1 // SPDX-License-Identifier: GPL-2.0
2 #include <errno.h>
3 #include <inttypes.h>
4 #include <linux/err.h>
5 #include <linux/kernel.h>
6 #include <linux/zalloc.h>
7 #include <api/fs/fs.h>
8 
9 #include <byteswap.h>
10 #include <unistd.h>
11 #include <sys/types.h>
12 #include <sys/mman.h>
13 #include <perf/cpumap.h>
14 
15 #include "map_symbol.h"
16 #include "branch.h"
17 #include "debug.h"
18 #include "env.h"
19 #include "evlist.h"
20 #include "evsel.h"
21 #include "memswap.h"
22 #include "map.h"
23 #include "symbol.h"
24 #include "session.h"
25 #include "tool.h"
26 #include "perf_regs.h"
27 #include "asm/bug.h"
28 #include "auxtrace.h"
29 #include "thread.h"
30 #include "thread-stack.h"
31 #include "sample-raw.h"
32 #include "stat.h"
33 #include "tsc.h"
34 #include "ui/progress.h"
35 #include "../perf.h"
36 #include "arch/common.h"
37 #include "units.h"
38 #include <internal/lib.h>
39 
40 #ifdef HAVE_ZSTD_SUPPORT
perf_session__process_compressed_event(struct perf_session * session,union perf_event * event,u64 file_offset,const char * file_path)41 static int perf_session__process_compressed_event(struct perf_session *session,
42 						  union perf_event *event, u64 file_offset,
43 						  const char *file_path)
44 {
45 	void *src;
46 	size_t decomp_size, src_size;
47 	u64 decomp_last_rem = 0;
48 	size_t mmap_len, decomp_len = session->header.env.comp_mmap_len;
49 	struct decomp *decomp, *decomp_last = session->active_decomp->decomp_last;
50 
51 	if (decomp_last) {
52 		decomp_last_rem = decomp_last->size - decomp_last->head;
53 		decomp_len += decomp_last_rem;
54 	}
55 
56 	mmap_len = sizeof(struct decomp) + decomp_len;
57 	decomp = mmap(NULL, mmap_len, PROT_READ|PROT_WRITE,
58 		      MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
59 	if (decomp == MAP_FAILED) {
60 		pr_err("Couldn't allocate memory for decompression\n");
61 		return -1;
62 	}
63 
64 	decomp->file_pos = file_offset;
65 	decomp->file_path = file_path;
66 	decomp->mmap_len = mmap_len;
67 	decomp->head = 0;
68 
69 	if (decomp_last_rem) {
70 		memcpy(decomp->data, &(decomp_last->data[decomp_last->head]), decomp_last_rem);
71 		decomp->size = decomp_last_rem;
72 	}
73 
74 	src = (void *)event + sizeof(struct perf_record_compressed);
75 	src_size = event->pack.header.size - sizeof(struct perf_record_compressed);
76 
77 	decomp_size = zstd_decompress_stream(session->active_decomp->zstd_decomp, src, src_size,
78 				&(decomp->data[decomp_last_rem]), decomp_len - decomp_last_rem);
79 	if (!decomp_size) {
80 		munmap(decomp, mmap_len);
81 		pr_err("Couldn't decompress data\n");
82 		return -1;
83 	}
84 
85 	decomp->size += decomp_size;
86 
87 	if (session->active_decomp->decomp == NULL)
88 		session->active_decomp->decomp = decomp;
89 	else
90 		session->active_decomp->decomp_last->next = decomp;
91 
92 	session->active_decomp->decomp_last = decomp;
93 
94 	pr_debug("decomp (B): %zd to %zd\n", src_size, decomp_size);
95 
96 	return 0;
97 }
98 #else /* !HAVE_ZSTD_SUPPORT */
99 #define perf_session__process_compressed_event perf_session__process_compressed_event_stub
100 #endif
101 
102 static int perf_session__deliver_event(struct perf_session *session,
103 				       union perf_event *event,
104 				       struct perf_tool *tool,
105 				       u64 file_offset,
106 				       const char *file_path);
107 
perf_session__open(struct perf_session * session,int repipe_fd)108 static int perf_session__open(struct perf_session *session, int repipe_fd)
109 {
110 	struct perf_data *data = session->data;
111 
112 	if (perf_session__read_header(session, repipe_fd) < 0) {
113 		pr_err("incompatible file format (rerun with -v to learn more)\n");
114 		return -1;
115 	}
116 
117 	if (perf_data__is_pipe(data))
118 		return 0;
119 
120 	if (perf_header__has_feat(&session->header, HEADER_STAT))
121 		return 0;
122 
123 	if (!evlist__valid_sample_type(session->evlist)) {
124 		pr_err("non matching sample_type\n");
125 		return -1;
126 	}
127 
128 	if (!evlist__valid_sample_id_all(session->evlist)) {
129 		pr_err("non matching sample_id_all\n");
130 		return -1;
131 	}
132 
133 	if (!evlist__valid_read_format(session->evlist)) {
134 		pr_err("non matching read_format\n");
135 		return -1;
136 	}
137 
138 	return 0;
139 }
140 
perf_session__set_id_hdr_size(struct perf_session * session)141 void perf_session__set_id_hdr_size(struct perf_session *session)
142 {
143 	u16 id_hdr_size = evlist__id_hdr_size(session->evlist);
144 
145 	machines__set_id_hdr_size(&session->machines, id_hdr_size);
146 }
147 
perf_session__create_kernel_maps(struct perf_session * session)148 int perf_session__create_kernel_maps(struct perf_session *session)
149 {
150 	int ret = machine__create_kernel_maps(&session->machines.host);
151 
152 	if (ret >= 0)
153 		ret = machines__create_guest_kernel_maps(&session->machines);
154 	return ret;
155 }
156 
perf_session__destroy_kernel_maps(struct perf_session * session)157 static void perf_session__destroy_kernel_maps(struct perf_session *session)
158 {
159 	machines__destroy_kernel_maps(&session->machines);
160 }
161 
perf_session__has_comm_exec(struct perf_session * session)162 static bool perf_session__has_comm_exec(struct perf_session *session)
163 {
164 	struct evsel *evsel;
165 
166 	evlist__for_each_entry(session->evlist, evsel) {
167 		if (evsel->core.attr.comm_exec)
168 			return true;
169 	}
170 
171 	return false;
172 }
173 
perf_session__set_comm_exec(struct perf_session * session)174 static void perf_session__set_comm_exec(struct perf_session *session)
175 {
176 	bool comm_exec = perf_session__has_comm_exec(session);
177 
178 	machines__set_comm_exec(&session->machines, comm_exec);
179 }
180 
ordered_events__deliver_event(struct ordered_events * oe,struct ordered_event * event)181 static int ordered_events__deliver_event(struct ordered_events *oe,
182 					 struct ordered_event *event)
183 {
184 	struct perf_session *session = container_of(oe, struct perf_session,
185 						    ordered_events);
186 
187 	return perf_session__deliver_event(session, event->event,
188 					   session->tool, event->file_offset,
189 					   event->file_path);
190 }
191 
__perf_session__new(struct perf_data * data,bool repipe,int repipe_fd,struct perf_tool * tool)192 struct perf_session *__perf_session__new(struct perf_data *data,
193 					 bool repipe, int repipe_fd,
194 					 struct perf_tool *tool)
195 {
196 	int ret = -ENOMEM;
197 	struct perf_session *session = zalloc(sizeof(*session));
198 
199 	if (!session)
200 		goto out;
201 
202 	session->repipe = repipe;
203 	session->tool   = tool;
204 	session->decomp_data.zstd_decomp = &session->zstd_data;
205 	session->active_decomp = &session->decomp_data;
206 	INIT_LIST_HEAD(&session->auxtrace_index);
207 	machines__init(&session->machines);
208 	ordered_events__init(&session->ordered_events,
209 			     ordered_events__deliver_event, NULL);
210 
211 	perf_env__init(&session->header.env);
212 	if (data) {
213 		ret = perf_data__open(data);
214 		if (ret < 0)
215 			goto out_delete;
216 
217 		session->data = data;
218 
219 		if (perf_data__is_read(data)) {
220 			ret = perf_session__open(session, repipe_fd);
221 			if (ret < 0)
222 				goto out_delete;
223 
224 			/*
225 			 * set session attributes that are present in perf.data
226 			 * but not in pipe-mode.
227 			 */
228 			if (!data->is_pipe) {
229 				perf_session__set_id_hdr_size(session);
230 				perf_session__set_comm_exec(session);
231 			}
232 
233 			evlist__init_trace_event_sample_raw(session->evlist);
234 
235 			/* Open the directory data. */
236 			if (data->is_dir) {
237 				ret = perf_data__open_dir(data);
238 				if (ret)
239 					goto out_delete;
240 			}
241 
242 			if (!symbol_conf.kallsyms_name &&
243 			    !symbol_conf.vmlinux_name)
244 				symbol_conf.kallsyms_name = perf_data__kallsyms_name(data);
245 		}
246 	} else  {
247 		session->machines.host.env = &perf_env;
248 	}
249 
250 	session->machines.host.single_address_space =
251 		perf_env__single_address_space(session->machines.host.env);
252 
253 	if (!data || perf_data__is_write(data)) {
254 		/*
255 		 * In O_RDONLY mode this will be performed when reading the
256 		 * kernel MMAP event, in perf_event__process_mmap().
257 		 */
258 		if (perf_session__create_kernel_maps(session) < 0)
259 			pr_warning("Cannot read kernel map\n");
260 	}
261 
262 	/*
263 	 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
264 	 * processed, so evlist__sample_id_all is not meaningful here.
265 	 */
266 	if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps &&
267 	    tool->ordered_events && !evlist__sample_id_all(session->evlist)) {
268 		dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
269 		tool->ordered_events = false;
270 	}
271 
272 	return session;
273 
274  out_delete:
275 	perf_session__delete(session);
276  out:
277 	return ERR_PTR(ret);
278 }
279 
perf_session__delete_threads(struct perf_session * session)280 static void perf_session__delete_threads(struct perf_session *session)
281 {
282 	machine__delete_threads(&session->machines.host);
283 }
284 
perf_decomp__release_events(struct decomp * next)285 static void perf_decomp__release_events(struct decomp *next)
286 {
287 	struct decomp *decomp;
288 	size_t mmap_len;
289 
290 	do {
291 		decomp = next;
292 		if (decomp == NULL)
293 			break;
294 		next = decomp->next;
295 		mmap_len = decomp->mmap_len;
296 		munmap(decomp, mmap_len);
297 	} while (1);
298 }
299 
perf_session__delete(struct perf_session * session)300 void perf_session__delete(struct perf_session *session)
301 {
302 	if (session == NULL)
303 		return;
304 	auxtrace__free(session);
305 	auxtrace_index__free(&session->auxtrace_index);
306 	perf_session__destroy_kernel_maps(session);
307 	perf_session__delete_threads(session);
308 	perf_decomp__release_events(session->decomp_data.decomp);
309 	perf_env__exit(&session->header.env);
310 	machines__exit(&session->machines);
311 	if (session->data) {
312 		if (perf_data__is_read(session->data))
313 			evlist__delete(session->evlist);
314 		perf_data__close(session->data);
315 	}
316 	trace_event__cleanup(&session->tevent);
317 	free(session);
318 }
319 
process_event_synth_tracing_data_stub(struct perf_session * session __maybe_unused,union perf_event * event __maybe_unused)320 static int process_event_synth_tracing_data_stub(struct perf_session *session
321 						 __maybe_unused,
322 						 union perf_event *event
323 						 __maybe_unused)
324 {
325 	dump_printf(": unhandled!\n");
326 	return 0;
327 }
328 
process_event_synth_attr_stub(struct perf_tool * tool __maybe_unused,union perf_event * event __maybe_unused,struct evlist ** pevlist __maybe_unused)329 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
330 					 union perf_event *event __maybe_unused,
331 					 struct evlist **pevlist
332 					 __maybe_unused)
333 {
334 	dump_printf(": unhandled!\n");
335 	return 0;
336 }
337 
process_event_synth_event_update_stub(struct perf_tool * tool __maybe_unused,union perf_event * event __maybe_unused,struct evlist ** pevlist __maybe_unused)338 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
339 						 union perf_event *event __maybe_unused,
340 						 struct evlist **pevlist
341 						 __maybe_unused)
342 {
343 	if (dump_trace)
344 		perf_event__fprintf_event_update(event, stdout);
345 
346 	dump_printf(": unhandled!\n");
347 	return 0;
348 }
349 
process_event_sample_stub(struct perf_tool * tool __maybe_unused,union perf_event * event __maybe_unused,struct perf_sample * sample __maybe_unused,struct evsel * evsel __maybe_unused,struct machine * machine __maybe_unused)350 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
351 				     union perf_event *event __maybe_unused,
352 				     struct perf_sample *sample __maybe_unused,
353 				     struct evsel *evsel __maybe_unused,
354 				     struct machine *machine __maybe_unused)
355 {
356 	dump_printf(": unhandled!\n");
357 	return 0;
358 }
359 
process_event_stub(struct perf_tool * tool __maybe_unused,union perf_event * event __maybe_unused,struct perf_sample * sample __maybe_unused,struct machine * machine __maybe_unused)360 static int process_event_stub(struct perf_tool *tool __maybe_unused,
361 			      union perf_event *event __maybe_unused,
362 			      struct perf_sample *sample __maybe_unused,
363 			      struct machine *machine __maybe_unused)
364 {
365 	dump_printf(": unhandled!\n");
366 	return 0;
367 }
368 
process_finished_round_stub(struct perf_tool * tool __maybe_unused,union perf_event * event __maybe_unused,struct ordered_events * oe __maybe_unused)369 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
370 				       union perf_event *event __maybe_unused,
371 				       struct ordered_events *oe __maybe_unused)
372 {
373 	dump_printf(": unhandled!\n");
374 	return 0;
375 }
376 
377 static int process_finished_round(struct perf_tool *tool,
378 				  union perf_event *event,
379 				  struct ordered_events *oe);
380 
skipn(int fd,off_t n)381 static int skipn(int fd, off_t n)
382 {
383 	char buf[4096];
384 	ssize_t ret;
385 
386 	while (n > 0) {
387 		ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
388 		if (ret <= 0)
389 			return ret;
390 		n -= ret;
391 	}
392 
393 	return 0;
394 }
395 
process_event_auxtrace_stub(struct perf_session * session __maybe_unused,union perf_event * event)396 static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused,
397 				       union perf_event *event)
398 {
399 	dump_printf(": unhandled!\n");
400 	if (perf_data__is_pipe(session->data))
401 		skipn(perf_data__fd(session->data), event->auxtrace.size);
402 	return event->auxtrace.size;
403 }
404 
process_event_op2_stub(struct perf_session * session __maybe_unused,union perf_event * event __maybe_unused)405 static int process_event_op2_stub(struct perf_session *session __maybe_unused,
406 				  union perf_event *event __maybe_unused)
407 {
408 	dump_printf(": unhandled!\n");
409 	return 0;
410 }
411 
412 
413 static
