xref: /openbmc/linux/tools/perf/util/intel-bts.c (revision 547840bd)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * intel-bts.c: Intel Processor Trace support
4  * Copyright (c) 2013-2015, Intel Corporation.
5  */
6 
7 #include <endian.h>
8 #include <errno.h>
9 #include <byteswap.h>
10 #include <inttypes.h>
11 #include <linux/kernel.h>
12 #include <linux/types.h>
13 #include <linux/bitops.h>
14 #include <linux/log2.h>
15 #include <linux/zalloc.h>
16 
17 #include "color.h"
18 #include "evsel.h"
19 #include "evlist.h"
20 #include "machine.h"
21 #include "symbol.h"
22 #include "session.h"
23 #include "tool.h"
24 #include "thread.h"
25 #include "thread-stack.h"
26 #include "debug.h"
27 #include "tsc.h"
28 #include "auxtrace.h"
29 #include "intel-pt-decoder/intel-pt-insn-decoder.h"
30 #include "intel-bts.h"
31 #include "util/synthetic-events.h"
32 
33 #define MAX_TIMESTAMP (~0ULL)
34 
35 #define INTEL_BTS_ERR_NOINSN  5
36 #define INTEL_BTS_ERR_LOST    9
37 
38 #if __BYTE_ORDER == __BIG_ENDIAN
39 #define le64_to_cpu bswap_64
40 #else
41 #define le64_to_cpu
42 #endif
43 
44 struct intel_bts {
45 	struct auxtrace			auxtrace;
46 	struct auxtrace_queues		queues;
47 	struct auxtrace_heap		heap;
48 	u32				auxtrace_type;
49 	struct perf_session		*session;
50 	struct machine			*machine;
51 	bool				sampling_mode;
52 	bool				snapshot_mode;
53 	bool				data_queued;
54 	u32				pmu_type;
55 	struct perf_tsc_conversion	tc;
56 	bool				cap_user_time_zero;
57 	struct itrace_synth_opts	synth_opts;
58 	bool				sample_branches;
59 	u32				branches_filter;
60 	u64				branches_sample_type;
61 	u64				branches_id;
62 	size_t				branches_event_size;
63 	unsigned long			num_events;
64 };
65 
66 struct intel_bts_queue {
67 	struct intel_bts	*bts;
68 	unsigned int		queue_nr;
69 	struct auxtrace_buffer	*buffer;
70 	bool			on_heap;
71 	bool			done;
72 	pid_t			pid;
73 	pid_t			tid;
74 	int			cpu;
75 	u64			time;
76 	struct intel_pt_insn	intel_pt_insn;
77 	u32			sample_flags;
78 };
79 
80 struct branch {
81 	u64 from;
82 	u64 to;
83 	u64 misc;
84 };
85 
86 static void intel_bts_dump(struct intel_bts *bts __maybe_unused,
87 			   unsigned char *buf, size_t len)
88 {
89 	struct branch *branch;
90 	size_t i, pos = 0, br_sz = sizeof(struct branch), sz;
91 	const char *color = PERF_COLOR_BLUE;
92 
93 	color_fprintf(stdout, color,
94 		      ". ... Intel BTS data: size %zu bytes\n",
95 		      len);
96 
97 	while (len) {
98 		if (len >= br_sz)
99 			sz = br_sz;
100 		else
101 			sz = len;
102 		printf(".");
103 		color_fprintf(stdout, color, "  %08x: ", pos);
104 		for (i = 0; i < sz; i++)
105 			color_fprintf(stdout, color, " %02x", buf[i]);
106 		for (; i < br_sz; i++)
107 			color_fprintf(stdout, color, "   ");
108 		if (len >= br_sz) {
109 			branch = (struct branch *)buf;
110 			color_fprintf(stdout, color, " %"PRIx64" -> %"PRIx64" %s\n",
111 				      le64_to_cpu(branch->from),
112 				      le64_to_cpu(branch->to),
113 				      le64_to_cpu(branch->misc) & 0x10 ?
