULL) {
2741 pr_err("Invalid cpu_list\n");
2742 return -1;
2743 }
2744
2745 for (i = 0; i < perf_cpu_map__nr(map); i++) {
2746 struct perf_cpu cpu = perf_cpu_map__cpu(map, i);
2747
2748 if (cpu.cpu >= nr_cpus) {
2749 pr_err("Requested CPU %d too large. "
2750 "Consider raising MAX_NR_CPUS\n", cpu.cpu);
2751 goto out_delete_map;
2752 }
2753
2754 __set_bit(cpu.cpu, cpu_bitmap);
2755 }
2756
2757 err = 0;
2758
2759out_delete_map:
2760 perf_cpu_map__put(map);
2761 return err;
2762}
2763
2764void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2765 bool full)
2766{
2767 if (session == NULL || fp == NULL)
2768 return;
2769
2770 fprintf(fp, "# ========\n");
2771 perf_header__fprintf_info(session, fp, full);
2772 fprintf(fp, "# ========\n#\n");
2773}
2774
2775static int perf_session__register_guest(struct perf_session *session, pid_t machine_pid)
2776{
2777 struct machine *machine = machines__findnew(&session->machines, machine_pid);
2778 struct thread *thread;
2779
2780 if (!machine)
2781 return -ENOMEM;
2782
2783 machine->single_address_space = session->machines.host.single_address_space;
2784
2785 thread = machine__idle_thread(machine);
2786 if (!thread)
2787 return -ENOMEM;
2788 thread__put(thread);
2789
2790 machine->kallsyms_filename = perf_data__guest_kallsyms_name(session->data, machine_pid);
2791
2792 return 0;
2793}
2794
2795static int perf_session__set_guest_cpu(struct perf_session *session, pid_t pid,
2796 pid_t tid, int guest_cpu)
2797{
2798 struct machine *machine = &session->machines.host;
2799 struct thread *thread = machine__findnew_thread(machine, pid, tid);
2800
2801 if (!thread)
2802 return -ENOMEM;
2803 thread->guest_cpu = guest_cpu;
2804 thread__put(thread);
2805
2806 return 0;
2807}
2808
2809int perf_event__process_id_index(struct perf_session *session,
2810 union perf_event *event)
2811{
2812 struct evlist *evlist = session->evlist;
2813 struct perf_record_id_index *ie = &event->id_index;
2814 size_t sz = ie->header.size - sizeof(*ie);
2815 size_t i, nr, max_nr;
2816 size_t e1_sz = sizeof(struct id_index_entry);
2817 size_t e2_sz = sizeof(struct id_index_entry_2);
2818 size_t etot_sz = e1_sz + e2_sz;
2819 struct id_index_entry_2 *e2;
2820 pid_t last_pid = 0;
2821
2822 max_nr = sz / e1_sz;
2823 nr = ie->nr;
2824 if (nr > max_nr) {
2825 printf("Too big: nr %zu max_nr %zu\n", nr, max_nr);
2826 return -EINVAL;
2827 }
2828
2829 if (sz >= nr * etot_sz) {
2830 max_nr = sz / etot_sz;
2831 if (nr > max_nr) {
2832 printf("Too big2: nr %zu max_nr %zu\n", nr, max_nr);
2833 return -EINVAL;
2834 }
2835 e2 = (void *)ie + sizeof(*ie) + nr * e1_sz;
2836 } else {
2837 e2 = NULL;
2838 }
2839
2840 if (dump_trace)
2841 fprintf(stdout, " nr: %zu\n", nr);
2842
2843 for (i = 0; i < nr; i++, (e2 ? e2++ : 0)) {
2844 struct id_index_entry *e = &ie->entries[i];
2845 struct perf_sample_id *sid;
2846 int ret;
2847
2848 if (dump_trace) {
2849 fprintf(stdout, " ... id: %"PRI_lu64, e->id);
2850 fprintf(stdout, " idx: %"PRI_lu64, e->idx);
2851 fprintf(stdout, " cpu: %"PRI_ld64, e->cpu);
2852 fprintf(stdout, " tid: %"PRI_ld64, e->tid);
2853 if (e2) {
2854 fprintf(stdout, " machine_pid: %"PRI_ld64, e2->machine_pid);
2855 fprintf(stdout, " vcpu: %"PRI_lu64"\n", e2->vcpu);
2856 } else {
2857 fprintf(stdout, "\n");
2858 }
2859 }
2860
2861 sid = evlist__id2sid(evlist, e->id);
2862 if (!sid)
2863 return -ENOENT;
2864
2865 sid->idx = e->idx;
2866 sid->cpu.cpu = e->cpu;
2867 sid->tid = e->tid;
2868
2869 if (!e2)
2870 continue;
2871
2872 sid->machine_pid = e2->machine_pid;
2873 sid->vcpu.cpu = e2->vcpu;
2874
2875 if (!sid->machine_pid)
2876 continue;
2877
2878 if (sid->machine_pid != last_pid) {
2879 ret = perf_session__register_guest(session, sid->machine_pid);
2880 if (ret)
2881 return ret;
2882 last_pid = sid->machine_pid;
2883 perf_guest = true;
2884 }
2885
2886 ret = perf_session__set_guest_cpu(session, sid->machine_pid, e->tid, e2->vcpu);
2887 if (ret)
2888 return ret;
2889 }
2890 return 0;
2891}
1#define _FILE_OFFSET_BITS 64
2
3#include <linux/kernel.h>
4
5#include <byteswap.h>
6#include <unistd.h>
7#include <sys/types.h>
8#include <sys/mman.h>
9
10#include "evlist.h"
11#include "evsel.h"
12#include "session.h"
13#include "sort.h"
14#include "util.h"
15#include "cpumap.h"
16
17static int perf_session__open(struct perf_session *self, bool force)
18{
19 struct stat input_stat;
20
21 if (!strcmp(self->filename, "-")) {
22 self->fd_pipe = true;
23 self->fd = STDIN_FILENO;
24
25 if (perf_session__read_header(self, self->fd) < 0)
26 pr_err("incompatible file format");
27
28 return 0;
29 }
30
31 self->fd = open(self->filename, O_RDONLY);
32 if (self->fd < 0) {
33 int err = errno;
34
35 pr_err("failed to open %s: %s", self->filename, strerror(err));
36 if (err == ENOENT && !strcmp(self->filename, "perf.data"))
37 pr_err(" (try 'perf record' first)");
38 pr_err("\n");
39 return -errno;
40 }
41
42 if (fstat(self->fd, &input_stat) < 0)
43 goto out_close;
44
45 if (!force && input_stat.st_uid && (input_stat.st_uid != geteuid())) {