process_event_thread_map_stub(struct perf_session * session __maybe_unused,union perf_event * event __maybe_unused)414 int process_event_thread_map_stub(struct perf_session *session __maybe_unused,
415 				  union perf_event *event __maybe_unused)
416 {
417 	if (dump_trace)
418 		perf_event__fprintf_thread_map(event, stdout);
419 
420 	dump_printf(": unhandled!\n");
421 	return 0;
422 }
423 
424 static
process_event_cpu_map_stub(struct perf_session * session __maybe_unused,union perf_event * event __maybe_unused)425 int process_event_cpu_map_stub(struct perf_session *session __maybe_unused,
426 			       union perf_event *event __maybe_unused)
427 {
428 	if (dump_trace)
429 		perf_event__fprintf_cpu_map(event, stdout);
430 
431 	dump_printf(": unhandled!\n");
432 	return 0;
433 }
434 
435 static
process_event_stat_config_stub(struct perf_session * session __maybe_unused,union perf_event * event __maybe_unused)436 int process_event_stat_config_stub(struct perf_session *session __maybe_unused,
437 				   union perf_event *event __maybe_unused)
438 {
439 	if (dump_trace)
440 		perf_event__fprintf_stat_config(event, stdout);
441 
442 	dump_printf(": unhandled!\n");
443 	return 0;
444 }
445 
process_stat_stub(struct perf_session * perf_session __maybe_unused,union perf_event * event)446 static int process_stat_stub(struct perf_session *perf_session __maybe_unused,
447 			     union perf_event *event)
448 {
449 	if (dump_trace)
450 		perf_event__fprintf_stat(event, stdout);
451 
452 	dump_printf(": unhandled!\n");
453 	return 0;
454 }
455 
process_stat_round_stub(struct perf_session * perf_session __maybe_unused,union perf_event * event)456 static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused,
457 				   union perf_event *event)
458 {
459 	if (dump_trace)
460 		perf_event__fprintf_stat_round(event, stdout);
461 
462 	dump_printf(": unhandled!\n");
463 	return 0;
464 }
465 
process_event_time_conv_stub(struct perf_session * perf_session __maybe_unused,union perf_event * event)466 static int process_event_time_conv_stub(struct perf_session *perf_session __maybe_unused,
467 					union perf_event *event)
468 {
469 	if (dump_trace)
470 		perf_event__fprintf_time_conv(event, stdout);
471 
472 	dump_printf(": unhandled!\n");
473 	return 0;
474 }
475 
perf_session__process_compressed_event_stub(struct perf_session * session __maybe_unused,union perf_event * event __maybe_unused,u64 file_offset __maybe_unused,const char * file_path __maybe_unused)476 static int perf_session__process_compressed_event_stub(struct perf_session *session __maybe_unused,
477 						       union perf_event *event __maybe_unused,
478 						       u64 file_offset __maybe_unused,
479 						       const char *file_path __maybe_unused)
480 {
481        dump_printf(": unhandled!\n");
482        return 0;
483 }
484 
perf_tool__fill_defaults(struct perf_tool * tool)485 void perf_tool__fill_defaults(struct perf_tool *tool)
486 {
487 	if (tool->sample == NULL)
488 		tool->sample = process_event_sample_stub;
489 	if (tool->mmap == NULL)
490 		tool->mmap = process_event_stub;
491 	if (tool->mmap2 == NULL)
492 		tool->mmap2 = process_event_stub;
493 	if (tool->comm == NULL)
494 		tool->comm = process_event_stub;
495 	if (tool->namespaces == NULL)
496 		tool->namespaces = process_event_stub;
497 	if (tool->cgroup == NULL)
498 		tool->cgroup = process_event_stub;
499 	if (tool->fork == NULL)
500 		tool->fork = process_event_stub;
501 	if (tool->exit == NULL)
502 		tool->exit = process_event_stub;
503 	if (tool->lost == NULL)
504 		tool->lost = perf_event__process_lost;
505 	if (tool->lost_samples == NULL)
506 		tool->lost_samples = perf_event__process_lost_samples;
507 	if (tool->aux == NULL)
508 		tool->aux = perf_event__process_aux;
509 	if (tool->itrace_start == NULL)
510 		tool->itrace_start = perf_event__process_itrace_start;
511 	if (tool->context_switch == NULL)
512 		tool->context_switch = perf_event__process_switch;
513 	if (tool->ksymbol == NULL)
514 		tool->ksymbol = perf_event__process_ksymbol;
515 	if (tool->bpf == NULL)
516 		tool->bpf = perf_event__process_bpf;
517 	if (tool->text_poke == NULL)
518 		tool->text_poke = perf_event__process_text_poke;
519 	if (tool->aux_output_hw_id == NULL)
520 		tool->aux_output_hw_id = perf_event__process_aux_output_hw_id;
521 	if (tool->read == NULL)
522 		tool->read = process_event_sample_stub;
523 	if (tool->throttle == NULL)
524 		tool->throttle = process_event_stub;
525 	if (tool->unthrottle == NULL)
526 		tool->unthrottle = process_event_stub;
527 	if (tool->attr == NULL)
528 		tool->attr = process_event_synth_attr_stub;
529 	if (tool->event_update == NULL)
530 		tool->event_update = process_event_synth_event_update_stub;
531 	if (tool->tracing_data == NULL)
532 		tool->tracing_data = process_event_synth_tracing_data_stub;
533 	if (tool->build_id == NULL)
534 		tool->build_id = process_event_op2_stub;
535 	if (tool->finished_round == NULL) {
536 		if (tool->ordered_events)
537 			tool->finished_round = process_finished_round;
538 		else
539 			tool->finished_round = process_finished_round_stub;
540 	}
541 	if (tool->id_index == NULL)
542 		tool->id_index = process_event_op2_stub;
543 	if (tool->auxtrace_info == NULL)
544 		tool->auxtrace_info = process_event_op2_stub;
545 	if (tool->auxtrace == NULL)
546 		tool->auxtrace = process_event_auxtrace_stub;
547 	if (tool->auxtrace_error == NULL)
548 		tool->auxtrace_error = process_event_op2_stub;
549 	if (tool->thread_map == NULL)
550 		tool->thread_map = process_event_thread_map_stub;
551 	if (tool->cpu_map == NULL)
552 		tool->cpu_map = process_event_cpu_map_stub;
553 	if (tool->stat_config == NULL)
554 		tool->stat_config = process_event_stat_config_stub;
555 	if (tool->stat == NULL)
556 		tool->stat = process_stat_stub;
557 	if (tool->stat_round == NULL)
558 		tool->stat_round = process_stat_round_stub;
559 	if (tool->time_conv == NULL)
560 		tool->time_conv = process_event_time_conv_stub;
561 	if (tool->feature == NULL)
562 		tool->feature = process_event_op2_stub;
563 	if (tool->compressed == NULL)
564 		tool->compressed = perf_session__process_compressed_event;
565 }
566 
swap_sample_id_all(union perf_event * event,void * data)567 static void swap_sample_id_all(union perf_event *event, void *data)
568 {
569 	void *end = (void *) event + event->header.size;
570 	int size = end - data;
571 
572 	BUG_ON(size % sizeof(u64));
573 	mem_bswap_64(data, size);
574 }
575 
perf_event__all64_swap(union perf_event * event,bool sample_id_all __maybe_unused)576 static void perf_event__all64_swap(union perf_event *event,
577 				   bool sample_id_all __maybe_unused)
578 {
579 	struct perf_event_header *hdr = &event->header;
580 	mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
581 }
582 
perf_event__comm_swap(union perf_event * event,bool sample_id_all)583 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
584 {
585 	event->comm.pid = bswap_32(event->comm.pid);
586 	event->comm.tid = bswap_32(event->comm.tid);
587 
588 	if (sample_id_all) {
589 		void *data = &event->comm.comm;
590 
591 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
592 		swap_sample_id_all(event, data);
593 	}
594 }
595 
perf_event__mmap_swap(union perf_event * event,bool sample_id_all)596 static void perf_event__mmap_swap(union perf_event *event,
597 				  bool sample_id_all)
598 {
599 	event->mmap.pid	  = bswap_32(event->mmap.pid);
600 	event->mmap.tid	  = bswap_32(event->mmap.tid);
601 	event->mmap.start = bswap_64(event->mmap.start);
602 	event->mmap.len	  = bswap_64(event->mmap.len);
603 	event->mmap.pgoff = bswap_64(event->mmap.pgoff);
604 
605 	if (sample_id_all) {
606 		void *data = &event->mmap.filename;
607 
608 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
609 		swap_sample_id_all(event, data);
610 	}
611 }
612 
perf_event__mmap2_swap(union perf_event * event,bool sample_id_all)613 static void perf_event__mmap2_swap(union perf_event *event,
614 				  bool sample_id_all)
615 {
616 	event->mmap2.pid   = bswap_32(event->mmap2.pid);
617 	event->mmap2.tid   = bswap_32(event->mmap2.tid);
618 	event->mmap2.start = bswap_64(event->mmap2.start);
619 	event->mmap2.len   = bswap_64(event->mmap2.len);
620 	event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
621 
622 	if (!(event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID)) {
623 		event->mmap2.maj   = bswap_32(event->mmap2.maj);
624 		event->mmap2.min   = bswap_32(event->mmap2.min);
625 		event->mmap2.ino   = bswap_64(event->mmap2.ino);
626 		event->mmap2.ino_generation = bswap_64(event->mmap2.ino_generation);
627 	}
628 
629 	if (sample_id_all) {
630 		void *data = &event->mmap2.filename;
631 
632 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
633 		swap_sample_id_all(event, data);
634 	}
635 }
perf_event__task_swap(union perf_event * event,bool sample_id_all)636 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
637 {
638 	event->fork.pid	 = bswap_32(event->fork.pid);
639 	event->fork.tid	 = bswap_32(event->fork.tid);
640 	event->fork.ppid = bswap_32(event->fork.ppid);
641 	event->fork.ptid = bswap_32(event->fork.ptid);
642 	event->fork.time = bswap_64(event->fork.time);
643 
644 	if (sample_id_all)
645 		swap_sample_id_all(event, &event->fork + 1);
646 }
647 
perf_event__read_swap(union perf_event * event,bool sample_id_all)648 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
649 {
650 	event->read.pid		 = bswap_32(event->read.pid);
651 	event->read.tid		 = bswap_32(event->read.tid);
652 	event->read.value	 = bswap_64(event->read.value);
653 	event->read.time_enabled = bswap_64(event->read.time_enabled);
654 	event->read.time_running = bswap_64(event->read.time_running);
655 	event->read.id		 = bswap_64(event->read.id);
656 
657 	if (sample_id_all)
658 		swap_sample_id_all(event, &event->read + 1);
659 }
660 
perf_event__aux_swap(union perf_event * event,bool sample_id_all)661 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
662 {
663 	event->aux.aux_offset = bswap_64(event->aux.aux_offset);
664 	event->aux.aux_size   = bswap_64(event->aux.aux_size);
665 	event->aux.flags      = bswap_64(event->aux.flags);
666 
667 	if (sample_id_all)
668 		swap_sample_id_all(event, &event->aux + 1);
669 }
670 
perf_event__itrace_start_swap(union perf_event * event,bool sample_id_all)671 static void perf_event__itrace_start_swap(union perf_event *event,
672 					  bool sample_id_all)
673 {
674 	event->itrace_start.pid	 = bswap_32(event->itrace_start.pid);
675 	event->itrace_start.tid	 = bswap_32(event->itrace_start.tid);
676 
677 	if (sample_id_all)
678 		swap_sample_id_all(event, &event->itrace_start + 1);
679 }
680 
perf_event__switch_swap(union perf_event * event,bool sample_id_all)681 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
682 {
683 	if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
684 		event->context_switch.next_prev_pid =
685 				bswap_32(event->context_switch.next_prev_pid);
686 		event->context_switch.next_prev_tid =
687 				bswap_32(event->context_switch.next_prev_tid);
688 	}
689 
690 	if (sample_id_all)
691 		swap_sample_id_all(event, &event->context_switch + 1);
692 }
693 
perf_event__text_poke_swap(union perf_event * event,bool sample_id_all)694 static void perf_event__text_poke_swap(union perf_event *event, bool sample_id_all)
695 {
696 	event->text_poke.addr    = bswap_64(event->text_poke.addr);
697 	event->text_poke.old_len = bswap_16(event->text_poke.old_len);
698 	event->text_poke.new_len = bswap_16(event->text_poke.new_len);
699 
700 	if (sample_id_all) {
701 		size_t len = sizeof(event->text_poke.old_len) +
702 			     sizeof(event->text_poke.new_len) +
703 			     event->text_poke.old_len +
704 			     event->text_poke.new_len;
705 		void *data = &event->text_poke.old_len;
706 
707 		data += PERF_ALIGN(len, sizeof(u64));
708 		swap_sample_id_all(event, data);
709 	}
710 }
711 
perf_event__throttle_swap(union perf_event * event,bool sample_id_all)712 static void perf_event__throttle_swap(union perf_event *event,
713 				      bool sample_id_all)
714 {
715 	event->throttle.time	  = bswap_64(event->throttle.time);
716 	event->throttle.id	  = bswap_64(event->throttle.id);
717 	event->throttle.stream_id = bswap_64(event->throttle.stream_id);
718 
719 	if (sample_id_all)
720 		swap_sample_id_all(event, &event->throttle + 1);
721 }
722 
perf_event__namespaces_swap(union perf_event * event,bool sample_id_all)723 static void perf_event__namespaces_swap(union perf_event *event,
724 					bool sample_id_all)
725 {
726 	u64 i;
727 
728 	event->namespaces.pid		= bswap_32(event->namespaces.pid);
729 	event->namespaces.tid		= bswap_32(event->namespaces.tid);
730 	event->namespaces.nr_namespaces	= bswap_64(event->namespaces.nr_namespaces);
731 
732 	for (i = 0; i < event->namespaces.nr_namespaces; i++) {
733 		struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
734 
735 		ns->dev = bswap_64(ns->dev);
736 		ns->ino = bswap_64(ns->ino);
737 	}
738 
739 	if (sample_id_all)
740 		swap_sample_id_all(event, &event->namespaces.link_info[i]);
741 }
742 
perf_event__cgroup_swap(union perf_event * event,bool sample_id_all)743 static void perf_event__cgroup_swap(union perf_event *event, bool sample_id_all)
744 {
745 	event->cgroup.id = bswap_64(event->cgroup.id);
746 
747 	if (sample_id_all) {
748 		void *data = &event->cgroup.path;
749 
750 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
751 		swap_sample_id_all(event, data);
752 	}
753 }
754 
revbyte(u8 b)755 static u8 revbyte(u8 b)
756 {
757 	int rev = (b >> 4) | ((b & 0xf) << 4);
758 	rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
759 	rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
760 	return (u8) rev;
761 }
762 
763 /*
764  * XXX this is hack in attempt to carry flags bitfield
765  * through endian village. ABI says:
766  *
767  * Bit-fields are allocated from right to left (least to most significant)
768  * on little-endian implementations and from left to right (most to least
769  * significant) on big-endian implementations.