114 							"pred" : "miss");
115 		} else {
116 			color_fprintf(stdout, color, " Bad record!\n");
117 		}
118 		pos += sz;
119 		buf += sz;
120 		len -= sz;
121 	}
122 }
123 
124 static void intel_bts_dump_event(struct intel_bts *bts, unsigned char *buf,
125 				 size_t len)
126 {
127 	printf(".\n");
128 	intel_bts_dump(bts, buf, len);
129 }
130 
131 static int intel_bts_lost(struct intel_bts *bts, struct perf_sample *sample)
132 {
133 	union perf_event event;
134 	int err;
135 
136 	auxtrace_synth_error(&event.auxtrace_error, PERF_AUXTRACE_ERROR_ITRACE,
137 			     INTEL_BTS_ERR_LOST, sample->cpu, sample->pid,
138 			     sample->tid, 0, "Lost trace data", sample->time);
139 
140 	err = perf_session__deliver_synth_event(bts->session, &event, NULL);
141 	if (err)
142 		pr_err("Intel BTS: failed to deliver error event, error %d\n",
143 		       err);
144 
145 	return err;
146 }
147 
148 static struct intel_bts_queue *intel_bts_alloc_queue(struct intel_bts *bts,
149 						     unsigned int queue_nr)
150 {
151 	struct intel_bts_queue *btsq;
152 
153 	btsq = zalloc(sizeof(struct intel_bts_queue));
154 	if (!btsq)
155 		return NULL;
156 
157 	btsq->bts = bts;
158 	btsq->queue_nr = queue_nr;
159 	btsq->pid = -1;
160 	btsq->tid = -1;
161 	btsq->cpu = -1;
162 
163 	return btsq;
164 }
165 
166 static int intel_bts_setup_queue(struct intel_bts *bts,
167 				 struct auxtrace_queue *queue,
168 				 unsigned int queue_nr)
169 {
170 	struct intel_bts_queue *btsq = queue->priv;
171 
172 	if (list_empty(&queue->head))
173 		return 0;
174 
175 	if (!btsq) {
176 		btsq = intel_bts_alloc_queue(bts, queue_nr);
177 		if (!btsq)
178 			return -ENOMEM;
179 		queue->priv = btsq;
180 
181 		if (queue->cpu != -1)
182 			btsq->cpu = queue->cpu;
183 		btsq->tid = queue->tid;
184 	}
185 
186 	if (bts->sampling_mode)
187 		return 0;
188 
189 	if (!btsq->on_heap && !btsq->buffer) {
190 		int ret;
191 
192 		btsq->buffer = auxtrace_buffer__next(queue, NULL);
193 		if (!btsq->buffer)
194 			return 0;
195 
196 		ret = auxtrace_heap__add(&bts->heap, queue_nr,
197 					 btsq->buffer->reference);
198 		if (ret)
199 			return ret;
200 		btsq->on_heap = true;
201 	}
202 
203 	return 0;
204 }
205 
206 static int intel_bts_setup_queues(struct intel_bts *bts)
207 {
208 	unsigned int i;
209 	int ret;
210 
211 	for (i = 0; i < bts->queues.nr_queues; i++) {
212 		ret = intel_bts_setup_queue(bts, &bts->queues.queue_array[i],
213 					    i);
214 		if (ret)
215 			return ret;
216 	}
217 	return 0;
218 }
219 
220 static inline int intel_bts_update_queues(struct intel_bts *bts)
221 {
222 	if (bts->queues.new_data) {
223 		bts->queues.new_data = false;
224 		return intel_bts_setup_queues(bts);
225 	}
226 	return 0;
227 }
228 
229 static unsigned char *intel_bts_find_overlap(unsigned char *buf_a, size_t len_a,
230 					     unsigned char *buf_b, size_t len_b)
231 {
232 	size_t offs, len;
233 
234 	if (len_a > len_b)
235 		offs = len_a - len_b;
236 	else
237 		offs = 0;
238 
239 	for (; offs < len_a; offs += sizeof(struct branch)) {
240 		len = len_a - offs;
241 		if (!memcmp(buf_a + offs, buf_b, len))
242 			return buf_b + len;
243 	}
244 
245 	return buf_b;
246 }
247 
248 static int intel_bts_do_fix_overlap(struct auxtrace_queue *queue,
249 				    struct auxtrace_buffer *b)
250 {
251 	struct auxtrace_buffer *a;
252 	void *start;
253 
254 	if (b->list.prev == &queue->head)
255 		return 0;
256 	a = list_entry(b->list.prev, struct auxtrace_buffer, list);
257 	start = intel_bts_find_overlap(a->data, a->size, b->data, b->size);
258 	if (!start)
259 		return -EINVAL;
260 	b->use_size = b->data + b->size - start;
261 	b->use_data = start;
262 	return 0;
263 }
264 
265 static inline u8 intel_bts_cpumode(struct intel_bts *bts, uint64_t ip)
266 {
267 	return machine__kernel_ip(bts->machine, ip) ?
