1 /*
2 * intel-bts.c: Intel Processor Trace support
3 * Copyright (c) 2013-2015, Intel Corporation.
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 *
14 */
15
16 #include <endian.h>
17 #include <byteswap.h>
18 #include <linux/kernel.h>
19 #include <linux/types.h>
20 #include <linux/bitops.h>
21 #include <linux/log2.h>
22
23 #include "cpumap.h"
24 #include "color.h"
25 #include "evsel.h"
26 #include "evlist.h"
27 #include "machine.h"
28 #include "session.h"
29 #include "util.h"
30 #include "thread.h"
31 #include "thread-stack.h"
32 #include "debug.h"
33 #include "tsc.h"
34 #include "auxtrace.h"
35 #include "intel-pt-decoder/intel-pt-insn-decoder.h"
36 #include "intel-bts.h"
37
38 #define MAX_TIMESTAMP (~0ULL)
39
40 #define INTEL_BTS_ERR_NOINSN 5
41 #define INTEL_BTS_ERR_LOST 9
42
43 #if __BYTE_ORDER == __BIG_ENDIAN
44 #define le64_to_cpu bswap_64
45 #else
46 #define le64_to_cpu
47 #endif
48
49 struct intel_bts {
50 struct auxtrace auxtrace;
51 struct auxtrace_queues queues;
52 struct auxtrace_heap heap;
53 u32 auxtrace_type;
54 struct perf_session *session;
55 struct machine *machine;
56 bool sampling_mode;
57 bool snapshot_mode;
58 bool data_queued;
59 u32 pmu_type;
60 struct perf_tsc_conversion tc;
61 bool cap_user_time_zero;
62 struct itrace_synth_opts synth_opts;
63 bool sample_branches;
64 u32 branches_filter;
65 u64 branches_sample_type;
66 u64 branches_id;
67 size_t branches_event_size;
68 bool synth_needs_swap;
69 unsigned long num_events;
70 };
71
72 struct intel_bts_queue {
73 struct intel_bts *bts;
74 unsigned int queue_nr;
75 struct auxtrace_buffer *buffer;
76 bool on_heap;
77 bool done;
78 pid_t pid;
79 pid_t tid;
80 int cpu;
81 u64 time;
82 struct intel_pt_insn intel_pt_insn;
83 u32 sample_flags;
84 };
85
86 struct branch {
87 u64 from;
88 u64 to;
89 u64 misc;
90 };
91
intel_bts_dump(struct intel_bts * bts __maybe_unused,unsigned char * buf,size_t len)92 static void intel_bts_dump(struct intel_bts *bts __maybe_unused,
93 unsigned char *buf, size_t len)
94 {
95 struct branch *branch;
96 size_t i, pos = 0, br_sz = sizeof(struct branch), sz;
97 const char *color = PERF_COLOR_BLUE;
98
99 color_fprintf(stdout, color,
100 ". ... Intel BTS data: size %zu bytes\n",
101 len);
102
103 while (len) {
104 if (len >= br_sz)
105 sz = br_sz;
106 else
107 sz = len;
108 printf(".");
109 color_fprintf(stdout, color, " %08x: ", pos);
110 for (i = 0; i < sz; i++)
111 color_fprintf(stdout, color, " %02x", buf[i]);
112 for (; i < br_sz; i++)
113 color_fprintf(stdout, color, " ");
114 if (len >= br_sz) {
115 branch = (struct branch *)buf;
116 color_fprintf(stdout, color, " %"PRIx64" -> %"PRIx64" %s\n",
117 le64_to_cpu(branch->from),
118 le64_to_cpu(branch->to),
119 le64_to_cpu(branch->misc) & 0x10 ?