770  *
771  * The above seems to be byte specific, so we need to reverse each
772  * byte of the bitfield. 'Internet' also says this might be implementation
773  * specific and we probably need proper fix and carry perf_event_attr
774  * bitfield flags in separate data file FEAT_ section. Thought this seems
775  * to work for now.
776  */
swap_bitfield(u8 * p,unsigned len)777 static void swap_bitfield(u8 *p, unsigned len)
778 {
779 	unsigned i;
780 
781 	for (i = 0; i < len; i++) {
782 		*p = revbyte(*p);
783 		p++;
784 	}
785 }
786 
787 /* exported for swapping attributes in file header */
perf_event__attr_swap(struct perf_event_attr * attr)788 void perf_event__attr_swap(struct perf_event_attr *attr)
789 {
790 	attr->type		= bswap_32(attr->type);
791 	attr->size		= bswap_32(attr->size);
792 
793 #define bswap_safe(f, n) 					\
794 	(attr->size > (offsetof(struct perf_event_attr, f) + 	\
795 		       sizeof(attr->f) * (n)))
796 #define bswap_field(f, sz) 			\
797 do { 						\
798 	if (bswap_safe(f, 0))			\
799 		attr->f = bswap_##sz(attr->f);	\
800 } while(0)
801 #define bswap_field_16(f) bswap_field(f, 16)
802 #define bswap_field_32(f) bswap_field(f, 32)
803 #define bswap_field_64(f) bswap_field(f, 64)
804 
805 	bswap_field_64(config);
806 	bswap_field_64(sample_period);
807 	bswap_field_64(sample_type);
808 	bswap_field_64(read_format);
809 	bswap_field_32(wakeup_events);
810 	bswap_field_32(bp_type);
811 	bswap_field_64(bp_addr);
812 	bswap_field_64(bp_len);
813 	bswap_field_64(branch_sample_type);
814 	bswap_field_64(sample_regs_user);
815 	bswap_field_32(sample_stack_user);
816 	bswap_field_32(aux_watermark);
817 	bswap_field_16(sample_max_stack);
818 	bswap_field_32(aux_sample_size);
819 
820 	/*
821 	 * After read_format are bitfields. Check read_format because
822 	 * we are unable to use offsetof on bitfield.
823 	 */
824 	if (bswap_safe(read_format, 1))
825 		swap_bitfield((u8 *) (&attr->read_format + 1),
826 			      sizeof(u64));
827 #undef bswap_field_64
828 #undef bswap_field_32
829 #undef bswap_field
830 #undef bswap_safe
831 }
832 
perf_event__hdr_attr_swap(union perf_event * event,bool sample_id_all __maybe_unused)833 static void perf_event__hdr_attr_swap(union perf_event *event,
834 				      bool sample_id_all __maybe_unused)
835 {
836 	size_t size;
837 
838 	perf_event__attr_swap(&event->attr.attr);
839 
840 	size = event->header.size;
841 	size -= (void *)&event->attr.id - (void *)event;
842 	mem_bswap_64(event->attr.id, size);
843 }
844 
perf_event__event_update_swap(union perf_event * event,bool sample_id_all __maybe_unused)845 static void perf_event__event_update_swap(union perf_event *event,
846 					  bool sample_id_all __maybe_unused)
847 {
848 	event->event_update.type = bswap_64(event->event_update.type);
849 	event->event_update.id   = bswap_64(event->event_update.id);
850 }
851 
perf_event__event_type_swap(union perf_event * event,bool sample_id_all __maybe_unused)852 static void perf_event__event_type_swap(union perf_event *event,
853 					bool sample_id_all __maybe_unused)
854 {
855 	event->event_type.event_type.event_id =
856 		bswap_64(event->event_type.event_type.event_id);
857 }
858 
perf_event__tracing_data_swap(union perf_event * event,bool sample_id_all __maybe_unused)859 static void perf_event__tracing_data_swap(union perf_event *event,
860 					  bool sample_id_all __maybe_unused)
861 {
862 	event->tracing_data.size = bswap_32(event->tracing_data.size);
863 }
864 
perf_event__auxtrace_info_swap(union perf_event * event,bool sample_id_all __maybe_unused)865 static void perf_event__auxtrace_info_swap(union perf_event *event,
866 					   bool sample_id_all __maybe_unused)
867 {
868 	size_t size;
869 
870 	event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
871 
872 	size = event->header.size;
873 	size -= (void *)&event->auxtrace_info.priv - (void *)event;
874 	mem_bswap_64(event->auxtrace_info.priv, size);
875 }
876 
perf_event__auxtrace_swap(union perf_event * event,bool sample_id_all __maybe_unused)877 static void perf_event__auxtrace_swap(union perf_event *event,
878 				      bool sample_id_all __maybe_unused)
879 {
880 	event->auxtrace.size      = bswap_64(event->auxtrace.size);
881 	event->auxtrace.offset    = bswap_64(event->auxtrace.offset);
882 	event->auxtrace.reference = bswap_64(event->auxtrace.reference);
883 	event->auxtrace.idx       = bswap_32(event->auxtrace.idx);
884 	event->auxtrace.tid       = bswap_32(event->auxtrace.tid);
885 	event->auxtrace.cpu       = bswap_32(event->auxtrace.cpu);
886 }
887 
perf_event__auxtrace_error_swap(union perf_event * event,bool sample_id_all __maybe_unused)888 static void perf_event__auxtrace_error_swap(union perf_event *event,
889 					    bool sample_id_all __maybe_unused)
890 {
891 	event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
892 	event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
893 	event->auxtrace_error.cpu  = bswap_32(event->auxtrace_error.cpu);
894 	event->auxtrace_error.pid  = bswap_32(event->auxtrace_error.pid);
895 	event->auxtrace_error.tid  = bswap_32(event->auxtrace_error.tid);
896 	event->auxtrace_error.fmt  = bswap_32(event->auxtrace_error.fmt);
897 	event->auxtrace_error.ip   = bswap_64(event->auxtrace_error.ip);
898 	if (event->auxtrace_error.fmt)
899 		event->auxtrace_error.time = bswap_64(event->auxtrace_error.time);
900 }
901 
perf_event__thread_map_swap(union perf_event * event,bool sample_id_all __maybe_unused)902 static void perf_event__thread_map_swap(union perf_event *event,
903 					bool sample_id_all __maybe_unused)
904 {
905 	unsigned i;
906 
907 	event->thread_map.nr = bswap_64(event->thread_map.nr);
908 
909 	for (i = 0; i < event->thread_map.nr; i++)
910 		event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
911 }
912 
perf_event__cpu_map_swap(union perf_event * event,bool sample_id_all __maybe_unused)913 static void perf_event__cpu_map_swap(union perf_event *event,
914 				     bool sample_id_all __maybe_unused)
915 {
916 	struct perf_record_cpu_map_data *data = &event->cpu_map.data;
917 	struct cpu_map_entries *cpus;
918 	struct perf_record_record_cpu_map *mask;
919 	unsigned i;
920 
921 	data->type = bswap_16(data->type);
922 
923 	switch (data->type) {
924 	case PERF_CPU_MAP__CPUS:
925 		cpus = (struct cpu_map_entries *)data->data;
926 
927 		cpus->nr = bswap_16(cpus->nr);
928 
929 		for (i = 0; i < cpus->nr; i++)
930 			cpus->cpu[i] = bswap_16(cpus->cpu[i]);
931 		break;
932 	case PERF_CPU_MAP__MASK:
933 		mask = (struct perf_record_record_cpu_map *)data->data;
934 
935 		mask->nr = bswap_16(mask->nr);
936 		mask->long_size = bswap_16(mask->long_size);
937 
938 		switch (mask->long_size) {
939 		case 4: mem_bswap_32(&mask->mask, mask->nr); break;
940 		case 8: mem_bswap_64(&mask->mask, mask->nr); break;
941 		default:
942 			pr_err("cpu_map swap: unsupported long size\n");
943 		}
944 	default:
945 		break;
946 	}
947 }
948 
perf_event__stat_config_swap(union perf_event * event,bool sample_id_all __maybe_unused)949 static void perf_event__stat_config_swap(union perf_event *event,
950 					 bool sample_id_all __maybe_unused)
951 {
952 	u64 size;
953 
954 	size  = bswap_64(event->stat_config.nr) * sizeof(event->stat_config.data[0]);
955 	size += 1; /* nr item itself */
956 	mem_bswap_64(&event->stat_config.nr, size);
957 }
958 
perf_event__stat_swap(union perf_event * event,bool sample_id_all __maybe_unused)959 static void perf_event__stat_swap(union perf_event *event,
960 				  bool sample_id_all __maybe_unused)
961 {
962 	event->stat.id     = bswap_64(event->stat.id);
963 	event->stat.thread = bswap_32(event->stat.thread);
964 	event->stat.cpu    = bswap_32(event->stat.cpu);
965 	event->stat.val    = bswap_64(event->stat.val);
966 	event->stat.ena    = bswap_64(event->stat.ena);
967 	event->stat.run    = bswap_64(event->stat.run);
968 }
969 
perf_event__stat_round_swap(union perf_event * event,bool sample_id_all __maybe_unused)970 static void perf_event__stat_round_swap(union perf_event *event,
971 					bool sample_id_all __maybe_unused)
972 {
973 	event->stat_round.type = bswap_64(event->stat_round.type);
974 	event->stat_round.time = bswap_64(event->stat_round.time);
975 }
976 
perf_event__time_conv_swap(union perf_event * event,bool sample_id_all __maybe_unused)977 static void perf_event__time_conv_swap(union perf_event *event,
978 				       bool sample_id_all __maybe_unused)
979 {
980 	event->time_conv.time_shift = bswap_64(event->time_conv.time_shift);
981 	event->time_conv.time_mult  = bswap_64(event->time_conv.time_mult);
982 	event->time_conv.time_zero  = bswap_64(event->time_conv.time_zero);
983 
984 	if (event_contains(event->time_conv, time_cycles)) {
985 		event->time_conv.time_cycles = bswap_64(event->time_conv.time_cycles);
986 		event->time_conv.time_mask = bswap_64(event->time_conv.time_mask);
987 	}
988 }
989 
990 typedef void (*perf_event__swap_op)(union perf_event *event,
991 				    bool sample_id_all);
992 
993 static perf_event__swap_op perf_event__swap_ops[] = {
994 	[PERF_RECORD_MMAP]		  = perf_event__mmap_swap,
995 	[PERF_RECORD_MMAP2]		  = perf_event__mmap2_swap,
996 	[PERF_RECORD_COMM]		  = perf_event__comm_swap,
997 	[PERF_RECORD_FORK]		  = perf_event__task_swap,
998 	[PERF_RECORD_EXIT]		  = perf_event__task_swap,
999 	[PERF_RECORD_LOST]		  = perf_event__all64_swap,
1000 	[PERF_RECORD_READ]		  = perf_event__read_swap,
1001 	[PERF_RECORD_THROTTLE]		  = perf_event__throttle_swap,
1002 	[PERF_RECORD_UNTHROTTLE]	  = perf_event__throttle_swap,
1003 	[PERF_RECORD_SAMPLE]		  = perf_event__all64_swap,
1004 	[PERF_RECORD_AUX]		  = perf_event__aux_swap,
1005 	[PERF_RECORD_ITRACE_START]	  = perf_event__itrace_start_swap,
1006 	[PERF_RECORD_LOST_SAMPLES]	  = perf_event__all64_swap,
1007 	[PERF_RECORD_SWITCH]		  = perf_event__switch_swap,
1008 	[PERF_RECORD_SWITCH_CPU_WIDE]	  = perf_event__switch_swap,
1009 	[PERF_RECORD_NAMESPACES]	  = perf_event__namespaces_swap,
1010 	[PERF_RECORD_CGROUP]		  = perf_event__cgroup_swap,
1011 	[PERF_RECORD_TEXT_POKE]		  = perf_event__text_poke_swap,
1012 	[PERF_RECORD_AUX_OUTPUT_HW_ID]	  = perf_event__all64_swap,
1013 	[PERF_RECORD_HEADER_ATTR]	  = perf_event__hdr_attr_swap,
1014 	[PERF_RECORD_HEADER_EVENT_TYPE]	  = perf_event__event_type_swap,
1015 	[PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
1016 	[PERF_RECORD_HEADER_BUILD_ID]	  = NULL,
1017 	[PERF_RECORD_ID_INDEX]		  = perf_event__all64_swap,
1018 	[PERF_RECORD_AUXTRACE_INFO]	  = perf_event__auxtrace_info_swap,
1019 	[PERF_RECORD_AUXTRACE]		  = perf_event__auxtrace_swap,
1020 	[PERF_RECORD_AUXTRACE_ERROR]	  = perf_event__auxtrace_error_swap,
1021 	[PERF_RECORD_THREAD_MAP]	  = perf_event__thread_map_swap,
1022 	[PERF_RECORD_CPU_MAP]		  = perf_event__cpu_map_swap,
1023 	[PERF_RECORD_STAT_CONFIG]	  = perf_event__stat_config_swap,
1024 	[PERF_RECORD_STAT]		  = perf_event__stat_swap,
1025 	[PERF_RECORD_STAT_ROUND]	  = perf_event__stat_round_swap,
1026 	[PERF_RECORD_EVENT_UPDATE]	  = perf_event__event_update_swap,
1027 	[PERF_RECORD_TIME_CONV]		  = perf_event__time_conv_swap,
1028 	[PERF_RECORD_HEADER_MAX]	  = NULL,
1029 };
1030 
1031 /*
1032  * When perf record finishes a pass on every buffers, it records this pseudo
1033  * event.