46 pr_err("file %s not owned by current user or root\n",
47 self->filename);
48 goto out_close;
49 }
50
51 if (!input_stat.st_size) {
52 pr_info("zero-sized file (%s), nothing to do!\n",
53 self->filename);
54 goto out_close;
55 }
56
57 if (perf_session__read_header(self, self->fd) < 0) {
58 pr_err("incompatible file format");
59 goto out_close;
60 }
61
62 if (!perf_evlist__valid_sample_type(self->evlist)) {
63 pr_err("non matching sample_type");
64 goto out_close;
65 }
66
67 if (!perf_evlist__valid_sample_id_all(self->evlist)) {
68 pr_err("non matching sample_id_all");
69 goto out_close;
70 }
71
72 self->size = input_stat.st_size;
73 return 0;
74
75out_close:
76 close(self->fd);
77 self->fd = -1;
78 return -1;
79}
80
81static void perf_session__id_header_size(struct perf_session *session)
82{
83 struct perf_sample *data;
84 u64 sample_type = session->sample_type;
85 u16 size = 0;
86
87 if (!session->sample_id_all)
88 goto out;
89
90 if (sample_type & PERF_SAMPLE_TID)
91 size += sizeof(data->tid) * 2;
92
93 if (sample_type & PERF_SAMPLE_TIME)
94 size += sizeof(data->time);
95
96 if (sample_type & PERF_SAMPLE_ID)
97 size += sizeof(data->id);
98
99 if (sample_type & PERF_SAMPLE_STREAM_ID)
100 size += sizeof(data->stream_id);
101
102 if (sample_type & PERF_SAMPLE_CPU)
103 size += sizeof(data->cpu) * 2;
104out:
105 session->id_hdr_size = size;
106}
107
108void perf_session__update_sample_type(struct perf_session *self)
109{
110 self->sample_type = perf_evlist__sample_type(self->evlist);
111 self->sample_size = __perf_evsel__sample_size(self->sample_type);
112 self->sample_id_all = perf_evlist__sample_id_all(self->evlist);
113 perf_session__id_header_size(self);
114}
115
116int perf_session__create_kernel_maps(struct perf_session *self)
117{
118 int ret = machine__create_kernel_maps(&self->host_machine);
119
120 if (ret >= 0)
121 ret = machines__create_guest_kernel_maps(&self->machines);
122 return ret;
123}
124
125static void perf_session__destroy_kernel_maps(struct perf_session *self)
126{
127 machine__destroy_kernel_maps(&self->host_machine);
128 machines__destroy_guest_kernel_maps(&self->machines);
129}
130
131struct perf_session *perf_session__new(const char *filename, int mode,
132 bool force, bool repipe,
133 struct perf_event_ops *ops)
134{
135 size_t len = filename ? strlen(filename) + 1 : 0;
136 struct perf_session *self = zalloc(sizeof(*self) + len);
137
138 if (self == NULL)
139 goto out;
140
141 memcpy(self->filename, filename, len);
142 self->threads = RB_ROOT;
143 INIT_LIST_HEAD(&self->dead_threads);
144 self->last_match = NULL;
145 /*
146 * On 64bit we can mmap the data file in one go. No need for tiny mmap
147 * slices. On 32bit we use 32MB.
148 */
149#if BITS_PER_LONG == 64
150 self->mmap_window = ULLONG_MAX;
151#else
152 self->mmap_window = 32 * 1024 * 1024ULL;
153#endif
154 self->machines = RB_ROOT;
155 self->repipe = repipe;
156 INIT_LIST_HEAD(&self->ordered_samples.samples);
157 INIT_LIST_HEAD(&self->ordered_samples.sample_cache);
158 INIT_LIST_HEAD(&self->ordered_samples.to_free);
159 machine__init(&self->host_machine, "", HOST_KERNEL_ID);
160
161 if (mode == O_RDONLY) {
162 if (perf_session__open(self, force) < 0)
163 goto out_delete;
164 perf_session__update_sample_type(self);
165 } else if (mode == O_WRONLY) {
166 /*
167 * In O_RDONLY mode this will be performed when reading the
168 * kernel MMAP event, in perf_event__process_mmap().
169 */
170 if (perf_session__create_kernel_maps(self) < 0)
171 goto out_delete;
172 }
173
174 if (ops && ops->ordering_requires_timestamps &&
175 ops->ordered_samples && !self->sample_id_all) {
176 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
177 ops->ordered_samples = false;
178 }
179
180out:
181 return self;
182out_delete:
183 perf_session__delete(self);
184 return NULL;
185}
186
187static void perf_session__delete_dead_threads(struct perf_session *self)
188{
189 struct thread *n, *t;
190
191 list_for_each_entry_safe(t, n, &self->dead_threads, node) {
192 list_del(&t->node);
193 thread__delete(t);
194 }
195}
196
197static void perf_session__delete_threads(struct perf_session *self)
198{
199 struct rb_node *nd = rb_first(&self->threads);
200
201 while (nd) {
202 struct thread *t = rb_entry(nd, struct thread, rb_node);
203
204 rb_erase(&t->rb_node, &self->threads);
205 nd = rb_next(nd);
206 thread__delete(t);
207 }
208}
209
210void perf_session__delete(struct perf_session *self)
211{
212 perf_session__destroy_kernel_maps(self);
213 perf_session__delete_dead_threads(self);
214 perf_session__delete_threads(self);
215 machine__exit(&self->host_machine);
216 close(self->fd);
217 free(self);
218}
219
220void perf_session__remove_thread(struct perf_session *self, struct thread *th)
221{
222 self->last_match = NULL;
223 rb_erase(&th->rb_node, &self->threads);
224 /*
225 * We may have references to this thread, for instance in some hist_entry
226 * instances, so just move them to a separate list.