268 	       PERF_RECORD_MISC_KERNEL :
269 	       PERF_RECORD_MISC_USER;
270 }
271 
272 static int intel_bts_synth_branch_sample(struct intel_bts_queue *btsq,
273 					 struct branch *branch)
274 {
275 	int ret;
276 	struct intel_bts *bts = btsq->bts;
277 	union perf_event event;
278 	struct perf_sample sample = { .ip = 0, };
279 
280 	if (bts->synth_opts.initial_skip &&
281 	    bts->num_events++ <= bts->synth_opts.initial_skip)
282 		return 0;
283 
284 	sample.ip = le64_to_cpu(branch->from);
285 	sample.cpumode = intel_bts_cpumode(bts, sample.ip);
286 	sample.pid = btsq->pid;
287 	sample.tid = btsq->tid;
288 	sample.addr = le64_to_cpu(branch->to);
289 	sample.id = btsq->bts->branches_id;
290 	sample.stream_id = btsq->bts->branches_id;
291 	sample.period = 1;
292 	sample.cpu = btsq->cpu;
293 	sample.flags = btsq->sample_flags;
294 	sample.insn_len = btsq->intel_pt_insn.length;
295 	memcpy(sample.insn, btsq->intel_pt_insn.buf, INTEL_PT_INSN_BUF_SZ);
296 
297 	event.sample.header.type = PERF_RECORD_SAMPLE;
298 	event.sample.header.misc = sample.cpumode;
299 	event.sample.header.size = sizeof(struct perf_event_header);
300 
301 	if (bts->synth_opts.inject) {
302 		event.sample.header.size = bts->branches_event_size;
303 		ret = perf_event__synthesize_sample(&event,
304 						    bts->branches_sample_type,
305 						    0, &sample);
306 		if (ret)
307 			return ret;
308 	}
309 
310 	ret = perf_session__deliver_synth_event(bts->session, &event, &sample);
311 	if (ret)
312 		pr_err("Intel BTS: failed to deliver branch event, error %d\n",
313 		       ret);
314 
315 	return ret;
316 }
317 
318 static int intel_bts_get_next_insn(struct intel_bts_queue *btsq, u64 ip)
319 {
320 	struct machine *machine = btsq->bts->machine;
321 	struct thread *thread;
322 	unsigned char buf[INTEL_PT_INSN_BUF_SZ];
323 	ssize_t len;
324 	bool x86_64;
325 	int err = -1;
326 
327 	thread = machine__find_thread(machine, -1, btsq->tid);
328 	if (!thread)
329 		return -1;
330 
331 	len = thread__memcpy(thread, machine, buf, ip, INTEL_PT_INSN_BUF_SZ, &x86_64);
332 	if (len <= 0)
333 		goto out_put;
334 
335 	if (intel_pt_get_insn(buf, len, x86_64, &btsq->intel_pt_insn))
336 		goto out_put;
337 
338 	err = 0;
339 out_put:
340 	thread__put(thread);
341 	return err;
342 }
343 
344 static int intel_bts_synth_error(struct intel_bts *bts, int cpu, pid_t pid,
345 				 pid_t tid, u64 ip)
346 {
347 	union perf_event event;
348 	int err;
349 
350 	auxtrace_synth_error(&event.auxtrace_error, PERF_AUXTRACE_ERROR_ITRACE,
351 			     INTEL_BTS_ERR_NOINSN, cpu, pid, tid, ip,
352 			     "Failed to get instruction", 0);
353 
354 	err = perf_session__deliver_synth_event(bts->session, &event, NULL);
355 	if (err)
356 		pr_err("Intel BTS: failed to deliver error event, error %d\n",
357 		       err);
358 
359 	return err;
360 }
361 
362 static int intel_bts_get_branch_type(struct intel_bts_queue *btsq,
363 				     struct branch *branch)
364 {
365 	int err;
366 
367 	if (!branch->from) {
368 		if (branch->to)
369 			btsq->sample_flags = PERF_IP_FLAG_BRANCH |
370 					     PERF_IP_FLAG_TRACE_BEGIN;
371 		else
372 			btsq->sample_flags = 0;
373 		btsq->intel_pt_insn.length = 0;
374 	} else if (!branch->to) {
375 		btsq->sample_flags = PERF_IP_FLAG_BRANCH |