1034  * We record the max timestamp t found in the pass n.
1035  * Assuming these timestamps are monotonic across cpus, we know that if
1036  * a buffer still has events with timestamps below t, they will be all
1037  * available and then read in the pass n + 1.
1038  * Hence when we start to read the pass n + 2, we can safely flush every
1039  * events with timestamps below t.
1040  *
1041  *    ============ PASS n =================
1042  *       CPU 0         |   CPU 1
1043  *                     |
1044  *    cnt1 timestamps  |   cnt2 timestamps
1045  *          1          |         2
1046  *          2          |         3
1047  *          -          |         4  <--- max recorded
1048  *
1049  *    ============ PASS n + 1 ==============
1050  *       CPU 0         |   CPU 1
1051  *                     |
1052  *    cnt1 timestamps  |   cnt2 timestamps
1053  *          3          |         5
1054  *          4          |         6
1055  *          5          |         7 <---- max recorded
1056  *
1057  *      Flush every events below timestamp 4
1058  *
1059  *    ============ PASS n + 2 ==============
1060  *       CPU 0         |   CPU 1
1061  *                     |
1062  *    cnt1 timestamps  |   cnt2 timestamps
1063  *          6          |         8
1064  *          7          |         9
1065  *          -          |         10
1066  *
1067  *      Flush every events below timestamp 7
1068  *      etc...
1069  */
process_finished_round(struct perf_tool * tool __maybe_unused,union perf_event * event __maybe_unused,struct ordered_events * oe)1070 static int process_finished_round(struct perf_tool *tool __maybe_unused,
1071 				  union perf_event *event __maybe_unused,
1072 				  struct ordered_events *oe)
1073 {
1074 	if (dump_trace)
1075 		fprintf(stdout, "\n");
1076 	return ordered_events__flush(oe, OE_FLUSH__ROUND);
1077 }
1078 
perf_session__queue_event(struct perf_session * s,union perf_event * event,u64 timestamp,u64 file_offset,const char * file_path)1079 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
1080 			      u64 timestamp, u64 file_offset, const char *file_path)
1081 {
1082 	return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset, file_path);
1083 }
1084 
callchain__lbr_callstack_printf(struct perf_sample * sample)1085 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
1086 {
1087 	struct ip_callchain *callchain = sample->callchain;
1088 	struct branch_stack *lbr_stack = sample->branch_stack;
1089 	struct branch_entry *entries = perf_sample__branch_entries(sample);
1090 	u64 kernel_callchain_nr = callchain->nr;
1091 	unsigned int i;
1092 
1093 	for (i = 0; i < kernel_callchain_nr; i++) {
1094 		if (callchain->ips[i] == PERF_CONTEXT_USER)
1095 			break;
1096 	}
1097 
1098 	if ((i != kernel_callchain_nr) && lbr_stack->nr) {
1099 		u64 total_nr;
1100 		/*
1101 		 * LBR callstack can only get user call chain,
1102 		 * i is kernel call chain number,
1103 		 * 1 is PERF_CONTEXT_USER.
1104 		 *
1105 		 * The user call chain is stored in LBR registers.
1106 		 * LBR are pair registers. The caller is stored
1107 		 * in "from" register, while the callee is stored
1108 		 * in "to" register.
1109 		 * For example, there is a call stack
1110 		 * "A"->"B"->"C"->"D".
1111 		 * The LBR registers will be recorded like
1112 		 * "C"->"D", "B"->"C", "A"->"B".
1113 		 * So only the first "to" register and all "from"
1114 		 * registers are needed to construct the whole stack.
1115 		 */
1116 		total_nr = i + 1 + lbr_stack->nr + 1;
1117 		kernel_callchain_nr = i + 1;
1118 
1119 		printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
1120 
1121 		for (i = 0; i < kernel_callchain_nr; i++)
1122 			printf("..... %2d: %016" PRIx64 "\n",
1123 			       i, callchain->ips[i]);
1124 
1125 		printf("..... %2d: %016" PRIx64 "\n",
1126 		       (int)(kernel_callchain_nr), entries[0].to);
1127 		for (i = 0; i < lbr_stack->nr; i++)
1128 			printf("..... %2d: %016" PRIx64 "\n",
1129 			       (int)(i + kernel_callchain_nr + 1), entries[i].from);
1130 	}
1131 }
1132 
callchain__printf(struct evsel * evsel,struct perf_sample * sample)1133 static void callchain__printf(struct evsel *evsel,
1134 			      struct perf_sample *sample)
1135 {
1136 	unsigned int i;
1137 	struct ip_callchain *callchain = sample->callchain;
1138 
1139 	if (evsel__has_branch_callstack(evsel))
1140 		callchain__lbr_callstack_printf(sample);
1141 
1142 	printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
1143 
1144 	for (i = 0; i < callchain->nr; i++)
1145 		printf("..... %2d: %016" PRIx64 "\n",
1146 		       i, callchain->ips[i]);
1147 }
1148 
branch_stack__printf(struct perf_sample * sample,bool callstack)1149 static void branch_stack__printf(struct perf_sample *sample, bool callstack)
1150 {
1151 	struct branch_entry *entries = perf_sample__branch_entries(sample);
1152 	uint64_t i;
1153 
1154 	if (!callstack) {
1155 		printf("%s: nr:%" PRIu64 "\n", "... branch stack", sample->branch_stack->nr);
1156 	} else {
1157 		/* the reason of adding 1 to nr is because after expanding
1158 		 * branch stack it generates nr + 1 callstack records. e.g.,
1159 		 *         B()->C()
1160 		 *         A()->B()
1161 		 * the final callstack should be:
1162 		 *         C()
1163 		 *         B()
1164 		 *         A()
1165 		 */
1166 		printf("%s: nr:%" PRIu64 "\n", "... branch callstack", sample->branch_stack->nr+1);
1167 	}
1168 
1169 	for (i = 0; i < sample->branch_stack->nr; i++) {
1170 		struct branch_entry *e = &entries[i];
1171 
1172 		if (!callstack) {
1173 			printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x %s\n",
1174 				i, e->from, e->to,
1175 				(unsigned short)e->flags.cycles,
1176 				e->flags.mispred ? "M" : " ",
1177 				e->flags.predicted ? "P" : " ",
1178 				e->flags.abort ? "A" : " ",
1179 				e->flags.in_tx ? "T" : " ",
1180 				(unsigned)e->flags.reserved,
1181 				e->flags.type ? branch_type_name(e->flags.type) : "");
1182 		} else {
1183 			if (i == 0) {
1184 				printf("..... %2"PRIu64": %016" PRIx64 "\n"
1185 				       "..... %2"PRIu64": %016" PRIx64 "\n",
1186 						i, e->to, i+1, e->from);
1187 			} else {
1188 				printf("..... %2"PRIu64": %016" PRIx64 "\n", i+1, e->from);
1189 			}
1190 		}
1191 	}
1192 }
1193 
regs_dump__printf(u64 mask,u64 * regs,const char * arch)1194 static void regs_dump__printf(u64 mask, u64 *regs, const char *arch)
1195 {
1196 	unsigned rid, i = 0;
1197 
1198 	for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
1199 		u64 val = regs[i++];
1200 
1201 		printf(".... %-5s 0x%016" PRIx64 "\n",
1202 		       perf_reg_name(rid, arch), val);
1203 	}
1204 }
1205 
1206 static const char *regs_abi[] = {
1207 	[PERF_SAMPLE_REGS_ABI_NONE] = "none",
1208 	[PERF_SAMPLE_REGS_ABI_32] = "32-bit",
1209 	[PERF_SAMPLE_REGS_ABI_64] = "64-bit",
1210 };
1211 
regs_dump_abi(struct regs_dump * d)1212 static inline const char *regs_dump_abi(struct regs_dump *d)
1213 {
1214 	if (d->abi > PERF_SAMPLE_REGS_ABI_64)
1215 		return "unknown";
1216 
1217 	return regs_abi[d->abi];
1218 }
1219 
regs__printf(const char * type,struct regs_dump * regs,const char * arch)1220 static void regs__printf(const char *type, struct regs_dump *regs, const char *arch)
1221 {
1222 	u64 mask = regs->mask;
1223 
1224 	printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
1225 	       type,
1226 	       mask,
1227 	       regs_dump_abi(regs));
1228 
1229 	regs_dump__printf(mask, regs->regs, arch);
1230 }
1231 
regs_user__printf(struct perf_sample * sample,const char * arch)1232 static void regs_user__printf(struct perf_sample *sample, const char *arch)
1233 {
1234 	struct regs_dump *user_regs = &sample->user_regs;
1235 
1236 	if (user_regs->regs)
1237 		regs__printf("user", user_regs, arch);
1238 }
1239 
regs_intr__printf(struct perf_sample * sample,const char * arch)1240 static void regs_intr__printf(struct perf_sample *sample, const char *arch)
1241 {
1242 	struct regs_dump *intr_regs = &sample->intr_regs;
1243 
1244 	if (intr_regs->regs)
1245 		regs__printf("intr", intr_regs, arch);
1246 }
1247 
stack_user__printf(struct stack_dump * dump)1248 static void stack_user__printf(struct stack_dump *dump)
1249 {
1250 	printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1251 	       dump->size, dump->offset);
1252 }
1253 
evlist__print_tstamp(struct evlist * evlist,union perf_event * event,struct perf_sample * sample)1254 static void evlist__print_tstamp(struct evlist *evlist, union perf_event *event, struct perf_sample *sample)
1255 {
1256 	u64 sample_type = __evlist__combined_sample_type(evlist);
1257 
1258 	if (event->header.type != PERF_RECORD_SAMPLE &&
1259 	    !evlist__sample_id_all(evlist)) {
1260 		fputs("-1 -1 ", stdout);
1261 		return;
1262 	}
1263 
1264 	if ((sample_type & PERF_SAMPLE_CPU))
1265 		printf("%u ", sample->cpu);
1266 
1267 	if (sample_type & PERF_SAMPLE_TIME)
1268 		printf("%" PRIu64 " ", sample->time);
1269 }
1270 
sample_read__printf(struct perf_sample * sample,u64 read_format)1271 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1272 {
1273 	printf("... sample_read:\n");
1274 
1275 	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1276 		printf("...... time enabled %016" PRIx64 "\n",
1277 		       sample->read.time_enabled);
1278 
1279 	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1280 		printf("...... time running %016" PRIx64 "\n",
1281 		       sample->read.time_running);
1282 
1283 	if (read_format & PERF_FORMAT_GROUP) {
1284 		u64 i;
1285 
1286 		printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1287 
1288 		for (i = 0; i < sample->read.group.nr; i++) {
1289 			struct sample_read_value *value;
1290 
1291 			value = &sample->read.group.values[i];
1292 			printf("..... id %016" PRIx64
1293 			       ", value %016" PRIx64 "\n",
1294 			       value->id, value->value);
1295 		}
1296 	} else
1297 		printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
1298 			sample->read.one.id, sample->read.one.value);
1299 }
1300 
dump_event(struct evlist * evlist,union perf_event * event,u64 file_offset,struct perf_sample * sample,const char * file_path)1301 static void dump_event(struct evlist *evlist, union perf_event *event,
1302 		       u64 file_offset, struct perf_sample *sample,
1303 		       const char *file_path)
1304 {
1305 	if (!dump_trace)
1306 		return;
1307 
1308 	printf("\n%#" PRIx64 "@%s [%#x]: event: %d\n",
1309 	       file_offset, file_path, event->header.size, event->header.type);
1310 
1311 	trace_event(event);
1312 	if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw)
1313 		evlist->trace_event_sample_raw(evlist, event, sample);
1314 
1315 	if (sample)
1316 		evlist__print_tstamp(evlist, event, sample);
1317 
1318 	printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1319 	       event->header.size, perf_event__name(event->header.type));
1320 }
1321 
get_page_size_name(u64 size,char * str)1322 char *get_page_size_name(u64 size, char *str)
1323 {
1324 	if (!size || !unit_number__scnprintf(str, PAGE_SIZE_NAME_LEN, size))
1325 		snprintf(str, PAGE_SIZE_NAME_LEN, "%s", "N/A");
1326 
1327 	return str;
1328 }
1329 
dump_sample(struct evsel * evsel,union perf_event * event,struct perf_sample * sample,const char * arch)1330 static void dump_sample(struct evsel *evsel, union perf_event *event,
1331 			struct perf_sample *sample, const char *arch)
1332 {
1333 	u64 sample_type;
1334 	char str[PAGE_SIZE_NAME_LEN];
1335 
1336 	if (!dump_trace)
1337 		return;
1338 
1339 	printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1340 	       event->header.misc, sample->pid, sample->tid, sample->ip,
1341 	       sample->period, sample->addr);
1342 
1343 	sample_type = evsel->core.attr.sample_type;
1344 
1345 	if (evsel__has_callchain(evsel))
1346 		callchain__printf(evsel, sample);
1347 
1348 	if (evsel__has_br_stack(evsel))
1349 		branch_stack__printf(sample, evsel__has_branch_callstack(evsel));
1350 
1351 	if (sample_type & PERF_SAMPLE_REGS_USER)
1352 		regs_user__printf(sample, arch);
1353 
1354 	if (sample_type & PERF_SAMPLE_REGS_INTR)
1355 		regs_intr__printf(sample, arch);