120 "pred" : "miss");
121 } else {
122 color_fprintf(stdout, color, " Bad record!\n");
123 }
124 pos += sz;
125 buf += sz;
126 len -= sz;
127 }
128 }
129
intel_bts_dump_event(struct intel_bts * bts,unsigned char * buf,size_t len)130 static void intel_bts_dump_event(struct intel_bts *bts, unsigned char *buf,
131 size_t len)
132 {
133 printf(".\n");
134 intel_bts_dump(bts, buf, len);
135 }
136
intel_bts_lost(struct intel_bts * bts,struct perf_sample * sample)137 static int intel_bts_lost(struct intel_bts *bts, struct perf_sample *sample)
138 {
139 union perf_event event;
140 int err;
141
142 auxtrace_synth_error(&event.auxtrace_error, PERF_AUXTRACE_ERROR_ITRACE,
143 INTEL_BTS_ERR_LOST, sample->cpu, sample->pid,
144 sample->tid, 0, "Lost trace data");
145
146 err = perf_session__deliver_synth_event(bts->session, &event, NULL);
147 if (err)
148 pr_err("Intel BTS: failed to deliver error event, error %d\n",
149 err);
150
151 return err;
152 }
153
intel_bts_alloc_queue(struct intel_bts * bts,unsigned int queue_nr)154 static struct intel_bts_queue *intel_bts_alloc_queue(struct intel_bts *bts,
155 unsigned int queue_nr)
156 {
157 struct intel_bts_queue *btsq;
158
159 btsq = zalloc(sizeof(struct intel_bts_queue));
160 if (!btsq)
161 return NULL;
162
163 btsq->bts = bts;
164 btsq->queue_nr = queue_nr;
165 btsq->pid = -1;
166 btsq->tid = -1;
167 btsq->cpu = -1;
168
169 return btsq;
170 }
171
intel_bts_setup_queue(struct intel_bts * bts,struct auxtrace_queue * queue,unsigned int queue_nr)172 static int intel_bts_setup_queue(struct intel_bts *bts,
173 struct auxtrace_queue *queue,
174 unsigned int queue_nr)
175 {
176 struct intel_bts_queue *btsq = queue->priv;
177
178 if (list_empty(&queue->head))
179 return 0;
180
181 if (!btsq) {
182 btsq = intel_bts_alloc_queue(bts, queue_nr);
183 if (!btsq)
184 return -ENOMEM;
185 queue->priv = btsq;
186
187 if (queue->cpu != -1)
188 btsq->cpu = queue->cpu;
189 btsq->tid = queue->tid;
190 }
191
192 if (bts->sampling_mode)
193 return 0;
194
195 if (!btsq->on_heap && !btsq->buffer) {
196 int ret;
197
198 btsq->buffer = auxtrace_buffer__next(queue, NULL);
199 if (!btsq->buffer)
200 return 0;
201
202 ret = auxtrace_heap__add(&bts->heap, queue_nr,
203 btsq->buffer->reference);
204 if (ret)
205 return ret;
206 btsq->on_heap = true;
207 }
208
209 return 0;
210 }
211
intel_bts_setup_queues(struct intel_bts * bts)212 static int intel_bts_setup_queues(struct intel_bts *bts)
213 {
214 unsigned int i;
215 int ret;
216
217 for (i = 0; i < bts->queues.nr_queues; i++) {
218 ret = intel_bts_setup_queue(bts, &bts->queues.queue_array[i],
219 i);
220 if (ret)
221 return ret;
222 }
223 return 0;
224 }
225
intel_bts_update_queues(struct intel_bts * bts)226 static inline int intel_bts_update_queues(struct intel_bts *bts)
227 {
228 if (bts->queues.new_data) {
229 bts->queues.new_data = false;
230 return intel_bts_setup_queues(bts);
231 }
232 return 0;
233 }
234
intel_bts_find_overlap(unsigned char * buf_a,size_t len_a,unsigned char * buf_b,size_t len_b)235 static unsigned char *intel_bts_find_overlap(unsigned char *buf_a, size_t len_a,
236 unsigned char *buf_b, size_t len_b)
237 {
238 size_t offs, len;
239
240 if (len_a > len_b)
241 offs = len_a - len_b;
242 else
243 offs = 0;
244
245 for (; offs < len_a; offs += sizeof(struct branch)) {
246 len = len_a - offs;
247 if (!memcmp(buf_a + offs, buf_b, len))
248 return buf_b + len;
249 }
250
251 return buf_b;
252 }
253
intel_bts_do_fix_overlap(struct auxtrace_queue * queue,struct auxtrace_buffer * b)254 static int intel_bts_do_fix_overlap(struct auxtrace_queue *queue,
255 struct auxtrace_buffer *b)
256 {
257 struct auxtrace_buffer *a;
258 void *start;
259
260 if (b->list.prev == &queue->head)
261 return 0;
262 a = list_entry(b->list.prev, struct auxtrace_buffer, list);
263 start = intel_bts_find_overlap(a->data, a->size, b->data, b->size);
264 if (!start)
265 return -EINVAL;
266 b->use_size = b->data + b->size - start;