376 				     PERF_IP_FLAG_TRACE_END;
377 		btsq->intel_pt_insn.length = 0;
378 	} else {
379 		err = intel_bts_get_next_insn(btsq, branch->from);
380 		if (err) {
381 			btsq->sample_flags = 0;
382 			btsq->intel_pt_insn.length = 0;
383 			if (!btsq->bts->synth_opts.errors)
384 				return 0;
385 			err = intel_bts_synth_error(btsq->bts, btsq->cpu,
386 						    btsq->pid, btsq->tid,
387 						    branch->from);
388 			return err;
389 		}
390 		btsq->sample_flags = intel_pt_insn_type(btsq->intel_pt_insn.op);
391 		/* Check for an async branch into the kernel */
392 		if (!machine__kernel_ip(btsq->bts->machine, branch->from) &&
393 		    machine__kernel_ip(btsq->bts->machine, branch->to) &&
394 		    btsq->sample_flags != (PERF_IP_FLAG_BRANCH |
395 					   PERF_IP_FLAG_CALL |
396 					   PERF_IP_FLAG_SYSCALLRET))
397 			btsq->sample_flags = PERF_IP_FLAG_BRANCH |
398 					     PERF_IP_FLAG_CALL |
399 					     PERF_IP_FLAG_ASYNC |
400 					     PERF_IP_FLAG_INTERRUPT;
401 	}
402 
403 	return 0;
404 }
405 
406 static int intel_bts_process_buffer(struct intel_bts_queue *btsq,
407 				    struct auxtrace_buffer *buffer,
408 				    struct thread *thread)
409 {
410 	struct branch *branch;
411 	size_t sz, bsz = sizeof(struct branch);
412 	u32 filter = btsq->bts->branches_filter;
413 	int err = 0;
414 
415 	if (buffer->use_data) {
416 		sz = buffer->use_size;
417 		branch = buffer->use_data;
418 	} else {
419 		sz = buffer->size;
420 		branch = buffer->data;
421 	}
422 
423 	if (!btsq->bts->sample_branches)
424 		return 0;
425 
426 	for (; sz > bsz; branch += 1, sz -= bsz) {
427 		if (!branch->from && !branch->to)
428 			continue;
429 		intel_bts_get_branch_type(btsq, branch);
430 		if (btsq->bts->synth_opts.thread_stack)
431 			thread_stack__event(thread, btsq->cpu, btsq->sample_flags,
432 					    le64_to_cpu(branch->from),
433 					    le64_to_cpu(branch->to),
434 					    btsq->intel_pt_insn.length,
435 					    buffer->buffer_nr + 1, true, 0, 0);
436 		if (filter && !(filter & btsq->sample_flags))
437 			continue;
438 		err = intel_bts_synth_branch_sample(btsq, branch);
439 		if (err)
440 			break;
441 	}
442 	return err;
443 }
444 
445 static int intel_bts_process_queue(struct intel_bts_queue *btsq, u64 *timestamp)
446 {
447 	struct auxtrace_buffer *buffer = btsq->buffer, *old_buffer = buffer;
448 	struct auxtrace_queue *queue;
449 	struct thread *thread;
450 	int err;
451 
452 	if (btsq->done)
453 		return 1;
454 
455 	if (btsq->pid == -1) {
456 		thread = machine__find_thread(btsq->bts->machine, -1,
457 					      btsq->tid);
458 		if (thread)
459 			btsq->pid = thread->pid_;
460 	} else {
461 		thread = machine__findnew_thread(btsq->bts->machine, btsq->pid,
462 						 btsq->tid);
463 	}
464 
465 	queue = &btsq->bts->queues.queue_array[btsq->queue_nr];
466 
467 	if (!buffer)
468 		buffer = auxtrace_buffer__next(queue, NULL);
469 
470 	if (!buffer) {
471 		if (!btsq->bts->sampling_mode)
472 			btsq->done = 1;
473 		err = 1;
474 		goto out_put;
475 	}
476 
477 	/* Currently there is no support for split buffers */
478 	if (buffer->consecutive) {
479 		err = -EINVAL;
480 		goto out_put;
481 	}
482 
483 	if (!buffer->data) {
484 		int fd = perf_data__fd(btsq->bts->session->data);