227 */
228 list_add_tail(&th->node, &self->dead_threads);
229}
230
231static bool symbol__match_parent_regex(struct symbol *sym)
232{
233 if (sym->name && !regexec(&parent_regex, sym->name, 0, NULL, 0))
234 return 1;
235
236 return 0;
237}
238
239int perf_session__resolve_callchain(struct perf_session *self,
240 struct thread *thread,
241 struct ip_callchain *chain,
242 struct symbol **parent)
243{
244 u8 cpumode = PERF_RECORD_MISC_USER;
245 unsigned int i;
246 int err;
247
248 callchain_cursor_reset(&self->callchain_cursor);
249
250 for (i = 0; i < chain->nr; i++) {
251 u64 ip;
252 struct addr_location al;
253
254 if (callchain_param.order == ORDER_CALLEE)
255 ip = chain->ips[i];
256 else
257 ip = chain->ips[chain->nr - i - 1];
258
259 if (ip >= PERF_CONTEXT_MAX) {
260 switch (ip) {
261 case PERF_CONTEXT_HV:
262 cpumode = PERF_RECORD_MISC_HYPERVISOR; break;
263 case PERF_CONTEXT_KERNEL:
264 cpumode = PERF_RECORD_MISC_KERNEL; break;
265 case PERF_CONTEXT_USER:
266 cpumode = PERF_RECORD_MISC_USER; break;
267 default:
268 break;
269 }
270 continue;
271 }
272
273 al.filtered = false;
274 thread__find_addr_location(thread, self, cpumode,
275 MAP__FUNCTION, thread->pid, ip, &al, NULL);
276 if (al.sym != NULL) {
277 if (sort__has_parent && !*parent &&
278 symbol__match_parent_regex(al.sym))
279 *parent = al.sym;
280 if (!symbol_conf.use_callchain)
281 break;
282 }
283
284 err = callchain_cursor_append(&self->callchain_cursor,
285 ip, al.map, al.sym);
286 if (err)
287 return err;
288 }
289
290 return 0;
291}
292
293static int process_event_synth_stub(union perf_event *event __used,
294 struct perf_session *session __used)
295{
296 dump_printf(": unhandled!\n");
297 return 0;
298}
299
300static int process_event_sample_stub(union perf_event *event __used,
301 struct perf_sample *sample __used,
302 struct perf_evsel *evsel __used,
303 struct perf_session *session __used)
304{
305 dump_printf(": unhandled!\n");
306 return 0;
307}
308
309static int process_event_stub(union perf_event *event __used,
310 struct perf_sample *sample __used,
311 struct perf_session *session __used)
312{
313 dump_printf(": unhandled!\n");
314 return 0;
315}
316
317static int process_finished_round_stub(union perf_event *event __used,
318 struct perf_session *session __used,
319 struct perf_event_ops *ops __used)
320{
321 dump_printf(": unhandled!\n");
322 return 0;
323}
324
325static int process_finished_round(union perf_event *event,
326 struct perf_session *session,
327 struct perf_event_ops *ops);
328
329static void perf_event_ops__fill_defaults(struct perf_event_ops *handler)
330{
331 if (handler->sample == NULL)
332 handler->sample = process_event_sample_stub;
333 if (handler->mmap == NULL)
334 handler->mmap = process_event_stub;
335 if (handler->comm == NULL)
336 handler->comm = process_event_stub;
337 if (handler->fork == NULL)
338 handler->fork = process_event_stub;
339 if (handler->exit == NULL)
340 handler->exit = process_event_stub;
341 if (handler->lost == NULL)
342 handler->lost = perf_event__process_lost;
343 if (handler->read == NULL)
344 handler->read = process_event_stub;
345 if (handler->throttle == NULL)
346 handler->throttle = process_event_stub;
347 if (handler->unthrottle == NULL)
348 handler->unthrottle = process_event_stub;
349 if (handler->attr == NULL)
350 handler->attr = process_event_synth_stub;
351 if (handler->event_type == NULL)
352 handler->event_type = process_event_synth_stub;
353 if (handler->tracing_data == NULL)
354 handler->tracing_data = process_event_synth_stub;
355 if (handler->build_id == NULL)
356 handler->build_id = process_event_synth_stub;
357 if (handler->finished_round == NULL) {
358 if (handler->ordered_samples)
359 handler->finished_round = process_finished_round;
360 else
361 handler->finished_round = process_finished_round_stub;
362 }
363}
364
365void mem_bswap_64(void *src, int byte_size)
366{
367 u64 *m = src;
368
369 while (byte_size > 0) {
370 *m = bswap_64(*m);
371 byte_size -= sizeof(u64);
372 ++m;
373 }
374}
375
376static void perf_event__all64_swap(union perf_event *event)
377{
378 struct perf_event_header *hdr = &event->header;
379 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
380}
381
382static void perf_event__comm_swap(union perf_event *event)
383{
384 event->comm.pid = bswap_32(event->comm.pid);
385 event->comm.tid = bswap_32(event->comm.tid);
386}
387
388static void perf_event__mmap_swap(union perf_event *event)
389{
390 event->mmap.pid = bswap_32(event->mmap.pid);
391 event->mmap.tid = bswap_32(event->mmap.tid);
392 event->mmap.start = bswap_64(event->mmap.start);
393 event->mmap.len = bswap_64(event->mmap.len);
394 event->mmap.pgoff = bswap_64(event->mmap.pgoff);
395}
396
397static void perf_event__task_swap(union perf_event *event)
398{
399 event->fork.pid = bswap_32(event->fork.pid);
400 event->fork.tid = bswap_32(event->fork.tid);
401 event->fork.ppid = bswap_32(event->fork.ppid);
402 event->fork.ptid = bswap_32(event->fork.ptid);
403 event->fork.time = bswap_64(event->fork.time);
404}
405
406static void perf_event__read_swap(union perf_event *event)
407{