1356 
1357 	if (sample_type & PERF_SAMPLE_STACK_USER)
1358 		stack_user__printf(&sample->user_stack);
1359 
1360 	if (sample_type & PERF_SAMPLE_WEIGHT_TYPE) {
1361 		printf("... weight: %" PRIu64 "", sample->weight);
1362 			if (sample_type & PERF_SAMPLE_WEIGHT_STRUCT) {
1363 				printf(",0x%"PRIx16"", sample->ins_lat);
1364 				printf(",0x%"PRIx16"", sample->p_stage_cyc);
1365 			}
1366 		printf("\n");
1367 	}
1368 
1369 	if (sample_type & PERF_SAMPLE_DATA_SRC)
1370 		printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1371 
1372 	if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1373 		printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1374 
1375 	if (sample_type & PERF_SAMPLE_DATA_PAGE_SIZE)
1376 		printf(" .. data page size: %s\n", get_page_size_name(sample->data_page_size, str));
1377 
1378 	if (sample_type & PERF_SAMPLE_CODE_PAGE_SIZE)
1379 		printf(" .. code page size: %s\n", get_page_size_name(sample->code_page_size, str));
1380 
1381 	if (sample_type & PERF_SAMPLE_TRANSACTION)
1382 		printf("... transaction: %" PRIx64 "\n", sample->transaction);
1383 
1384 	if (sample_type & PERF_SAMPLE_READ)
1385 		sample_read__printf(sample, evsel->core.attr.read_format);
1386 }
1387 
dump_read(struct evsel * evsel,union perf_event * event)1388 static void dump_read(struct evsel *evsel, union perf_event *event)
1389 {
1390 	struct perf_record_read *read_event = &event->read;
1391 	u64 read_format;
1392 
1393 	if (!dump_trace)
1394 		return;
1395 
1396 	printf(": %d %d %s %" PRI_lu64 "\n", event->read.pid, event->read.tid,
1397 	       evsel__name(evsel), event->read.value);
1398 
1399 	if (!evsel)
1400 		return;
1401 
1402 	read_format = evsel->core.attr.read_format;
1403 
1404 	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1405 		printf("... time enabled : %" PRI_lu64 "\n", read_event->time_enabled);
1406 
1407 	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1408 		printf("... time running : %" PRI_lu64 "\n", read_event->time_running);
1409 
1410 	if (read_format & PERF_FORMAT_ID)
1411 		printf("... id           : %" PRI_lu64 "\n", read_event->id);
1412 }
1413 
machines__find_for_cpumode(struct machines * machines,union perf_event * event,struct perf_sample * sample)1414 static struct machine *machines__find_for_cpumode(struct machines *machines,
1415 					       union perf_event *event,
1416 					       struct perf_sample *sample)
1417 {
1418 	if (perf_guest &&
1419 	    ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1420 	     (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1421 		u32 pid;
1422 
1423 		if (event->header.type == PERF_RECORD_MMAP
1424 		    || event->header.type == PERF_RECORD_MMAP2)
1425 			pid = event->mmap.pid;
1426 		else
1427 			pid = sample->pid;
1428 
1429 		/*
1430 		 * Guest code machine is created as needed and does not use
1431 		 * DEFAULT_GUEST_KERNEL_ID.
1432 		 */
1433 		if (symbol_conf.guest_code)
1434 			return machines__findnew(machines, pid);
1435 
1436 		return machines__find_guest(machines, pid);
1437 	}
1438 
1439 	return &machines->host;
1440 }
1441 
deliver_sample_value(struct evlist * evlist,struct perf_tool * tool,union perf_event * event,struct perf_sample * sample,struct sample_read_value * v,struct machine * machine)1442 static int deliver_sample_value(struct evlist *evlist,
1443 				struct perf_tool *tool,
1444 				union perf_event *event,
1445 				struct perf_sample *sample,
1446 				struct sample_read_value *v,
1447 				struct machine *machine)
1448 {
1449 	struct perf_sample_id *sid = evlist__id2sid(evlist, v->id);
1450 	struct evsel *evsel;
1451 
1452 	if (sid) {
1453 		sample->id     = v->id;
1454 		sample->period = v->value - sid->period;
1455 		sid->period    = v->value;
1456 	}
1457 
1458 	if (!sid || sid->evsel == NULL) {
1459 		++evlist->stats.nr_unknown_id;
1460 		return 0;
1461 	}
1462 
1463 	/*
1464 	 * There's no reason to deliver sample
1465 	 * for zero period, bail out.
1466 	 */
1467 	if (!sample->period)
1468 		return 0;
1469 
1470 	evsel = container_of(sid->evsel, struct evsel, core);
1471 	return tool->sample(tool, event, sample, evsel, machine);
1472 }
1473 
deliver_sample_group(struct evlist * evlist,struct perf_tool * tool,union perf_event * event,struct perf_sample * sample,struct machine * machine)1474 static int deliver_sample_group(struct evlist *evlist,
1475 				struct perf_tool *tool,
1476 				union  perf_event *event,
1477 				struct perf_sample *sample,
1478 				struct machine *machine)
1479 {
1480 	int ret = -EINVAL;
1481 	u64 i;
1482 
1483 	for (i = 0; i < sample->read.group.nr; i++) {
1484 		ret = deliver_sample_value(evlist, tool, event, sample,
1485 					   &sample->read.group.values[i],
1486 					   machine);
1487 		if (ret)
1488 			break;
1489 	}
1490 
1491 	return ret;
1492 }
1493 
evlist__deliver_sample(struct evlist * evlist,struct perf_tool * tool,union perf_event * event,struct perf_sample * sample,struct evsel * evsel,struct machine * machine)1494 static int evlist__deliver_sample(struct evlist *evlist, struct perf_tool *tool,
1495 				  union  perf_event *event, struct perf_sample *sample,
1496 				  struct evsel *evsel, struct machine *machine)
1497 {
1498 	/* We know evsel != NULL. */
1499 	u64 sample_type = evsel->core.attr.sample_type;
1500 	u64 read_format = evsel->core.attr.read_format;
1501 
1502 	/* Standard sample delivery. */
1503 	if (!(sample_type & PERF_SAMPLE_READ))
1504 		return tool->sample(tool, event, sample, evsel, machine);
1505 
1506 	/* For PERF_SAMPLE_READ we have either single or group mode. */
1507 	if (read_format & PERF_FORMAT_GROUP)
1508 		return deliver_sample_group(evlist, tool, event, sample,
1509 					    machine);
1510 	else
1511 		return deliver_sample_value(evlist, tool, event, sample,
1512 					    &sample->read.one, machine);
1513 }
1514 
machines__deliver_event(struct machines * machines,struct evlist * evlist,union perf_event * event,struct perf_sample * sample,struct perf_tool * tool,u64 file_offset,const char * file_path)1515 static int machines__deliver_event(struct machines *machines,
1516 				   struct evlist *evlist,
1517 				   union perf_event *event,
1518 				   struct perf_sample *sample,
1519 				   struct perf_tool *tool, u64 file_offset,
1520 				   const char *file_path)
1521 {
1522 	struct evsel *evsel;
1523 	struct machine *machine;
1524 
1525 	dump_event(evlist, event, file_offset, sample, file_path);
1526 
1527 	evsel = evlist__id2evsel(evlist, sample->id);
1528 
1529 	machine = machines__find_for_cpumode(machines, event, sample);
1530 
1531 	switch (event->header.type) {
1532 	case PERF_RECORD_SAMPLE:
1533 		if (evsel == NULL) {
1534 			++evlist->stats.nr_unknown_id;
1535 			return 0;
1536 		}
1537 		if (machine == NULL) {
1538 			++evlist->stats.nr_unprocessable_samples;
1539 			dump_sample(evsel, event, sample, perf_env__arch(NULL));
1540 			return 0;
1541 		}
1542 		dump_sample(evsel, event, sample, perf_env__arch(machine->env));
1543 		return evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1544 	case PERF_RECORD_MMAP:
1545 		return tool->mmap(tool, event, sample, machine);
1546 	case PERF_RECORD_MMAP2:
1547 		if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1548 			++evlist->stats.nr_proc_map_timeout;
1549 		return tool->mmap2(tool, event, sample, machine);
1550 	case PERF_RECORD_COMM:
1551 		return tool->comm(tool, event, sample, machine);
1552 	case PERF_RECORD_NAMESPACES:
1553 		return tool->namespaces(tool, event, sample, machine);
1554 	case PERF_RECORD_CGROUP:
1555 		return tool->cgroup(tool, event, sample, machine);
1556 	case PERF_RECORD_FORK:
1557 		return tool->fork(tool, event, sample, machine);
1558 	case PERF_RECORD_EXIT:
1559 		return tool->exit(tool, event, sample, machine);
1560 	case PERF_RECORD_LOST:
1561 		if (tool->lost == perf_event__process_lost)
1562 			evlist->stats.total_lost += event->lost.lost;
1563 		return tool->lost(tool, event, sample, machine);
1564 	case PERF_RECORD_LOST_SAMPLES:
1565 		if (tool->lost_samples == perf_event__process_lost_samples)
1566 			evlist->stats.total_lost_samples += event->lost_samples.lost;
1567 		return tool->lost_samples(tool, event, sample, machine);
1568 	case PERF_RECORD_READ:
1569 		dump_read(evsel, event);
1570 		return tool->read(tool, event, sample, evsel, machine);
1571 	case PERF_RECORD_THROTTLE:
1572 		return tool->throttle(tool, event, sample, machine);
1573 	case PERF_RECORD_UNTHROTTLE:
1574 		return tool->unthrottle(tool, event, sample, machine);
1575 	case PERF_RECORD_AUX:
1576 		if (tool->aux == perf_event__process_aux) {
1577 			if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1578 				evlist->stats.total_aux_lost += 1;
1579 			if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1580 				evlist->stats.total_aux_partial += 1;
1581 			if (event->aux.flags & PERF_AUX_FLAG_COLLISION)
1582 				evlist->stats.total_aux_collision += 1;
1583 		}
1584 		return tool->aux(tool, event, sample, machine);
1585 	case PERF_RECORD_ITRACE_START:
1586 		return tool->itrace_start(tool, event, sample, machine);
1587 	case PERF_RECORD_SWITCH:
1588 	case PERF_RECORD_SWITCH_CPU_WIDE:
1589 		return tool->context_switch(tool, event, sample, machine);
1590 	case PERF_RECORD_KSYMBOL:
1591 		return tool->ksymbol(tool, event, sample, machine);
1592 	case PERF_RECORD_BPF_EVENT:
1593 		return tool->bpf(tool, event, sample, machine);
1594 	case PERF_RECORD_TEXT_POKE:
1595 		return tool->text_poke(tool, event, sample, machine);
1596 	case PERF_RECORD_AUX_OUTPUT_HW_ID:
1597 		return tool->aux_output_hw_id(tool, event, sample, machine);
1598 	default:
1599 		++evlist->stats.nr_unknown_events;
1600 		return -1;
1601 	}
1602 }
1603 
perf_session__deliver_event(struct perf_session * session,union perf_event * event,struct perf_tool * tool,u64 file_offset,const char * file_path)1604 static int perf_session__deliver_event(struct perf_session *session,
1605 				       union perf_event *event,
1606 				       struct perf_tool *tool,
1607 				       u64 file_offset,
1608 				       const char *file_path)
1609 {
1610 	struct perf_sample sample;
1611 	int ret = evlist__parse_sample(session->evlist, event, &sample);
1612 
1613 	if (ret) {
1614 		pr_err("Can't parse sample, err = %d\n", ret);
1615 		return ret;
1616 	}
1617 
1618 	ret = auxtrace__process_event(session, event, &sample, tool);
1619 	if (ret < 0)
1620 		return ret;
1621 	if (ret > 0)
1622 		return 0;
1623 
1624 	ret = machines__deliver_event(&session->machines, session->evlist,
1625 				      event, &sample, tool, file_offset, file_path);
1626 
1627 	if (dump_trace && sample.aux_sample.size)
1628 		auxtrace__dump_auxtrace_sample(session, &sample);
1629 
1630 	return ret;
1631 }
1632 
perf_session__process_user_event(struct perf_session * session,union perf_event * event,u64 file_offset,const char * file_path)1633 static s64 perf_session__process_user_event(struct perf_session *session,
1634 					    union perf_event *event,
1635 					    u64 file_offset,
1636 					    const char *file_path)
1637 {
1638 	struct ordered_events *oe = &session->ordered_events;
1639 	struct perf_tool *tool = session->tool;
1640 	struct perf_sample sample = { .time = 0, };
1641 	int fd = perf_data__fd(session->data);
1642 	int err;
1643 
1644 	if (event->header.type != PERF_RECORD_COMPRESSED ||
1645 	    tool->compressed == perf_session__process_compressed_event_stub)
1646 		dump_event(session->evlist, event, file_offset, &sample, file_path);
1647 
1648 	/* These events are processed right away */
1649 	switch (event->header.type) {
1650 	case PERF_RECORD_HEADER_ATTR:
1651 		err = tool->attr(tool, event, &session->evlist);
1652 		if (err == 0) {
1653 			perf_session__set_id_hdr_size(session);
1654 			perf_session__set_comm_exec(session);
1655 		}
1656 		return err;
1657 	case PERF_RECORD_EVENT_UPDATE:
1658 		return tool->event_update(tool, event, &session->evlist);
1659 	case PERF_RECORD_HEADER_EVENT_TYPE:
1660 		/*
1661 		 * Deprecated, but we need to handle it for sake
1662 		 * of old data files create in pipe mode.