485 
486 		buffer->data = auxtrace_buffer__get_data(buffer, fd);
487 		if (!buffer->data) {
488 			err = -ENOMEM;
489 			goto out_put;
490 		}
491 	}
492 
493 	if (btsq->bts->snapshot_mode && !buffer->consecutive &&
494 	    intel_bts_do_fix_overlap(queue, buffer)) {
495 		err = -ENOMEM;
496 		goto out_put;
497 	}
498 
499 	if (!btsq->bts->synth_opts.callchain &&
500 	    !btsq->bts->synth_opts.thread_stack && thread &&
501 	    (!old_buffer || btsq->bts->sampling_mode ||
502 	     (btsq->bts->snapshot_mode && !buffer->consecutive)))
503 		thread_stack__set_trace_nr(thread, btsq->cpu, buffer->buffer_nr + 1);
504 
505 	err = intel_bts_process_buffer(btsq, buffer, thread);
506 
507 	auxtrace_buffer__drop_data(buffer);
508 
509 	btsq->buffer = auxtrace_buffer__next(queue, buffer);
510 	if (btsq->buffer) {
511 		if (timestamp)
512 			*timestamp = btsq->buffer->reference;
513 	} else {
514 		if (!btsq->bts->sampling_mode)
515 			btsq->done = 1;
516 	}
517 out_put:
518 	thread__put(thread);
519 	return err;
520 }
521 
522 static int intel_bts_flush_queue(struct intel_bts_queue *btsq)
523 {
524 	u64 ts = 0;
525 	int ret;
526 
527 	while (1) {
528 		ret = intel_bts_process_queue(btsq, &ts);
529 		if (ret < 0)
530 			return ret;
531 		if (ret)
532 			break;
533 	}
534 	return 0;
535 }
536 
537 static int intel_bts_process_tid_exit(struct intel_bts *bts, pid_t tid)
538 {
539 	struct auxtrace_queues *queues = &bts->queues;
540 	unsigned int i;
541 
542 	for (i = 0; i < queues->nr_queues; i++) {
543 		struct auxtrace_queue *queue = &bts->queues.queue_array[i];
544 		struct intel_bts_queue *btsq = queue->priv;
545 
546 		if (btsq && btsq->tid == tid)
547 			return intel_bts_flush_queue(btsq);
548 	}
549 	return 0;
550 }
551 
552 static int intel_bts_process_queues(struct intel_bts *bts, u64 timestamp)
553 {
554 	while (1) {
555 		unsigned int queue_nr;
556 		struct auxtrace_queue *queue;
557 		struct intel_bts_queue *btsq;
558 		u64 ts = 0;
559 		int ret;
560 
561 		if (!bts->heap.heap_cnt)
562 			return 0;
563 
564 		if (bts->heap.heap_array[0].ordinal > timestamp)
565 			return 0;
566 
567 		queue_nr = bts->heap.heap_array[0].queue_nr;
568 		queue = &bts->queues.queue_array[queue_nr];
569 		btsq = queue->priv;
570 
571 		auxtrace_heap__pop(&bts->heap);
572 
573 		ret = intel_bts_process_queue(btsq, &ts);
574 		if (ret < 0) {
575 			auxtrace_heap__add(&bts->heap, queue_nr, ts);
576 			return ret;
577 		}
578 
579 		if (!ret) {
580 			ret = auxtrace_heap__add(&bts->heap, queue_nr, ts);
581 			if (ret < 0)
582 				return ret;
583 		} else {
584 			btsq->on_heap = false;
585 		}
586 	}
587 
588 	return 0;
589 }
590 
591 static int intel_bts_process_event(struct perf_session *session,
592 				   union perf_event *event,
593 				   struct perf_sample *sample,
594 				   struct perf_tool *tool)
595 {
596 	struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts,
597 					     auxtrace);
598 	u64 timestamp;
599 	int err;
600 
601 	if (dump_trace)
602 		return 0;
603 
604 	if (!tool->ordered_events) {
605 		pr_err("Intel BTS requires ordered events\n");
606 		return -EINVAL;
607 	}
608 
609 	if (sample->time && sample->time != (u64)-1)
610 		timestamp = perf_time_to_tsc(sample->time, &bts->tc);