408 event->read.pid = bswap_32(event->read.pid);
409 event->read.tid = bswap_32(event->read.tid);
410 event->read.value = bswap_64(event->read.value);
411 event->read.time_enabled = bswap_64(event->read.time_enabled);
412 event->read.time_running = bswap_64(event->read.time_running);
413 event->read.id = bswap_64(event->read.id);
414}
415
416/* exported for swapping attributes in file header */
417void perf_event__attr_swap(struct perf_event_attr *attr)
418{
419 attr->type = bswap_32(attr->type);
420 attr->size = bswap_32(attr->size);
421 attr->config = bswap_64(attr->config);
422 attr->sample_period = bswap_64(attr->sample_period);
423 attr->sample_type = bswap_64(attr->sample_type);
424 attr->read_format = bswap_64(attr->read_format);
425 attr->wakeup_events = bswap_32(attr->wakeup_events);
426 attr->bp_type = bswap_32(attr->bp_type);
427 attr->bp_addr = bswap_64(attr->bp_addr);
428 attr->bp_len = bswap_64(attr->bp_len);
429}
430
431static void perf_event__hdr_attr_swap(union perf_event *event)
432{
433 size_t size;
434
435 perf_event__attr_swap(&event->attr.attr);
436
437 size = event->header.size;
438 size -= (void *)&event->attr.id - (void *)event;
439 mem_bswap_64(event->attr.id, size);
440}
441
442static void perf_event__event_type_swap(union perf_event *event)
443{
444 event->event_type.event_type.event_id =
445 bswap_64(event->event_type.event_type.event_id);
446}
447
448static void perf_event__tracing_data_swap(union perf_event *event)
449{
450 event->tracing_data.size = bswap_32(event->tracing_data.size);
451}
452
453typedef void (*perf_event__swap_op)(union perf_event *event);
454
455static perf_event__swap_op perf_event__swap_ops[] = {
456 [PERF_RECORD_MMAP] = perf_event__mmap_swap,
457 [PERF_RECORD_COMM] = perf_event__comm_swap,
458 [PERF_RECORD_FORK] = perf_event__task_swap,
459 [PERF_RECORD_EXIT] = perf_event__task_swap,
460 [PERF_RECORD_LOST] = perf_event__all64_swap,
461 [PERF_RECORD_READ] = perf_event__read_swap,
462 [PERF_RECORD_SAMPLE] = perf_event__all64_swap,
463 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap,
464 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap,
465 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
466 [PERF_RECORD_HEADER_BUILD_ID] = NULL,
467 [PERF_RECORD_HEADER_MAX] = NULL,
468};
469
470struct sample_queue {
471 u64 timestamp;
472 u64 file_offset;
473 union perf_event *event;
474 struct list_head list;
475};
476
477static void perf_session_free_sample_buffers(struct perf_session *session)
478{
479 struct ordered_samples *os = &session->ordered_samples;
480
481 while (!list_empty(&os->to_free)) {
482 struct sample_queue *sq;
483
484 sq = list_entry(os->to_free.next, struct sample_queue, list);
485 list_del(&sq->list);
486 free(sq);
487 }
488}
489
490static int perf_session_deliver_event(struct perf_session *session,
491 union perf_event *event,
492 struct perf_sample *sample,
493 struct perf_event_ops *ops,
494 u64 file_offset);
495
496static void flush_sample_queue(struct perf_session *s,
497 struct perf_event_ops *ops)
498{
499 struct ordered_samples *os = &s->ordered_samples;
500 struct list_head *head = &os->samples;
501 struct sample_queue *tmp, *iter;
502 struct perf_sample sample;
503 u64 limit = os->next_flush;
504 u64 last_ts = os->last_sample ? os->last_sample->timestamp : 0ULL;
505 int ret;
506
507 if (!ops->ordered_samples || !limit)
508 return;
509
510 list_for_each_entry_safe(iter, tmp, head, list) {
511 if (iter->timestamp > limit)
512 break;
513
514 ret = perf_session__parse_sample(s, iter->event, &sample);
515 if (ret)
516 pr_err("Can't parse sample, err = %d\n", ret);
517 else
518 perf_session_deliver_event(s, iter->event, &sample, ops,
519 iter->file_offset);
520
521 os->last_flush = iter->timestamp;
522 list_del(&iter->list);
523 list_add(&iter->list, &os->sample_cache);
524 }
525
526 if (list_empty(head)) {
527 os->last_sample = NULL;
528 } else if (last_ts <= limit) {
529 os->last_sample =
530 list_entry(head->prev, struct sample_queue, list);
531 }
532}
533
534/*
535 * When perf record finishes a pass on every buffers, it records this pseudo
536 * event.
537 * We record the max timestamp t found in the pass n.
538 * Assuming these timestamps are monotonic across cpus, we know that if
539 * a buffer still has events with timestamps below t, they will be all
540 * available and then read in the pass n + 1.
541 * Hence when we start to read the pass n + 2, we can safely flush every
542 * events with timestamps below t.
543 *
544 * ============ PASS n =================
545 * CPU 0 | CPU 1
546 * |
547 * cnt1 timestamps | cnt2 timestamps
548 * 1 | 2
549 * 2 | 3
550 * - | 4 <--- max recorded
551 *
552 * ============ PASS n + 1 ==============
553 * CPU 0 | CPU 1
554 * |
555 * cnt1 timestamps | cnt2 timestamps
556 * 3 | 5
557 * 4 | 6
558 * 5 | 7 <---- max recorded
559 *
560 * Flush every events below timestamp 4
561 *
562 * ============ PASS n + 2 ==============
563 * CPU 0 | CPU 1
564 * |
565 * cnt1 timestamps | cnt2 timestamps
566 * 6 | 8
567 * 7 | 9
568 * - | 10
569 *
570 * Flush every events below timestamp 7
571 * etc...