1663 		 */
1664 		return 0;
1665 	case PERF_RECORD_HEADER_TRACING_DATA:
1666 		/*
1667 		 * Setup for reading amidst mmap, but only when we
1668 		 * are in 'file' mode. The 'pipe' fd is in proper
1669 		 * place already.
1670 		 */
1671 		if (!perf_data__is_pipe(session->data))
1672 			lseek(fd, file_offset, SEEK_SET);
1673 		return tool->tracing_data(session, event);
1674 	case PERF_RECORD_HEADER_BUILD_ID:
1675 		return tool->build_id(session, event);
1676 	case PERF_RECORD_FINISHED_ROUND:
1677 		return tool->finished_round(tool, event, oe);
1678 	case PERF_RECORD_ID_INDEX:
1679 		return tool->id_index(session, event);
1680 	case PERF_RECORD_AUXTRACE_INFO:
1681 		return tool->auxtrace_info(session, event);
1682 	case PERF_RECORD_AUXTRACE:
1683 		/* setup for reading amidst mmap */
1684 		lseek(fd, file_offset + event->header.size, SEEK_SET);
1685 		return tool->auxtrace(session, event);
1686 	case PERF_RECORD_AUXTRACE_ERROR:
1687 		perf_session__auxtrace_error_inc(session, event);
1688 		return tool->auxtrace_error(session, event);
1689 	case PERF_RECORD_THREAD_MAP:
1690 		return tool->thread_map(session, event);
1691 	case PERF_RECORD_CPU_MAP:
1692 		return tool->cpu_map(session, event);
1693 	case PERF_RECORD_STAT_CONFIG:
1694 		return tool->stat_config(session, event);
1695 	case PERF_RECORD_STAT:
1696 		return tool->stat(session, event);
1697 	case PERF_RECORD_STAT_ROUND:
1698 		return tool->stat_round(session, event);
1699 	case PERF_RECORD_TIME_CONV:
1700 		session->time_conv = event->time_conv;
1701 		return tool->time_conv(session, event);
1702 	case PERF_RECORD_HEADER_FEATURE:
1703 		return tool->feature(session, event);
1704 	case PERF_RECORD_COMPRESSED:
1705 		err = tool->compressed(session, event, file_offset, file_path);
1706 		if (err)
1707 			dump_event(session->evlist, event, file_offset, &sample, file_path);
1708 		return err;
1709 	default:
1710 		return -EINVAL;
1711 	}
1712 }
1713 
perf_session__deliver_synth_event(struct perf_session * session,union perf_event * event,struct perf_sample * sample)1714 int perf_session__deliver_synth_event(struct perf_session *session,
1715 				      union perf_event *event,
1716 				      struct perf_sample *sample)
1717 {
1718 	struct evlist *evlist = session->evlist;
1719 	struct perf_tool *tool = session->tool;
1720 
1721 	events_stats__inc(&evlist->stats, event->header.type);
1722 
1723 	if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1724 		return perf_session__process_user_event(session, event, 0, NULL);
1725 
1726 	return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0, NULL);
1727 }
1728 
event_swap(union perf_event * event,bool sample_id_all)1729 static void event_swap(union perf_event *event, bool sample_id_all)
1730 {
1731 	perf_event__swap_op swap;
1732 
1733 	swap = perf_event__swap_ops[event->header.type];
1734 	if (swap)
1735 		swap(event, sample_id_all);
1736 }
1737 
perf_session__peek_event(struct perf_session * session,off_t file_offset,void * buf,size_t buf_sz,union perf_event ** event_ptr,struct perf_sample * sample)1738 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1739 			     void *buf, size_t buf_sz,
1740 			     union perf_event **event_ptr,
1741 			     struct perf_sample *sample)
1742 {
1743 	union perf_event *event;
1744 	size_t hdr_sz, rest;
1745 	int fd;
1746 
1747 	if (session->one_mmap && !session->header.needs_swap) {
1748 		event = file_offset - session->one_mmap_offset +
1749 			session->one_mmap_addr;
1750 		goto out_parse_sample;
1751 	}
1752 
1753 	if (perf_data__is_pipe(session->data))
1754 		return -1;
1755 
1756 	fd = perf_data__fd(session->data);
1757 	hdr_sz = sizeof(struct perf_event_header);
1758 
1759 	if (buf_sz < hdr_sz)
1760 		return -1;
1761 
1762 	if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1763 	    readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1764 		return -1;
1765 
1766 	event = (union perf_event *)buf;
1767 
1768 	if (session->header.needs_swap)
1769 		perf_event_header__bswap(&event->header);
1770 
1771 	if (event->header.size < hdr_sz || event->header.size > buf_sz)
1772 		return -1;
1773 
1774 	buf += hdr_sz;
1775 	rest = event->header.size - hdr_sz;
1776 
1777 	if (readn(fd, buf, rest) != (ssize_t)rest)
1778 		return -1;
1779 
1780 	if (session->header.needs_swap)
1781 		event_swap(event, evlist__sample_id_all(session->evlist));
1782 
1783 out_parse_sample:
1784 
1785 	if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1786 	    evlist__parse_sample(session->evlist, event, sample))
1787 		return -1;
1788 
1789 	*event_ptr = event;
1790 
1791 	return 0;
1792 }
1793 
perf_session__peek_events(struct perf_session * session,u64 offset,u64 size,peek_events_cb_t cb,void * data)1794 int perf_session__peek_events(struct perf_session *session, u64 offset,
1795 			      u64 size, peek_events_cb_t cb, void *data)
1796 {
1797 	u64 max_offset = offset + size;
1798 	char buf[PERF_SAMPLE_MAX_SIZE];
1799 	union perf_event *event;
1800 	int err;
1801 
1802 	do {
1803 		err = perf_session__peek_event(session, offset, buf,
1804 					       PERF_SAMPLE_MAX_SIZE, &event,
1805 					       NULL);
1806 		if (err)
1807 			return err;
1808 
1809 		err = cb(session, event, offset, data);
1810 		if (err)
1811 			return err;
1812 
1813 		offset += event->header.size;
1814 		if (event->header.type == PERF_RECORD_AUXTRACE)
1815 			offset += event->auxtrace.size;
1816 
1817 	} while (offset < max_offset);
1818 
1819 	return err;
1820 }
1821 
perf_session__process_event(struct perf_session * session,union perf_event * event,u64 file_offset,const char * file_path)1822 static s64 perf_session__process_event(struct perf_session *session,
1823 				       union perf_event *event, u64 file_offset,
1824 				       const char *file_path)
1825 {
1826 	struct evlist *evlist = session->evlist;
1827 	struct perf_tool *tool = session->tool;
1828 	int ret;
1829 
1830 	if (session->header.needs_swap)
1831 		event_swap(event, evlist__sample_id_all(evlist));
1832 
1833 	if (event->header.type >= PERF_RECORD_HEADER_MAX)
1834 		return -EINVAL;
1835 
1836 	events_stats__inc(&evlist->stats, event->header.type);
1837 
1838 	if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1839 		return perf_session__process_user_event(session, event, file_offset, file_path);
1840 
1841 	if (tool->ordered_events) {
1842 		u64 timestamp = -1ULL;
1843 
1844 		ret = evlist__parse_sample_timestamp(evlist, event, &timestamp);
1845 		if (ret && ret != -1)
1846 			return ret;
1847 
1848 		ret = perf_session__queue_event(session, event, timestamp, file_offset, file_path);
1849 		if (ret != -ETIME)
1850 			return ret;
1851 	}
1852 
1853 	return perf_session__deliver_event(session, event, tool, file_offset, file_path);
1854 }
1855 
perf_event_header__bswap(struct perf_event_header * hdr)1856 void perf_event_header__bswap(struct perf_event_header *hdr)
1857 {
1858 	hdr->type = bswap_32(hdr->type);
1859 	hdr->misc = bswap_16(hdr->misc);
1860 	hdr->size = bswap_16(hdr->size);
1861 }
1862 
perf_session__findnew(struct perf_session * session,pid_t pid)1863 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1864 {
1865 	return machine__findnew_thread(&session->machines.host, -1, pid);
1866 }
1867 
perf_session__register_idle_thread(struct perf_session * session)1868 int perf_session__register_idle_thread(struct perf_session *session)
1869 {
1870 	struct thread *thread = machine__idle_thread(&session->machines.host);
1871 
1872 	/* machine__idle_thread() got the thread, so put it */
1873 	thread__put(thread);
1874 	return thread ? 0 : -1;
1875 }
1876 
1877 static void
perf_session__warn_order(const struct perf_session * session)1878 perf_session__warn_order(const struct perf_session *session)
1879 {
1880 	const struct ordered_events *oe = &session->ordered_events;
1881 	struct evsel *evsel;
1882 	bool should_warn = true;
1883 
1884 	evlist__for_each_entry(session->evlist, evsel) {
1885 		if (evsel->core.attr.write_backward)
1886 			should_warn = false;
1887 	}
1888 
1889 	if (!should_warn)
1890 		return;
1891 	if (oe->nr_unordered_events != 0)
1892 		ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1893 }
1894 
perf_session__warn_about_errors(const struct perf_session * session)1895 static void perf_session__warn_about_errors(const struct perf_session *session)
1896 {
1897 	const struct events_stats *stats = &session->evlist->stats;
1898 
1899 	if (session->tool->lost == perf_event__process_lost &&
1900 	    stats->nr_events[PERF_RECORD_LOST] != 0) {
1901 		ui__warning("Processed %d events and lost %d chunks!\n\n"
1902 			    "Check IO/CPU overload!\n\n",
1903 			    stats->nr_events[0],
1904 			    stats->nr_events[PERF_RECORD_LOST]);
1905 	}
1906 
1907 	if (session->tool->lost_samples == perf_event__process_lost_samples) {
1908 		double drop_rate;
1909 
1910 		drop_rate = (double)stats->total_lost_samples /
1911 			    (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1912 		if (drop_rate > 0.05) {
1913 			ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n",
1914 				    stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1915 				    drop_rate * 100.0);
1916 		}
1917 	}
1918 
1919 	if (session->tool->aux == perf_event__process_aux &&
1920 	    stats->total_aux_lost != 0) {
1921 		ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1922 			    stats->total_aux_lost,
1923 			    stats->nr_events[PERF_RECORD_AUX]);
1924 	}
1925 
1926 	if (session->tool->aux == perf_event__process_aux &&
1927 	    stats->total_aux_partial != 0) {
1928 		bool vmm_exclusive = false;
1929 
1930 		(void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1931 		                       &vmm_exclusive);
1932 
1933 		ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1934 		            "Are you running a KVM guest in the background?%s\n\n",
1935 			    stats->total_aux_partial,
1936 			    stats->nr_events[PERF_RECORD_AUX],
1937 			    vmm_exclusive ?