611 	else
612 		timestamp = 0;
613 
614 	err = intel_bts_update_queues(bts);
615 	if (err)
616 		return err;
617 
618 	err = intel_bts_process_queues(bts, timestamp);
619 	if (err)
620 		return err;
621 	if (event->header.type == PERF_RECORD_EXIT) {
622 		err = intel_bts_process_tid_exit(bts, event->fork.tid);
623 		if (err)
624 			return err;
625 	}
626 
627 	if (event->header.type == PERF_RECORD_AUX &&
628 	    (event->aux.flags & PERF_AUX_FLAG_TRUNCATED) &&
629 	    bts->synth_opts.errors)
630 		err = intel_bts_lost(bts, sample);
631 
632 	return err;
633 }
634 
635 static int intel_bts_process_auxtrace_event(struct perf_session *session,
636 					    union perf_event *event,
637 					    struct perf_tool *tool __maybe_unused)
638 {
639 	struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts,
640 					     auxtrace);
641 
642 	if (bts->sampling_mode)
643 		return 0;
644 
645 	if (!bts->data_queued) {
646 		struct auxtrace_buffer *buffer;
647 		off_t data_offset;
648 		int fd = perf_data__fd(session->data);
649 		int err;
650 
651 		if (perf_data__is_pipe(session->data)) {
652 			data_offset = 0;
653 		} else {
654 			data_offset = lseek(fd, 0, SEEK_CUR);
655 			if (data_offset == -1)
656 				return -errno;
657 		}
658 
659 		err = auxtrace_queues__add_event(&bts->queues, session, event,
660 						 data_offset, &buffer);
661 		if (err)
662 			return err;
663 
664 		/* Dump here now we have copied a piped trace out of the pipe */
665 		if (dump_trace) {
666 			if (auxtrace_buffer__get_data(buffer, fd)) {
667 				intel_bts_dump_event(bts, buffer->data,
668 						     buffer->size);
669 				auxtrace_buffer__put_data(buffer);
670 			}
671 		}
672 	}
673 
674 	return 0;
675 }
676 
677 static int intel_bts_flush(struct perf_session *session,
678 			   struct perf_tool *tool __maybe_unused)
679 {
680 	struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts,
681 					     auxtrace);
682 	int ret;
683 
684 	if (dump_trace || bts->sampling_mode)
685 		return 0;
686 
687 	if (!tool->ordered_events)
688 		return -EINVAL;
689 
690 	ret = intel_bts_update_queues(bts);
691 	if (ret < 0)
692 		return ret;
693 
694 	return intel_bts_process_queues(bts, MAX_TIMESTAMP);
695 }
696 
697 static void intel_bts_free_queue(void *priv)
698 {
699 	struct intel_bts_queue *btsq = priv;
700 
701 	if (!btsq)
702 		return;
703 	free(btsq);
704 }
705 
706 static void intel_bts_free_events(struct perf_session *session)
707 {
708 	struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts,
709 					     auxtrace);
710 	struct auxtrace_queues *queues = &bts->queues;
711 	unsigned int i;
712 
713 	for (i = 0; i < queues->nr_queues; i++) {
714 		intel_bts_free_queue(queues->queue_array[i].priv);
715 		queues->queue_array[i].priv = NULL;
716 	}
717 	auxtrace_queues__free(queues);
718 }
719 
720 static void intel_bts_free(struct perf_session *session)
721 {
722 	struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts,
723 					     auxtrace);
724 
725 	auxtrace_heap__free(&bts->heap);
726 	intel_bts_free_events(session);
727 	session->auxtrace = NULL;
728 	free(bts);
729 }
730 
731 static bool intel_bts_evsel_is_auxtrace(struct perf_session *session,
732 					struct evsel *evsel)
733 {