572 */
573static int process_finished_round(union perf_event *event __used,
574 struct perf_session *session,
575 struct perf_event_ops *ops)
576{
577 flush_sample_queue(session, ops);
578 session->ordered_samples.next_flush = session->ordered_samples.max_timestamp;
579
580 return 0;
581}
582
583/* The queue is ordered by time */
584static void __queue_event(struct sample_queue *new, struct perf_session *s)
585{
586 struct ordered_samples *os = &s->ordered_samples;
587 struct sample_queue *sample = os->last_sample;
588 u64 timestamp = new->timestamp;
589 struct list_head *p;
590
591 os->last_sample = new;
592
593 if (!sample) {
594 list_add(&new->list, &os->samples);
595 os->max_timestamp = timestamp;
596 return;
597 }
598
599 /*
600 * last_sample might point to some random place in the list as it's
601 * the last queued event. We expect that the new event is close to
602 * this.
603 */
604 if (sample->timestamp <= timestamp) {
605 while (sample->timestamp <= timestamp) {
606 p = sample->list.next;
607 if (p == &os->samples) {
608 list_add_tail(&new->list, &os->samples);
609 os->max_timestamp = timestamp;
610 return;
611 }
612 sample = list_entry(p, struct sample_queue, list);
613 }
614 list_add_tail(&new->list, &sample->list);
615 } else {
616 while (sample->timestamp > timestamp) {
617 p = sample->list.prev;
618 if (p == &os->samples) {
619 list_add(&new->list, &os->samples);
620 return;
621 }
622 sample = list_entry(p, struct sample_queue, list);
623 }
624 list_add(&new->list, &sample->list);
625 }
626}
627
628#define MAX_SAMPLE_BUFFER (64 * 1024 / sizeof(struct sample_queue))
629
630static int perf_session_queue_event(struct perf_session *s, union perf_event *event,
631 struct perf_sample *sample, u64 file_offset)
632{
633 struct ordered_samples *os = &s->ordered_samples;
634 struct list_head *sc = &os->sample_cache;
635 u64 timestamp = sample->time;
636 struct sample_queue *new;
637
638 if (!timestamp || timestamp == ~0ULL)
639 return -ETIME;
640
641 if (timestamp < s->ordered_samples.last_flush) {
642 printf("Warning: Timestamp below last timeslice flush\n");
643 return -EINVAL;
644 }
645
646 if (!list_empty(sc)) {
647 new = list_entry(sc->next, struct sample_queue, list);
648 list_del(&new->list);
649 } else if (os->sample_buffer) {
650 new = os->sample_buffer + os->sample_buffer_idx;
651 if (++os->sample_buffer_idx == MAX_SAMPLE_BUFFER)
652 os->sample_buffer = NULL;
653 } else {
654 os->sample_buffer = malloc(MAX_SAMPLE_BUFFER * sizeof(*new));
655 if (!os->sample_buffer)
656 return -ENOMEM;
657 list_add(&os->sample_buffer->list, &os->to_free);
658 os->sample_buffer_idx = 2;
659 new = os->sample_buffer + 1;
660 }
661
662 new->timestamp = timestamp;
663 new->file_offset = file_offset;
664 new->event = event;
665
666 __queue_event(new, s);
667
668 return 0;
669}
670
671static void callchain__printf(struct perf_sample *sample)
672{
673 unsigned int i;
674
675 printf("... chain: nr:%" PRIu64 "\n", sample->callchain->nr);
676
677 for (i = 0; i < sample->callchain->nr; i++)
678 printf("..... %2d: %016" PRIx64 "\n",
679 i, sample->callchain->ips[i]);
680}
681
682static void perf_session__print_tstamp(struct perf_session *session,
683 union perf_event *event,
684 struct perf_sample *sample)
685{
686 if (event->header.type != PERF_RECORD_SAMPLE &&
687 !session->sample_id_all) {
688 fputs("-1 -1 ", stdout);
689 return;
690 }
691
692 if ((session->sample_type & PERF_SAMPLE_CPU))
693 printf("%u ", sample->cpu);
694
695 if (session->sample_type & PERF_SAMPLE_TIME)
696 printf("%" PRIu64 " ", sample->time);
697}
698
699static void dump_event(struct perf_session *session, union perf_event *event,
700 u64 file_offset, struct perf_sample *sample)
701{
702 if (!dump_trace)
703 return;
704
705 printf("\n%#" PRIx64 " [%#x]: event: %d\n",
706 file_offset, event->header.size, event->header.type);
707
708 trace_event(event);
709
710 if (sample)
711 perf_session__print_tstamp(session, event, sample);
712
713 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
714 event->header.size, perf_event__name(event->header.type));
715}
716
717static void dump_sample(struct perf_session *session, union perf_event *event,
718 struct perf_sample *sample)
719{
720 if (!dump_trace)
721 return;
722
723 printf("(IP, %d): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
724 event->header.misc, sample->pid, sample->tid, sample->ip,
725 sample->period, sample->addr);
726
727 if (session->sample_type & PERF_SAMPLE_CALLCHAIN)
728 callchain__printf(sample);
729}
730
731static int perf_session_deliver_event(struct perf_session *session,
732 union perf_event *event,
733 struct perf_sample *sample,
734 struct perf_event_ops *ops,
735 u64 file_offset)
736{
737 struct perf_evsel *evsel;
738
739 dump_event(session, event, file_offset, sample);
740
741 switch (event->header.type) {
742 case PERF_RECORD_SAMPLE:
743 dump_sample(session, event, sample);
744 evsel = perf_evlist__id2evsel(session->evlist, sample->id);
745 if (evsel == NULL) {
746 ++session->hists.stats.nr_unknown_id;
747 return -1;
748 }
749 return ops->sample(event, sample, evsel, session);
750 case PERF_RECORD_MMAP:
751 return ops->mmap(event, sample, session);
752 case PERF_RECORD_COMM:
753 return ops->comm(event, sample, session);
754 case PERF_RECORD_FORK:
755 return ops->fork(event, sample, session);
756 case PERF_RECORD_EXIT:
757 return ops->exit(event, sample, session);
758 case PERF_RECORD_LOST:
759 return ops->lost(event, sample, session);
760 case PERF_RECORD_READ:
761 return ops->read(event, sample, session);
762 case PERF_RECORD_THROTTLE:
763 return ops->throttle(event, sample, session);
764 case PERF_RECORD_UNTHROTTLE:
765 return ops->unthrottle(event, sample, session);
766 default:
767 ++session->hists.stats.nr_unknown_events;
768 return -1;
769 }
770}
771