1938 			    "\nReloading kvm_intel module with vmm_exclusive=0\n"
1939 			    "will reduce the gaps to only guest's timeslices." :
1940 			    "");
1941 	}
1942 
1943 	if (session->tool->aux == perf_event__process_aux &&
1944 	    stats->total_aux_collision != 0) {
1945 		ui__warning("AUX data detected collision  %" PRIu64 " times out of %u!\n\n",
1946 			    stats->total_aux_collision,
1947 			    stats->nr_events[PERF_RECORD_AUX]);
1948 	}
1949 
1950 	if (stats->nr_unknown_events != 0) {
1951 		ui__warning("Found %u unknown events!\n\n"
1952 			    "Is this an older tool processing a perf.data "
1953 			    "file generated by a more recent tool?\n\n"
1954 			    "If that is not the case, consider "
1955 			    "reporting to linux-kernel@vger.kernel.org.\n\n",
1956 			    stats->nr_unknown_events);
1957 	}
1958 
1959 	if (stats->nr_unknown_id != 0) {
1960 		ui__warning("%u samples with id not present in the header\n",
1961 			    stats->nr_unknown_id);
1962 	}
1963 
1964 	if (stats->nr_invalid_chains != 0) {
1965 		ui__warning("Found invalid callchains!\n\n"
1966 			    "%u out of %u events were discarded for this reason.\n\n"
1967 			    "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1968 			    stats->nr_invalid_chains,
1969 			    stats->nr_events[PERF_RECORD_SAMPLE]);
1970 	}
1971 
1972 	if (stats->nr_unprocessable_samples != 0) {
1973 		ui__warning("%u unprocessable samples recorded.\n"
1974 			    "Do you have a KVM guest running and not using 'perf kvm'?\n",
1975 			    stats->nr_unprocessable_samples);
1976 	}
1977 
1978 	perf_session__warn_order(session);
1979 
1980 	events_stats__auxtrace_error_warn(stats);
1981 
1982 	if (stats->nr_proc_map_timeout != 0) {
1983 		ui__warning("%d map information files for pre-existing threads were\n"
1984 			    "not processed, if there are samples for addresses they\n"
1985 			    "will not be resolved, you may find out which are these\n"
1986 			    "threads by running with -v and redirecting the output\n"
1987 			    "to a file.\n"
1988 			    "The time limit to process proc map is too short?\n"
1989 			    "Increase it by --proc-map-timeout\n",
1990 			    stats->nr_proc_map_timeout);
1991 	}
1992 }
1993 
perf_session__flush_thread_stack(struct thread * thread,void * p __maybe_unused)1994 static int perf_session__flush_thread_stack(struct thread *thread,
1995 					    void *p __maybe_unused)
1996 {
1997 	return thread_stack__flush(thread);
1998 }
1999 
perf_session__flush_thread_stacks(struct perf_session * session)2000 static int perf_session__flush_thread_stacks(struct perf_session *session)
2001 {
2002 	return machines__for_each_thread(&session->machines,
2003 					 perf_session__flush_thread_stack,
2004 					 NULL);
2005 }
2006 
2007 volatile int session_done;
2008 
2009 static int __perf_session__process_decomp_events(struct perf_session *session);
2010 
__perf_session__process_pipe_events(struct perf_session * session)2011 static int __perf_session__process_pipe_events(struct perf_session *session)
2012 {
2013 	struct ordered_events *oe = &session->ordered_events;
2014 	struct perf_tool *tool = session->tool;
2015 	union perf_event *event;
2016 	uint32_t size, cur_size = 0;
2017 	void *buf = NULL;
2018 	s64 skip = 0;
2019 	u64 head;
2020 	ssize_t err;
2021 	void *p;
2022 
2023 	perf_tool__fill_defaults(tool);
2024 
2025 	head = 0;
2026 	cur_size = sizeof(union perf_event);
2027 
2028 	buf = malloc(cur_size);
2029 	if (!buf)
2030 		return -errno;
2031 	ordered_events__set_copy_on_queue(oe, true);
2032 more:
2033 	event = buf;
2034 	err = perf_data__read(session->data, event,
2035 			      sizeof(struct perf_event_header));
2036 	if (err <= 0) {
2037 		if (err == 0)
2038 			goto done;
2039 
2040 		pr_err("failed to read event header\n");
2041 		goto out_err;
2042 	}
2043 
2044 	if (session->header.needs_swap)
2045 		perf_event_header__bswap(&event->header);
2046 
2047 	size = event->header.size;
2048 	if (size < sizeof(struct perf_event_header)) {
2049 		pr_err("bad event header size\n");
2050 		goto out_err;
2051 	}
2052 
2053 	if (size > cur_size) {
2054 		void *new = realloc(buf, size);
2055 		if (!new) {
2056 			pr_err("failed to allocate memory to read event\n");
2057 			goto out_err;
2058 		}
2059 		buf = new;
2060 		cur_size = size;
2061 		event = buf;
2062 	}
2063 	p = event;
2064 	p += sizeof(struct perf_event_header);
2065 
2066 	if (size - sizeof(struct perf_event_header)) {
2067 		err = perf_data__read(session->data, p,
2068 				      size - sizeof(struct perf_event_header));
2069 		if (err <= 0) {
2070 			if (err == 0) {
2071 				pr_err("unexpected end of event stream\n");
2072 				goto done;
2073 			}
2074 
2075 			pr_err("failed to read event data\n");
2076 			goto out_err;
2077 		}
2078 	}
2079 
2080 	if ((skip = perf_session__process_event(session, event, head, "pipe")) < 0) {
2081 		pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2082 		       head, event->header.size, event->header.type);
2083 		err = -EINVAL;
2084 		goto out_err;
2085 	}
2086 
2087 	head += size;
2088 
2089 	if (skip > 0)
2090 		head += skip;
2091 
2092 	err = __perf_session__process_decomp_events(session);
2093 	if (err)
2094 		goto out_err;
2095 
2096 	if (!session_done())
2097 		goto more;
2098 done:
2099 	/* do the final flush for ordered samples */
2100 	err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2101 	if (err)
2102 		goto out_err;
2103 	err = auxtrace__flush_events(session, tool);
2104 	if (err)
2105 		goto out_err;
2106 	err = perf_session__flush_thread_stacks(session);
2107 out_err:
2108 	free(buf);
2109 	if (!tool->no_warn)
2110 		perf_session__warn_about_errors(session);
2111 	ordered_events__free(&session->ordered_events);
2112 	auxtrace__free_events(session);
2113 	return err;
2114 }
2115 
2116 static union perf_event *
prefetch_event(char * buf,u64 head,size_t mmap_size,bool needs_swap,union perf_event * error)2117 prefetch_event(char *buf, u64 head, size_t mmap_size,
2118 	       bool needs_swap, union perf_event *error)
2119 {
2120 	union perf_event *event;
2121 	u16 event_size;
2122 
2123 	/*
2124 	 * Ensure we have enough space remaining to read
2125 	 * the size of the event in the headers.
2126 	 */
2127 	if (head + sizeof(event->header) > mmap_size)
2128 		return NULL;
2129 
2130 	event = (union perf_event *)(buf + head);
2131 	if (needs_swap)
2132 		perf_event_header__bswap(&event->header);
2133 
2134 	event_size = event->header.size;
2135 	if (head + event_size <= mmap_size)
2136 		return event;
2137 
2138 	/* We're not fetching the event so swap back again */
2139 	if (needs_swap)
2140 		perf_event_header__bswap(&event->header);
2141 
2142 	/* Check if the event fits into the next mmapped buf. */
2143 	if (event_size <= mmap_size - head % page_size) {
2144 		/* Remap buf and fetch again. */
2145 		return NULL;
2146 	}
2147 
2148 	/* Invalid input. Event size should never exceed mmap_size. */
2149 	pr_debug("%s: head=%#" PRIx64 " event->header.size=%#x, mmap_size=%#zx:"
2150 		 " fuzzed or compressed perf.data?\n", __func__, head, event_size, mmap_size);
2151 
2152 	return error;
2153 }
2154 
2155 static union perf_event *
fetch_mmaped_event(u64 head,size_t mmap_size,char * buf,bool needs_swap)2156 fetch_mmaped_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2157 {
2158 	return prefetch_event(buf, head, mmap_size, needs_swap, ERR_PTR(-EINVAL));
2159 }
2160 
2161 static union perf_event *
fetch_decomp_event(u64 head,size_t mmap_size,char * buf,bool needs_swap)2162 fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2163 {
2164 	return prefetch_event(buf, head, mmap_size, needs_swap, NULL);
2165 }
2166 
__perf_session__process_decomp_events(struct perf_session * session)2167 static int __perf_session__process_decomp_events(struct perf_session *session)
2168 {
2169 	s64 skip;
2170 	u64 size;
2171 	struct decomp *decomp = session->active_decomp->decomp_last;
2172 
2173 	if (!decomp)
2174 		return 0;
2175 
2176 	while (decomp->head < decomp->size && !session_done()) {
2177 		union perf_event *event = fetch_decomp_event(decomp->head, decomp->size, decomp->data,
2178 							     session->header.needs_swap);
2179 
2180 		if (!event)
2181 			break;
2182 
2183 		size = event->header.size;
2184 
2185 		if (size < sizeof(struct perf_event_header) ||
2186 		    (skip = perf_session__process_event(session, event, decomp->file_pos,
2187 							decomp->file_path)) < 0) {
2188 			pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2189 				decomp->file_pos + decomp->head, event->header.size, event->header.type);
2190 			return -EINVAL;
2191 		}
2192 
2193 		if (skip)
2194 			size += skip;
2195 
2196 		decomp->head += size;
2197 	}
2198 
2199 	return 0;
2200 }
2201 
2202 /*
2203  * On 64bit we can mmap the data file in one go. No need for tiny mmap
2204  * slices. On 32bit we use 32MB.
2205  */
2206 #if BITS_PER_LONG == 64
2207 #define MMAP_SIZE ULLONG_MAX
2208 #define NUM_MMAPS 1
2209 #else
2210 #define MMAP_SIZE (32 * 1024 * 1024ULL)
2211 #define NUM_MMAPS 128
2212 #endif
2213 
2214 struct reader;
2215 
2216 typedef s64 (*reader_cb_t)(struct perf_session *session,
2217 			   union perf_event *event,
2218 			   u64 file_offset,
2219 			   const char *file_path);
2220 
2221 struct reader {
2222 	int		 fd;
2223 	const char	 *path;
2224 	u64		 data_size;
2225 	u64		 data_offset;
2226 	reader_cb_t	 process;
2227 	bool		 in_place_update;
2228 	char		 *mmaps[NUM_MMAPS];
2229 	size_t		 mmap_size;
2230 	int		 mmap_idx;
2231 	char		 *mmap_cur;
2232 	u64		 file_pos;
2233 	u64		 file_offset;
2234 	u64		 head;
2235 	u64		 size;
2236 	bool		 done;
2237 	struct zstd_data   zstd_data;
2238 	struct decomp_data decomp_data;
2239 };
2240 
2241 static int
reader__init(struct reader * rd,bool * one_mmap)2242 reader__init(struct reader *rd, bool *one_mmap)
2243 {
2244 	u64 data_size = rd->data_size;
2245 	char **mmaps = rd->mmaps;
2246 
2247 	rd->head = rd->data_offset;
2248 	data_size += rd->data_offset;
2249 
2250 	rd->mmap_size = MMAP_SIZE;
2251 	if (rd->mmap_size > data_size) {
2252 		rd->mmap_size = data_size;
2253 		if (one_mmap)
2254 			*one_mmap = true;
2255 	}
2256 
2257 	memset(mmaps, 0, sizeof(rd->mmaps));
2258 
2259 	if (zstd_init(&rd->zstd_data, 0))
2260 		return -1;
2261 	rd->decomp_data.zstd_decomp = &rd->zstd_data;
2262 
2263 	return 0;
2264 }
2265 
2266 static void
reader__release_decomp(struct reader * rd)2267 reader__release_decomp(struct reader *rd)
2268 {
2269 	perf_decomp__release_events(rd->decomp_data.decomp);
2270 	zstd_fini(&rd->zstd_data);
2271 }
2272 
2273 static int
reader__mmap(struct reader * rd,struct perf_session * session)2274 reader__mmap(struct reader *rd, struct perf_session *session)
2275 {
2276 	int mmap_prot, mmap_flags;
2277 	char *buf, **mmaps = rd->mmaps;
2278 	u64 page_offset;
2279 
2280 	mmap_prot  = PROT_READ;
2281 	mmap_flags = MAP_SHARED;
2282 
2283 	if (rd->in_place_update) {
2284 		mmap_prot  |= PROT_WRITE;
2285 	} else if (session->header.needs_swap) {
2286 		mmap_prot  |= PROT_WRITE;
2287 		mmap_flags = MAP_PRIVATE;
2288 	}
2289 
2290 	if (mmaps[rd->mmap_idx]) {
2291 		munmap(mmaps[rd->mmap_idx], rd->mmap_size);
2292 		mmaps[rd->mmap_idx] = NULL;
2293 	}
2294 
2295 	page_offset = page_size * (rd->head / page_size);
2296 	rd->file_offset += page_offset;
2297 	rd->head -= page_offset;
2298 
2299 	buf = mmap(NULL, rd->mmap_size, mmap_prot, mmap_flags, rd->fd,
2300 		   rd->file_offset);
2301 	if (buf == MAP_FAILED) {
2302 		pr_err("failed to mmap file\n");
2303 		return -errno;
2304 	}
2305 	mmaps[rd->mmap_idx] = rd->mmap_cur = buf;
2306 	rd->mmap_idx = (rd->mmap_idx + 1) & (ARRAY_SIZE(rd->mmaps) - 1);
2307 	rd->file_pos = rd->file_offset + rd->head;
2308 	if (session->one_mmap) {
2309 		session->one_mmap_addr = buf;
2310 		session->one_mmap_offset = rd->file_offset;
2311 	}
2312 
2313 	return 0;
2314 }
2315 
2316 enum {
2317 	READER_OK,
2318 	READER_NODATA,
2319 };
2320 
2321 static int
reader__read_event(struct reader * rd,struct perf_session * session,struct ui_progress * prog)2322 reader__read_event(struct reader *rd, struct perf_session *session,
2323 		   struct ui_progress *prog)
2324 {
2325 	u64 size;
2326 	int err = READER_OK;
2327 	union perf_event *event;
2328 	s64 skip;
2329 
2330 	event = fetch_mmaped_event(rd->head, rd->mmap_size, rd->mmap_cur,
2331 				   session->header.needs_swap);
2332 	if (IS_ERR(event))
2333 		return PTR_ERR(event);
2334 
2335 	if (!event)
2336 		return READER_NODATA;
2337 
2338 	size = event->header.size;
2339 
2340 	skip = -EINVAL;
2341 
2342 	if (size < sizeof(struct perf_event_header) ||
2343 	    (skip = rd->process(session, event, rd->file_pos, rd->path)) < 0) {
2344 		pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n",
2345 		       rd->file_offset + rd->head, event->header.size,
2346 		       event->header.type, strerror(-skip));
2347 		err = skip;
2348 		goto out;
2349 	}
2350 
2351 	if (skip)
2352 		size += skip;
2353 
2354 	rd->size += size;
2355 	rd->head += size;
2356 	rd->file_pos += size;
2357 
2358 	err = __perf_session__process_decomp_events(session);
2359 	if (err)
2360 		goto out;
2361 
2362 	ui_progress__update(prog, size);
2363 
2364 out:
2365 	return err;
2366 }
2367 
2368 static inline bool
reader__eof(struct reader * rd)2369 reader__eof(struct reader *rd)
2370 {
2371 	return (rd->file_pos >= rd->data_size + rd->data_offset);
2372 }
2373 
2374 static int
reader__process_events(struct reader * rd,struct perf_session * session,struct ui_progress * prog)2375 reader__process_events(struct reader *rd, struct perf_session *session,
2376 		       struct ui_progress *prog)
2377 {
2378 	int err;
2379 
2380 	err = reader__init(rd, &session->one_mmap);
2381 	if (err)
2382 		goto out;
2383 
2384 	session->active_decomp = &rd->decomp_data;
2385 
2386 remap:
2387 	err = reader__mmap(rd, session);
2388 	if (err)
2389 		goto out;
2390 
2391 more:
2392 	err = reader__read_event(rd, session, prog);
2393 	if (err < 0)
2394 		goto out;
2395 	else if (err == READER_NODATA)
2396 		goto remap;
2397 
2398 	if (session_done())
2399 		goto out;
2400 
2401 	if (!reader__eof(rd))
2402 		goto more;
2403 
2404 out:
2405 	session->active_decomp = &session->decomp_data;
2406 	return err;
2407 }
2408 
process_simple(struct perf_session * session,union perf_event * event,u64 file_offset,const char * file_path)2409 static s64 process_simple(struct perf_session *session,
2410 			  union perf_event *event,
2411 			  u64 file_offset,
2412 			  const char *file_path)
2413 {
2414 	return perf_session__process_event(session, event, file_offset, file_path);
2415 }
2416 
__perf_session__process_events(struct perf_session * session)2417 static int __perf_session__process_events(struct perf_session *session)
2418 {
2419 	struct reader rd = {
2420 		.fd		= perf_data__fd(session->data),
2421 		.path		= session->data->file.path,
2422 		.data_size	= session->header.data_size,
2423 		.data_offset	= session->header.data_offset,
2424 		.process	= process_simple,
2425 		.in_place_update = session->data->in_place_update,
2426 	};
2427 	struct ordered_events *oe = &session->ordered_events;
2428 	struct perf_tool *tool = session->tool;
2429 	struct ui_progress prog;
2430 	int err;
2431 
2432 	perf_tool__fill_defaults(tool);
2433 
2434 	if (rd.data_size == 0)
2435 		return -1;
2436 
2437 	ui_progress__init_size(&prog, rd.data_size, "Processing events...");
2438 
2439 	err = reader__process_events(&rd, session, &prog);
2440 	if (err)
2441 		goto out_err;
2442 	/* do the final flush for ordered samples */
2443 	err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2444 	if (err)
2445 		goto out_err;
2446 	err = auxtrace__flush_events(session, tool);
2447 	if (err)
2448 		goto out_err;
2449 	err = perf_session__flush_thread_stacks(session);
2450 out_err:
2451 	ui_progress__finish();
2452 	if (!tool->no_warn)
2453 		perf_session__warn_about_errors(session);
2454 	/*
2455 	 * We may switching perf.data output, make ordered_events
2456 	 * reusable.