734 	struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts,
735 					     auxtrace);
736 
737 	return evsel->core.attr.type == bts->pmu_type;
738 }
739 
740 struct intel_bts_synth {
741 	struct perf_tool dummy_tool;
742 	struct perf_session *session;
743 };
744 
745 static int intel_bts_event_synth(struct perf_tool *tool,
746 				 union perf_event *event,
747 				 struct perf_sample *sample __maybe_unused,
748 				 struct machine *machine __maybe_unused)
749 {
750 	struct intel_bts_synth *intel_bts_synth =
751 			container_of(tool, struct intel_bts_synth, dummy_tool);
752 
753 	return perf_session__deliver_synth_event(intel_bts_synth->session,
754 						 event, NULL);
755 }
756 
757 static int intel_bts_synth_event(struct perf_session *session,
758 				 struct perf_event_attr *attr, u64 id)
759 {
760 	struct intel_bts_synth intel_bts_synth;
761 
762 	memset(&intel_bts_synth, 0, sizeof(struct intel_bts_synth));
763 	intel_bts_synth.session = session;
764 
765 	return perf_event__synthesize_attr(&intel_bts_synth.dummy_tool, attr, 1,
766 					   &id, intel_bts_event_synth);
767 }
768 
769 static int intel_bts_synth_events(struct intel_bts *bts,
770 				  struct perf_session *session)
771 {
772 	struct evlist *evlist = session->evlist;
773 	struct evsel *evsel;
774 	struct perf_event_attr attr;
775 	bool found = false;
776 	u64 id;
777 	int err;
778 
779 	evlist__for_each_entry(evlist, evsel) {
780 		if (evsel->core.attr.type == bts->pmu_type && evsel->core.ids) {
781 			found = true;
782 			break;
783 		}
784 	}
785 
786 	if (!found) {
787 		pr_debug("There are no selected events with Intel BTS data\n");
788 		return 0;
789 	}
790 
791 	memset(&attr, 0, sizeof(struct perf_event_attr));
792 	attr.size = sizeof(struct perf_event_attr);
793 	attr.type = PERF_TYPE_HARDWARE;
794 	attr.sample_type = evsel->core.attr.sample_type & PERF_SAMPLE_MASK;
795 	attr.sample_type |= PERF_SAMPLE_IP | PERF_SAMPLE_TID |
796 			    PERF_SAMPLE_PERIOD;
797 	attr.sample_type &= ~(u64)PERF_SAMPLE_TIME;
798 	attr.sample_type &= ~(u64)PERF_SAMPLE_CPU;
799 	attr.exclude_user = evsel->core.attr.exclude_user;
800 	attr.exclude_kernel = evsel->core.attr.exclude_kernel;
801 	attr.exclude_hv = evsel->core.attr.exclude_hv;
802 	attr.exclude_host = evsel->core.attr.exclude_host;
803 	attr.exclude_guest = evsel->core.attr.exclude_guest;
804 	attr.sample_id_all = evsel->core.attr.sample_id_all;
805 	attr.read_format = evsel->core.attr.read_format;
806 
807 	id = evsel->core.id[0] + 1000000000;
808 	if (!id)
809 		id = 1;
810 
811 	if (bts->synth_opts.branches) {
812 		attr.config = PERF_COUNT_HW_BRANCH_INSTRUCTIONS;
813 		attr.sample_period = 1;
814 		attr.sample_type |= PERF_SAMPLE_ADDR;
815 		pr_debug("Synthesizing 'branches' event with id %" PRIu64 " sample type %#" PRIx64 "\n",
816 			 id, (u64)attr.sample_type);
817 		err = intel_bts_synth_event(session, &attr, id);
818 		if (err) {
819 			pr_err("%s: failed to synthesize 'branches' event type\n",
820 			       __func__);
821 			return err;
822 		}
823 		bts->sample_branches = true;
824 		bts->branches_sample_type = attr.sample_type;
825 		bts->branches_id = id;
826 		/*
827 		 * We only use sample types from PERF_SAMPLE_MASK so we can use
828 		 * __evsel__sample_size() here.