772static int perf_session__preprocess_sample(struct perf_session *session,
773 union perf_event *event, struct perf_sample *sample)
774{
775 if (event->header.type != PERF_RECORD_SAMPLE ||
776 !(session->sample_type & PERF_SAMPLE_CALLCHAIN))
777 return 0;
778
779 if (!ip_callchain__valid(sample->callchain, event)) {
780 pr_debug("call-chain problem with event, skipping it.\n");
781 ++session->hists.stats.nr_invalid_chains;
782 session->hists.stats.total_invalid_chains += sample->period;
783 return -EINVAL;
784 }
785 return 0;
786}
787
788static int perf_session__process_user_event(struct perf_session *session, union perf_event *event,
789 struct perf_event_ops *ops, u64 file_offset)
790{
791 dump_event(session, event, file_offset, NULL);
792
793 /* These events are processed right away */
794 switch (event->header.type) {
795 case PERF_RECORD_HEADER_ATTR:
796 return ops->attr(event, session);
797 case PERF_RECORD_HEADER_EVENT_TYPE:
798 return ops->event_type(event, session);
799 case PERF_RECORD_HEADER_TRACING_DATA:
800 /* setup for reading amidst mmap */
801 lseek(session->fd, file_offset, SEEK_SET);
802 return ops->tracing_data(event, session);
803 case PERF_RECORD_HEADER_BUILD_ID:
804 return ops->build_id(event, session);
805 case PERF_RECORD_FINISHED_ROUND:
806 return ops->finished_round(event, session, ops);
807 default:
808 return -EINVAL;
809 }
810}
811
812static int perf_session__process_event(struct perf_session *session,
813 union perf_event *event,
814 struct perf_event_ops *ops,
815 u64 file_offset)
816{
817 struct perf_sample sample;
818 int ret;
819
820 if (session->header.needs_swap &&
821 perf_event__swap_ops[event->header.type])
822 perf_event__swap_ops[event->header.type](event);
823
824 if (event->header.type >= PERF_RECORD_HEADER_MAX)
825 return -EINVAL;
826
827 hists__inc_nr_events(&session->hists, event->header.type);
828
829 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
830 return perf_session__process_user_event(session, event, ops, file_offset);
831
832 /*
833 * For all kernel events we get the sample data
834 */
835 ret = perf_session__parse_sample(session, event, &sample);
836 if (ret)
837 return ret;
838
839 /* Preprocess sample records - precheck callchains */
840 if (perf_session__preprocess_sample(session, event, &sample))
841 return 0;
842
843 if (ops->ordered_samples) {
844 ret = perf_session_queue_event(session, event, &sample,
845 file_offset);
846 if (ret != -ETIME)
847 return ret;
848 }
849
850 return perf_session_deliver_event(session, event, &sample, ops,
851 file_offset);
852}
853
854void perf_event_header__bswap(struct perf_event_header *self)
855{
856 self->type = bswap_32(self->type);
857 self->misc = bswap_16(self->misc);
858 self->size = bswap_16(self->size);
859}
860
861static struct thread *perf_session__register_idle_thread(struct perf_session *self)
862{
863 struct thread *thread = perf_session__findnew(self, 0);
864
865 if (thread == NULL || thread__set_comm(thread, "swapper")) {
866 pr_err("problem inserting idle task.\n");
867 thread = NULL;
868 }
869
870 return thread;
871}
872
873static void perf_session__warn_about_errors(const struct perf_session *session,
874 const struct perf_event_ops *ops)
875{
876 if (ops->lost == perf_event__process_lost &&
877 session->hists.stats.total_lost != 0) {
878 ui__warning("Processed %" PRIu64 " events and LOST %" PRIu64
879 "!\n\nCheck IO/CPU overload!\n\n",
880 session->hists.stats.total_period,
881 session->hists.stats.total_lost);
882 }
883
884 if (session->hists.stats.nr_unknown_events != 0) {
885 ui__warning("Found %u unknown events!\n\n"
886 "Is this an older tool processing a perf.data "
887 "file generated by a more recent tool?\n\n"
888 "If that is not the case, consider "
889 "reporting to linux-kernel@vger.kernel.org.\n\n",
890 session->hists.stats.nr_unknown_events);
891 }
892
893 if (session->hists.stats.nr_unknown_id != 0) {
894 ui__warning("%u samples with id not present in the header\n",
895 session->hists.stats.nr_unknown_id);
896 }
897
898 if (session->hists.stats.nr_invalid_chains != 0) {
899 ui__warning("Found invalid callchains!\n\n"
900 "%u out of %u events were discarded for this reason.\n\n"
901 "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
902 session->hists.stats.nr_invalid_chains,
903 session->hists.stats.nr_events[PERF_RECORD_SAMPLE]);
904 }
905}
906
907#define session_done() (*(volatile int *)(&session_done))
908volatile int session_done;
909
910static int __perf_session__process_pipe_events(struct perf_session *self,
911 struct perf_event_ops *ops)
912{
913 union perf_event event;
914 uint32_t size;
915 int skip = 0;
916 u64 head;
917 int err;
918 void *p;
919
920 perf_event_ops__fill_defaults(ops);
921
922 head = 0;
923more:
924 err = readn(self->fd, &event, sizeof(struct perf_event_header));
925 if (err <= 0) {
926 if (err == 0)
927 goto done;
928
929 pr_err("failed to read event header\n");
930 goto out_err;
931 }
932
933 if (self->header.needs_swap)
934 perf_event_header__bswap(&event.header);
935
936 size = event.header.size;
937 if (size == 0)
938 size = 8;
939
940 p = &event;
941 p += sizeof(struct perf_event_header);
942
943 if (size - sizeof(struct perf_event_header)) {
944 err = readn(self->fd, p, size - sizeof(struct perf_event_header));
945 if (err <= 0) {
946 if (err == 0) {
947 pr_err("unexpected end of event stream\n");
948 goto done;
949 }
950
951 pr_err("failed to read event data\n");
952 goto out_err;
953 }
954 }
955
956 if (size == 0 ||
957 (skip = perf_session__process_event(self, &event, ops, head)) < 0) {
958 dump_printf("%#" PRIx64 " [%#x]: skipping unknown header type: %d\n",
959 head, event.header.size, event.header.type);
960 /*
961 * assume we lost track of the stream, check alignment, and
962 * increment a single u64 in the hope to catch on again 'soon'.