267 b->use_data = start;
268 return 0;
269 }
270
intel_bts_synth_branch_sample(struct intel_bts_queue * btsq,struct branch * branch)271 static int intel_bts_synth_branch_sample(struct intel_bts_queue *btsq,
272 struct branch *branch)
273 {
274 int ret;
275 struct intel_bts *bts = btsq->bts;
276 union perf_event event;
277 struct perf_sample sample = { .ip = 0, };
278
279 if (bts->synth_opts.initial_skip &&
280 bts->num_events++ <= bts->synth_opts.initial_skip)
281 return 0;
282
283 event.sample.header.type = PERF_RECORD_SAMPLE;
284 event.sample.header.misc = PERF_RECORD_MISC_USER;
285 event.sample.header.size = sizeof(struct perf_event_header);
286
287 sample.cpumode = PERF_RECORD_MISC_USER;
288 sample.ip = le64_to_cpu(branch->from);
289 sample.pid = btsq->pid;
290 sample.tid = btsq->tid;
291 sample.addr = le64_to_cpu(branch->to);
292 sample.id = btsq->bts->branches_id;
293 sample.stream_id = btsq->bts->branches_id;
294 sample.period = 1;
295 sample.cpu = btsq->cpu;
296 sample.flags = btsq->sample_flags;
297 sample.insn_len = btsq->intel_pt_insn.length;
298
299 if (bts->synth_opts.inject) {
300 event.sample.header.size = bts->branches_event_size;
301 ret = perf_event__synthesize_sample(&event,
302 bts->branches_sample_type,
303 0, &sample,
304 bts->synth_needs_swap);
305 if (ret)
306 return ret;
307 }
308
309 ret = perf_session__deliver_synth_event(bts->session, &event, &sample);
310 if (ret)
311 pr_err("Intel BTS: failed to deliver branch event, error %d\n",
312 ret);
313
314 return ret;
315 }
316
intel_bts_get_next_insn(struct intel_bts_queue * btsq,u64 ip)317 static int intel_bts_get_next_insn(struct intel_bts_queue *btsq, u64 ip)
318 {
319 struct machine *machine = btsq->bts->machine;
320 struct thread *thread;
321 struct addr_location al;
322 unsigned char buf[1024];
323 size_t bufsz;
324 ssize_t len;
325 int x86_64;
326 uint8_t cpumode;
327 int err = -1;
328
329 bufsz = intel_pt_insn_max_size();
330
331 if (machine__kernel_ip(machine, ip))
332 cpumode = PERF_RECORD_MISC_KERNEL;
333 else
334 cpumode = PERF_RECORD_MISC_USER;
335
336 thread = machine__find_thread(machine, -1, btsq->tid);
337 if (!thread)
338 return -1;
339
340 thread__find_addr_map(thread, cpumode, MAP__FUNCTION, ip, &al);
341 if (!al.map || !al.map->dso)
342 goto out_put;
343
344 len = dso__data_read_addr(al.map->dso, al.map, machine, ip, buf, bufsz);
345 if (len <= 0)
346 goto out_put;
347
348 /* Load maps to ensure dso->is_64_bit has been updated */
349 map__load(al.map);
350
351 x86_64 = al.map->dso->is_64_bit;
352
353 if (intel_pt_get_insn(buf, len, x86_64, &btsq->intel_pt_insn))
354 goto out_put;
355
356 err = 0;
357 out_put:
358 thread__put(thread);
359 return err;
360 }
361
intel_bts_synth_error(struct intel_bts * bts,int cpu,pid_t pid,pid_t tid,u64 ip)362 static int intel_bts_synth_error(struct intel_bts *bts, int cpu, pid_t pid,
363 pid_t tid, u64 ip)
364 {
365 union perf_event event;
366 int err;
367
368 auxtrace_synth_error(&event.auxtrace_error, PERF_AUXTRACE_ERROR_ITRACE,
369 INTEL_BTS_ERR_NOINSN, cpu, pid, tid, ip,
370 "Failed to get instruction");
371
372 err = perf_session__deliver_synth_event(bts->session, &event, NULL);
373 if (err)
374 pr_err("Intel BTS: failed to deliver error event, error %d\n",
375 err);
376
377 return err;
378 }
379
intel_bts_get_branch_type(struct intel_bts_queue * btsq,struct branch * branch)380 static int intel_bts_get_branch_type(struct intel_bts_queue *btsq,
381 struct branch *branch)
382 {
383 int err;
384
385 if (!branch->from) {
386 if (branch->to)
387 btsq->sample_flags = PERF_IP_FLAG_BRANCH |
388 PERF_IP_FLAG_TRACE_BEGIN;
389 else
390 btsq->sample_flags = 0;
391 btsq->intel_pt_insn.length = 0;
392 } else if (!branch->to) {
393 btsq->sample_flags = PERF_IP_FLAG_BRANCH |
394 PERF_IP_FLAG_TRACE_END;
395 btsq->intel_pt_insn.length = 0;
396 } else {
397 err = intel_bts_get_next_insn(btsq, branch->from);
398 if (err) {
399 btsq->sample_flags = 0;