829 		 */
830 		bts->branches_event_size = sizeof(struct perf_record_sample) +
831 					   __evsel__sample_size(attr.sample_type);
832 	}
833 
834 	return 0;
835 }
836 
837 static const char * const intel_bts_info_fmts[] = {
838 	[INTEL_BTS_PMU_TYPE]		= "  PMU Type           %"PRId64"\n",
839 	[INTEL_BTS_TIME_SHIFT]		= "  Time Shift         %"PRIu64"\n",
840 	[INTEL_BTS_TIME_MULT]		= "  Time Muliplier     %"PRIu64"\n",
841 	[INTEL_BTS_TIME_ZERO]		= "  Time Zero          %"PRIu64"\n",
842 	[INTEL_BTS_CAP_USER_TIME_ZERO]	= "  Cap Time Zero      %"PRId64"\n",
843 	[INTEL_BTS_SNAPSHOT_MODE]	= "  Snapshot mode      %"PRId64"\n",
844 };
845 
846 static void intel_bts_print_info(__u64 *arr, int start, int finish)
847 {
848 	int i;
849 
850 	if (!dump_trace)
851 		return;
852 
853 	for (i = start; i <= finish; i++)
854 		fprintf(stdout, intel_bts_info_fmts[i], arr[i]);
855 }
856 
857 int intel_bts_process_auxtrace_info(union perf_event *event,
858 				    struct perf_session *session)
859 {
860 	struct perf_record_auxtrace_info *auxtrace_info = &event->auxtrace_info;
861 	size_t min_sz = sizeof(u64) * INTEL_BTS_SNAPSHOT_MODE;
862 	struct intel_bts *bts;
863 	int err;
864 
865 	if (auxtrace_info->header.size < sizeof(struct perf_record_auxtrace_info) +
866 					min_sz)
867 		return -EINVAL;
868 
869 	bts = zalloc(sizeof(struct intel_bts));
870 	if (!bts)
871 		return -ENOMEM;
872 
873 	err = auxtrace_queues__init(&bts->queues);
874 	if (err)
875 		goto err_free;
876 
877 	bts->session = session;
878 	bts->machine = &session->machines.host; /* No kvm support */
879 	bts->auxtrace_type = auxtrace_info->type;
880 	bts->pmu_type = auxtrace_info->priv[INTEL_BTS_PMU_TYPE];
881 	bts->tc.time_shift = auxtrace_info->priv[INTEL_BTS_TIME_SHIFT];
882 	bts->tc.time_mult = auxtrace_info->priv[INTEL_BTS_TIME_MULT];
883 	bts->tc.time_zero = auxtrace_info->priv[INTEL_BTS_TIME_ZERO];
884 	bts->cap_user_time_zero =
885 			auxtrace_info->priv[INTEL_BTS_CAP_USER_TIME_ZERO];
886 	bts->snapshot_mode = auxtrace_info->priv[INTEL_BTS_SNAPSHOT_MODE];
887 
888 	bts->sampling_mode = false;
889 
890 	bts->auxtrace.process_event = intel_bts_process_event;
891 	bts->auxtrace.process_auxtrace_event = intel_bts_process_auxtrace_event;
892 	bts->auxtrace.flush_events = intel_bts_flush;
893 	bts->auxtrace.free_events = intel_bts_free_events;
894 	bts->auxtrace.free = intel_bts_free;
895 	bts->auxtrace.evsel_is_auxtrace = intel_bts_evsel_is_auxtrace;
896 	session->auxtrace = &bts->auxtrace;
897 
898 	intel_bts_print_info(&auxtrace_info->priv[0], INTEL_BTS_PMU_TYPE,
899 			     INTEL_BTS_SNAPSHOT_MODE);
900 
901 	if (dump_trace)
902 		return 0;
903 
904 	if (session->itrace_synth_opts->set) {
905 		bts->synth_opts = *session->itrace_synth_opts;
906 	} else {
907 		itrace_synth_opts__set_default(&bts->synth_opts,
908 				session->itrace_synth_opts->default_no_sample);
909 		bts->synth_opts.thread_stack =
910 				session->itrace_synth_opts->thread_stack;
911 	}
912 
913 	if (bts->synth_opts.calls)
914 		bts->branches_filter |= PERF_IP_FLAG_CALL | PERF_IP_FLAG_ASYNC |
915 					PERF_IP_FLAG_TRACE_END;
916 	if (bts->synth_opts.returns)
917 		bts->branches_filter |= PERF_IP_FLAG_RETURN |
918 					PERF_IP_FLAG_TRACE_BEGIN;
919 
920 	err = intel_bts_synth_events(bts, session);
921 	if (err)
922 		goto err_free_queues;
923 
924 	err = auxtrace_queues__process_index(&bts->queues, session);
925 	if (err)
926 		goto err_free_queues;
927 
928 	if (bts->queues.populated)
929 		bts->data_queued = true;
930 
931 	return 0;
932 
933 err_free_queues:
934 	auxtrace_queues__free(&bts->queues);
935 	session->auxtrace = NULL;
936 err_free:
937 	free(bts);
938 	return err;
939 }
940