963 */
964 if (unlikely(head & 7))
965 head &= ~7ULL;
966
967 size = 8;
968 }
969
970 head += size;
971
972 if (skip > 0)
973 head += skip;
974
975 if (!session_done())
976 goto more;
977done:
978 err = 0;
979out_err:
980 perf_session__warn_about_errors(self, ops);
981 perf_session_free_sample_buffers(self);
982 return err;
983}
984
985static union perf_event *
986fetch_mmaped_event(struct perf_session *session,
987 u64 head, size_t mmap_size, char *buf)
988{
989 union perf_event *event;
990
991 /*
992 * Ensure we have enough space remaining to read
993 * the size of the event in the headers.
994 */
995 if (head + sizeof(event->header) > mmap_size)
996 return NULL;
997
998 event = (union perf_event *)(buf + head);
999
1000 if (session->header.needs_swap)
1001 perf_event_header__bswap(&event->header);
1002
1003 if (head + event->header.size > mmap_size)
1004 return NULL;
1005
1006 return event;
1007}
1008
1009int __perf_session__process_events(struct perf_session *session,
1010 u64 data_offset, u64 data_size,
1011 u64 file_size, struct perf_event_ops *ops)
1012{
1013 u64 head, page_offset, file_offset, file_pos, progress_next;
1014 int err, mmap_prot, mmap_flags, map_idx = 0;
1015 struct ui_progress *progress;
1016 size_t page_size, mmap_size;
1017 char *buf, *mmaps[8];
1018 union perf_event *event;
1019 uint32_t size;
1020
1021 perf_event_ops__fill_defaults(ops);
1022
1023 page_size = sysconf(_SC_PAGESIZE);
1024
1025 page_offset = page_size * (data_offset / page_size);
1026 file_offset = page_offset;
1027 head = data_offset - page_offset;
1028
1029 if (data_offset + data_size < file_size)
1030 file_size = data_offset + data_size;
1031
1032 progress_next = file_size / 16;
1033 progress = ui_progress__new("Processing events...", file_size);
1034 if (progress == NULL)
1035 return -1;
1036
1037 mmap_size = session->mmap_window;
1038 if (mmap_size > file_size)
1039 mmap_size = file_size;
1040
1041 memset(mmaps, 0, sizeof(mmaps));
1042
1043 mmap_prot = PROT_READ;
1044 mmap_flags = MAP_SHARED;
1045
1046 if (session->header.needs_swap) {
1047 mmap_prot |= PROT_WRITE;
1048 mmap_flags = MAP_PRIVATE;
1049 }
1050remap:
1051 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, session->fd,
1052 file_offset);
1053 if (buf == MAP_FAILED) {
1054 pr_err("failed to mmap file\n");
1055 err = -errno;
1056 goto out_err;
1057 }
1058 mmaps[map_idx] = buf;
1059 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
1060 file_pos = file_offset + head;
1061
1062more:
1063 event = fetch_mmaped_event(session, head, mmap_size, buf);
1064 if (!event) {
1065 if (mmaps[map_idx]) {
1066 munmap(mmaps[map_idx], mmap_size);
1067 mmaps[map_idx] = NULL;
1068 }
1069
1070 page_offset = page_size * (head / page_size);
1071 file_offset += page_offset;
1072 head -= page_offset;
1073 goto remap;
1074 }
1075
1076 size = event->header.size;
1077
1078 if (size == 0 ||
1079 perf_session__process_event(session, event, ops, file_pos) < 0) {
1080 dump_printf("%#" PRIx64 " [%#x]: skipping unknown header type: %d\n",
1081 file_offset + head, event->header.size,
1082 event->header.type);
1083 /*
1084 * assume we lost track of the stream, check alignment, and
1085 * increment a single u64 in the hope to catch on again 'soon'.