2457 	 */
2458 	ordered_events__reinit(&session->ordered_events);
2459 	auxtrace__free_events(session);
2460 	reader__release_decomp(&rd);
2461 	session->one_mmap = false;
2462 	return err;
2463 }
2464 
2465 /*
2466  * Processing 2 MB of data from each reader in sequence,
2467  * because that's the way the ordered events sorting works
2468  * most efficiently.
2469  */
2470 #define READER_MAX_SIZE (2 * 1024 * 1024)
2471 
2472 /*
2473  * This function reads, merge and process directory data.
2474  * It assumens the version 1 of directory data, where each
2475  * data file holds per-cpu data, already sorted by kernel.
2476  */
__perf_session__process_dir_events(struct perf_session * session)2477 static int __perf_session__process_dir_events(struct perf_session *session)
2478 {
2479 	struct perf_data *data = session->data;
2480 	struct perf_tool *tool = session->tool;
2481 	int i, ret, readers, nr_readers;
2482 	struct ui_progress prog;
2483 	u64 total_size = perf_data__size(session->data);
2484 	struct reader *rd;
2485 
2486 	perf_tool__fill_defaults(tool);
2487 
2488 	ui_progress__init_size(&prog, total_size, "Sorting events...");
2489 
2490 	nr_readers = 1;
2491 	for (i = 0; i < data->dir.nr; i++) {
2492 		if (data->dir.files[i].size)
2493 			nr_readers++;
2494 	}
2495 
2496 	rd = zalloc(nr_readers * sizeof(struct reader));
2497 	if (!rd)
2498 		return -ENOMEM;
2499 
2500 	rd[0] = (struct reader) {
2501 		.fd		 = perf_data__fd(session->data),
2502 		.path		 = session->data->file.path,
2503 		.data_size	 = session->header.data_size,
2504 		.data_offset	 = session->header.data_offset,
2505 		.process	 = process_simple,
2506 		.in_place_update = session->data->in_place_update,
2507 	};
2508 	ret = reader__init(&rd[0], NULL);
2509 	if (ret)
2510 		goto out_err;
2511 	ret = reader__mmap(&rd[0], session);
2512 	if (ret)
2513 		goto out_err;
2514 	readers = 1;
2515 
2516 	for (i = 0; i < data->dir.nr; i++) {
2517 		if (!data->dir.files[i].size)
2518 			continue;
2519 		rd[readers] = (struct reader) {
2520 			.fd		 = data->dir.files[i].fd,
2521 			.path		 = data->dir.files[i].path,
2522 			.data_size	 = data->dir.files[i].size,
2523 			.data_offset	 = 0,
2524 			.process	 = process_simple,
2525 			.in_place_update = session->data->in_place_update,
2526 		};
2527 		ret = reader__init(&rd[readers], NULL);
2528 		if (ret)
2529 			goto out_err;
2530 		ret = reader__mmap(&rd[readers], session);
2531 		if (ret)
2532 			goto out_err;
2533 		readers++;
2534 	}
2535 
2536 	i = 0;
2537 	while (readers) {
2538 		if (session_done())
2539 			break;
2540 
2541 		if (rd[i].done) {
2542 			i = (i + 1) % nr_readers;
2543 			continue;
2544 		}
2545 		if (reader__eof(&rd[i])) {
2546 			rd[i].done = true;
2547 			readers--;
2548 			continue;
2549 		}
2550 
2551 		session->active_decomp = &rd[i].decomp_data;
2552 		ret = reader__read_event(&rd[i], session, &prog);
2553 		if (ret < 0) {
2554 			goto out_err;
2555 		} else if (ret == READER_NODATA) {
2556 			ret = reader__mmap(&rd[i], session);
2557 			if (ret)
2558 				goto out_err;
2559 		}
2560 
2561 		if (rd[i].size >= READER_MAX_SIZE) {
2562 			rd[i].size = 0;
2563 			i = (i + 1) % nr_readers;
2564 		}
2565 	}
2566 
2567 	ret = ordered_events__flush(&session->ordered_events, OE_FLUSH__FINAL);
2568 	if (ret)
2569 		goto out_err;
2570 
2571 	ret = perf_session__flush_thread_stacks(session);
2572 out_err:
2573 	ui_progress__finish();
2574 
2575 	if (!tool->no_warn)
2576 		perf_session__warn_about_errors(session);
2577 
2578 	/*
2579 	 * We may switching perf.data output, make ordered_events
2580 	 * reusable.
2581 	 */
2582 	ordered_events__reinit(&session->ordered_events);
2583 
2584 	session->one_mmap = false;
2585 
2586 	session->active_decomp = &session->decomp_data;
2587 	for (i = 0; i < nr_readers; i++)
2588 		reader__release_decomp(&rd[i]);
2589 	zfree(&rd);
2590 
2591 	return ret;
2592 }
2593 
perf_session__process_events(struct perf_session * session)2594 int perf_session__process_events(struct perf_session *session)
2595 {
2596 	if (perf_session__register_idle_thread(session) < 0)
2597 		return -ENOMEM;
2598 
2599 	if (perf_data__is_pipe(session->data))
2600 		return __perf_session__process_pipe_events(session);
2601 
2602 	if (perf_data__is_dir(session->data) && session->data->dir.nr)
2603 		return __perf_session__process_dir_events(session);
2604 
2605 	return __perf_session__process_events(session);
2606 }
2607 
perf_session__has_traces(struct perf_session * session,const char * msg)2608 bool perf_session__has_traces(struct perf_session *session, const char *msg)
2609 {
2610 	struct evsel *evsel;
2611 
2612 	evlist__for_each_entry(session->evlist, evsel) {
2613 		if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT)
2614 			return true;
2615 	}
2616 
2617 	pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
2618 	return false;
2619 }
2620 
map__set_kallsyms_ref_reloc_sym(struct map * map,const char * symbol_name,u64 addr)2621 int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr)
2622 {
2623 	char *bracket;
2624 	struct ref_reloc_sym *ref;
2625 	struct kmap *kmap;
2626 
2627 	ref = zalloc(sizeof(struct ref_reloc_sym));
2628 	if (ref == NULL)
2629 		return -ENOMEM;
2630 
2631 	ref->name = strdup(symbol_name);
2632 	if (ref->name == NULL) {
2633 		free(ref);
2634 		return -ENOMEM;
2635 	}
2636 
2637 	bracket = strchr(ref->name, ']');
2638 	if (bracket)
2639 		*bracket = '\0';
2640 
2641 	ref->addr = addr;
2642 
2643 	kmap = map__kmap(map);
2644 	if (kmap)
2645 		kmap->ref_reloc_sym = ref;
2646 
2647 	return 0;
2648 }
2649 
perf_session__fprintf_dsos(struct perf_session * session,FILE * fp)2650 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2651 {
2652 	return machines__fprintf_dsos(&session->machines, fp);
2653 }
2654 
perf_session__fprintf_dsos_buildid(struct perf_session * session,FILE * fp,bool (skip)(struct dso * dso,int parm),int parm)2655 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2656 					  bool (skip)(struct dso *dso, int parm), int parm)
2657 {
2658 	return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2659 }
2660 
perf_session__fprintf_nr_events(struct perf_session * session,FILE * fp,bool skip_empty)2661 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp,
2662 				       bool skip_empty)
2663 {
2664 	size_t ret;
2665 	const char *msg = "";
2666 
2667 	if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2668 		msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2669 
2670 	ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2671 
2672 	ret += events_stats__fprintf(&session->evlist->stats, fp, skip_empty);
2673 	return ret;
2674 }
2675 
perf_session__fprintf(struct perf_session * session,FILE * fp)2676 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2677 {
2678 	/*
2679 	 * FIXME: Here we have to actually print all the machines in this
2680 	 * session, not just the host...
2681 	 */
2682 	return machine__fprintf(&session->machines.host, fp);
2683 }
2684 
perf_session__find_first_evtype(struct perf_session * session,unsigned int type)2685 struct evsel *perf_session__find_first_evtype(struct perf_session *session,
2686 					      unsigned int type)
2687 {
2688 	struct evsel *pos;
2689 
2690 	evlist__for_each_entry(session->evlist, pos) {
2691 		if (pos->core.attr.type == type)
2692 			return pos;
2693 	}
2694 	return NULL;
2695 }
2696 
perf_session__cpu_bitmap(struct perf_session * session,const char * cpu_list,unsigned long * cpu_bitmap)2697 int perf_session__cpu_bitmap(struct perf_session *session,
2698 			     const char *cpu_list, unsigned long *cpu_bitmap)
2699 {
2700 	int i, err = -1;
2701 	struct perf_cpu_map *map;
2702 	int nr_cpus = min(session->header.env.nr_cpus_avail, MAX_NR_CPUS);
2703 
2704 	for (i = 0; i < PERF_TYPE_MAX; ++i) {
2705 		struct evsel *evsel;
2706 
2707 		evsel = perf_session__find_first_evtype(session, i);
2708 		if (!evsel)
2709 			continue;
2710 
2711 		if (!(evsel->core.attr.sample_type & PERF_SAMPLE_CPU)) {
2712 			pr_err("File does not contain CPU events. "
2713 			       "Remove -C option to proceed.\n");
2714 			return -1;
2715 		}
2716 	}
2717 
2718 	map = perf_cpu_map__new(cpu_list);
2719 	if (map == NULL) {
2720 		pr_err("Invalid cpu_list\n");
2721 		return -1;
2722 	}
2723 
2724 	for (i = 0; i < perf_cpu_map__nr(map); i++) {
2725 		struct perf_cpu cpu = perf_cpu_map__cpu(map, i);
2726 
2727 		if (cpu.cpu >= nr_cpus) {
2728 			pr_err("Requested CPU %d too large. "
2729 			       "Consider raising MAX_NR_CPUS\n", cpu.cpu);
2730 			goto out_delete_map;
2731 		}
2732 
2733 		set_bit(cpu.cpu, cpu_bitmap);
2734 	}
2735 
2736 	err = 0;
2737 
2738 out_delete_map:
2739 	perf_cpu_map__put(map);
2740 	return err;
2741 }
2742 
perf_session__fprintf_info(struct perf_session * session,FILE * fp,bool full)2743 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2744 				bool full)
2745 {
2746 	if (session == NULL || fp == NULL)
2747 		return;
2748 
2749 	fprintf(fp, "# ========\n");
2750 	perf_header__fprintf_info(session, fp, full);
2751 	fprintf(fp, "# ========\n#\n");
2752 }
2753 
perf_event__process_id_index(struct perf_session * session,union perf_event * event)2754 int perf_event__process_id_index(struct perf_session *session,
2755 				 union perf_event *event)
2756 {
2757 	struct evlist *evlist = session->evlist;
2758 	struct perf_record_id_index *ie = &event->id_index;
2759 	size_t i, nr, max_nr;
2760 
2761 	max_nr = (ie->header.size - sizeof(struct perf_record_id_index)) /
2762 		 sizeof(struct id_index_entry);
2763 	nr = ie->nr;
2764 	if (nr > max_nr)
2765 		return -EINVAL;
2766 
2767 	if (dump_trace)
2768 		fprintf(stdout, " nr: %zu\n", nr);
2769 
2770 	for (i = 0; i < nr; i++) {
2771 		struct id_index_entry *e = &ie->entries[i];
2772 		struct perf_sample_id *sid;
2773 
2774 		if (dump_trace) {
2775 			fprintf(stdout,	" ... id: %"PRI_lu64, e->id);
2776 			fprintf(stdout,	"  idx: %"PRI_lu64, e->idx);
2777 			fprintf(stdout,	"  cpu: %"PRI_ld64, e->cpu);
2778 			fprintf(stdout,	"  tid: %"PRI_ld64"\n", e->tid);
2779 		}
2780 
2781 		sid = evlist__id2sid(evlist, e->id);
2782 		if (!sid)
2783 			return -ENOENT;
2784 		sid->idx = e->idx;
2785 		sid->cpu.cpu = e->cpu;
2786 		sid->tid = e->tid;
2787 	}
2788 	return 0;
2789 }
2790