1086 */
1087 if (unlikely(head & 7))
1088 head &= ~7ULL;
1089
1090 size = 8;
1091 }
1092
1093 head += size;
1094 file_pos += size;
1095
1096 if (file_pos >= progress_next) {
1097 progress_next += file_size / 16;
1098 ui_progress__update(progress, file_pos);
1099 }
1100
1101 if (file_pos < file_size)
1102 goto more;
1103
1104 err = 0;
1105 /* do the final flush for ordered samples */
1106 session->ordered_samples.next_flush = ULLONG_MAX;
1107 flush_sample_queue(session, ops);
1108out_err:
1109 ui_progress__delete(progress);
1110 perf_session__warn_about_errors(session, ops);
1111 perf_session_free_sample_buffers(session);
1112 return err;
1113}
1114
1115int perf_session__process_events(struct perf_session *self,
1116 struct perf_event_ops *ops)
1117{
1118 int err;
1119
1120 if (perf_session__register_idle_thread(self) == NULL)
1121 return -ENOMEM;
1122
1123 if (!self->fd_pipe)
1124 err = __perf_session__process_events(self,
1125 self->header.data_offset,
1126 self->header.data_size,
1127 self->size, ops);
1128 else
1129 err = __perf_session__process_pipe_events(self, ops);
1130
1131 return err;
1132}
1133
1134bool perf_session__has_traces(struct perf_session *self, const char *msg)
1135{
1136 if (!(self->sample_type & PERF_SAMPLE_RAW)) {
1137 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
1138 return false;
1139 }
1140
1141 return true;
1142}
1143
1144int perf_session__set_kallsyms_ref_reloc_sym(struct map **maps,
1145 const char *symbol_name,
1146 u64 addr)
1147{
1148 char *bracket;
1149 enum map_type i;
1150 struct ref_reloc_sym *ref;
1151
1152 ref = zalloc(sizeof(struct ref_reloc_sym));
1153 if (ref == NULL)
1154 return -ENOMEM;
1155
1156 ref->name = strdup(symbol_name);
1157 if (ref->name == NULL) {
1158 free(ref);
1159 return -ENOMEM;
1160 }
1161
1162 bracket = strchr(ref->name, ']');
1163 if (bracket)
1164 *bracket = '\0';
1165
1166 ref->addr = addr;
1167
1168 for (i = 0; i < MAP__NR_TYPES; ++i) {
1169 struct kmap *kmap = map__kmap(maps[i]);
1170 kmap->ref_reloc_sym = ref;
1171 }
1172
1173 return 0;
1174}
1175
1176size_t perf_session__fprintf_dsos(struct perf_session *self, FILE *fp)
1177{
1178 return __dsos__fprintf(&self->host_machine.kernel_dsos, fp) +
1179 __dsos__fprintf(&self->host_machine.user_dsos, fp) +
1180 machines__fprintf_dsos(&self->machines, fp);
1181}
1182
1183size_t perf_session__fprintf_dsos_buildid(struct perf_session *self, FILE *fp,
1184 bool with_hits)
1185{
1186 size_t ret = machine__fprintf_dsos_buildid(&self->host_machine, fp, with_hits);
1187 return ret + machines__fprintf_dsos_buildid(&self->machines, fp, with_hits);
1188}
1189
1190size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
1191{
1192 struct perf_evsel *pos;
1193 size_t ret = fprintf(fp, "Aggregated stats:\n");
1194
1195 ret += hists__fprintf_nr_events(&session->hists, fp);
1196
1197 list_for_each_entry(pos, &session->evlist->entries, node) {
1198 ret += fprintf(fp, "%s stats:\n", event_name(pos));
1199 ret += hists__fprintf_nr_events(&pos->hists, fp);
1200 }
1201
1202 return ret;
1203}
1204
1205struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session,
1206 unsigned int type)
1207{
1208 struct perf_evsel *pos;
1209
1210 list_for_each_entry(pos, &session->evlist->entries, node) {
1211 if (pos->attr.type == type)
1212 return pos;
1213 }
1214 return NULL;
1215}
1216
1217void perf_session__print_ip(union perf_event *event,
1218 struct perf_sample *sample,
1219 struct perf_session *session,
1220 int print_sym, int print_dso)
1221{
1222 struct addr_location al;
1223 const char *symname, *dsoname;
1224 struct callchain_cursor *cursor = &session->callchain_cursor;
1225 struct callchain_cursor_node *node;
1226
1227 if (perf_event__preprocess_sample(event, session, &al, sample,
1228 NULL) < 0) {
1229 error("problem processing %d event, skipping it.\n",
1230 event->header.type);
1231 return;
1232 }
1233
1234 if (symbol_conf.use_callchain && sample->callchain) {
1235
1236 if (perf_session__resolve_callchain(session, al.thread,
1237 sample->callchain, NULL) != 0) {
1238 if (verbose)
1239 error("Failed to resolve callchain. Skipping\n");
1240 return;
1241 }
1242 callchain_cursor_commit(cursor);
1243
1244 while (1) {
1245 node = callchain_cursor_current(cursor);
1246 if (!node)
1247 break;
1248
1249 printf("\t%16" PRIx64, node->ip);
1250 if (print_sym) {
1251 if (node->sym && node->sym->name)
1252 symname = node->sym->name;
1253 else
1254 symname = "";
1255
1256 printf(" %s", symname);
1257 }
1258 if (print_dso) {
1259 if (node->map && node->map->dso && node->map->dso->name)
1260 dsoname = node->map->dso->name;
1261 else
1262 dsoname = "";
1263
1264 printf(" (%s)", dsoname);
1265 }
1266 printf("\n");
1267
1268 callchain_cursor_advance(cursor);
1269 }
1270
1271 } else {
1272 printf("%16" PRIx64, sample->ip);
1273 if (print_sym) {
1274 if (al.sym && al.sym->name)
1275 symname = al.sym->name;
1276 else
1277 symname = "";
1278
1279 printf(" %s", symname);
1280 }
1281
1282 if (print_dso) {
1283 if (al.map && al.map->dso && al.map->dso->name)
1284 dsoname = al.map->dso->name;
1285 else
1286 dsoname = "";
1287
1288 printf(" (%s)", dsoname);
1289 }
1290 }
1291}
1292
1293int perf_session__cpu_bitmap(struct perf_session *session,
1294 const char *cpu_list, unsigned long *cpu_bitmap)
1295{
1296 int i;
1297 struct cpu_map *map;
1298
1299 for (i = 0; i < PERF_TYPE_MAX; ++i) {
1300 struct perf_evsel *evsel;
1301
1302 evsel = perf_session__find_first_evtype(session, i);
1303 if (!evsel)
1304 continue;
1305
1306 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) {
1307 pr_err("File does not contain CPU events. "
1308 "Remove -c option to proceed.\n");
1309 return -1;
1310 }
1311 }
1312
1313 map = cpu_map__new(cpu_list);
1314
1315 for (i = 0; i < map->nr; i++) {
1316 int cpu = map->map[i];
1317
1318 if (cpu >= MAX_NR_CPUS) {
1319 pr_err("Requested CPU %d too large. "
1320 "Consider raising MAX_NR_CPUS\n", cpu);
1321 return -1;
1322 }
1323
1324 set_bit(cpu, cpu_bitmap);
1325 }
1326
1327 return 0;
1328}