400 btsq->intel_pt_insn.length = 0;
401 if (!btsq->bts->synth_opts.errors)
402 return 0;
403 err = intel_bts_synth_error(btsq->bts, btsq->cpu,
404 btsq->pid, btsq->tid,
405 branch->from);
406 return err;
407 }
408 btsq->sample_flags = intel_pt_insn_type(btsq->intel_pt_insn.op);
409 /* Check for an async branch into the kernel */
410 if (!machine__kernel_ip(btsq->bts->machine, branch->from) &&
411 machine__kernel_ip(btsq->bts->machine, branch->to) &&
412 btsq->sample_flags != (PERF_IP_FLAG_BRANCH |
413 PERF_IP_FLAG_CALL |
414 PERF_IP_FLAG_SYSCALLRET))
415 btsq->sample_flags = PERF_IP_FLAG_BRANCH |
416 PERF_IP_FLAG_CALL |
417 PERF_IP_FLAG_ASYNC |
418 PERF_IP_FLAG_INTERRUPT;
419 }
420
421 return 0;
422 }
423
intel_bts_process_buffer(struct intel_bts_queue * btsq,struct auxtrace_buffer * buffer,struct thread * thread)424 static int intel_bts_process_buffer(struct intel_bts_queue *btsq,
425 struct auxtrace_buffer *buffer,
426 struct thread *thread)
427 {
428 struct branch *branch;
429 size_t sz, bsz = sizeof(struct branch);
430 u32 filter = btsq->bts->branches_filter;
431 int err = 0;
432
433 if (buffer->use_data) {
434 sz = buffer->use_size;
435 branch = buffer->use_data;
436 } else {
437 sz = buffer->size;
438 branch = buffer->data;
439 }
440
441 if (!btsq->bts->sample_branches)
442 return 0;
443
444 for (; sz > bsz; branch += 1, sz -= bsz) {
445 if (!branch->from && !branch->to)
446 continue;
447 intel_bts_get_branch_type(btsq, branch);
448 if (btsq->bts->synth_opts.thread_stack)
449 thread_stack__event(thread, btsq->sample_flags,
450 le64_to_cpu(branch->from),
451 le64_to_cpu(branch->to),
452 btsq->intel_pt_insn.length,
453 buffer->buffer_nr + 1);
454 if (filter && !(filter & btsq->sample_flags))
455 continue;
456 err = intel_bts_synth_branch_sample(btsq, branch);
457 if (err)
458 break;
459 }
460 return err;
461 }
462
intel_bts_process_queue(struct intel_bts_queue * btsq,u64 * timestamp)463 static int intel_bts_process_queue(struct intel_bts_queue *btsq, u64 *timestamp)
464 {
465 struct auxtrace_buffer *buffer = btsq->buffer, *old_buffer = buffer;
466 struct auxtrace_queue *queue;
467 struct thread *thread;
468 int err;
469
470 if (btsq->done)
471 return 1;
472
473 if (btsq->pid == -1) {
474 thread = machine__find_thread(btsq->bts->machine, -1,
475 btsq->tid);
476 if (thread)
477 btsq->pid = thread->pid_;
478 } else {
479 thread = machine__findnew_thread(btsq->bts->machine, btsq->pid,
480 btsq->tid);
481 }
482
483 queue = &btsq->bts->queues.queue_array[btsq->queue_nr];
484
485 if (!buffer)
486 buffer = auxtrace_buffer__next(queue, NULL);
487
488 if (!buffer) {
489 if (!btsq->bts->sampling_mode)
490 btsq->done = 1;
491 err = 1;
492 goto out_put;
493 }
494
495 /* Currently there is no support for split buffers */
496 if (buffer->consecutive) {
497 err = -EINVAL;
498 goto out_put;
499 }
500
501 if (!buffer->data) {
502 int fd = perf_data_file__fd(btsq->bts->session->file);
503
504 buffer->data = auxtrace_buffer__get_data(buffer, fd);
505 if (!buffer->data) {
506 err = -ENOMEM;
507 goto out_put;
508 }
509 }
510
511 if (btsq->bts->snapshot_mode && !buffer->consecutive &&
512 intel_bts_do_fix_overlap(queue, buffer)) {
513 err = -ENOMEM;
514 goto out_put;
515 }
516
517 if (!btsq->bts->synth_opts.callchain &&
518 !btsq->bts->synth_opts.thread_stack && thread &&
519 (!old_buffer || btsq->bts->sampling_mode ||
520 (btsq->bts->snapshot_mode && !buffer->consecutive)))
521 thread_stack__set_trace_nr(thread, buffer->buffer_nr + 1);
522
523 err = intel_bts_process_buffer(btsq, buffer, thread);
524
525 auxtrace_buffer__drop_data(buffer);
526
527 btsq->buffer = auxtrace_buffer__next(queue, buffer);
528 if (btsq->buffer) {
529 if (timestamp)
530 *timestamp = btsq->buffer->reference;
531 } else {
532 if (!btsq->bts->sampling_mode)
533 btsq->done = 1;
534 }
535 out_put:
536 thread__put(thread);
537 return err;
538 }
539
intel_bts_flush_queue(struct intel_bts_queue * btsq)540 static int intel_bts_flush_queue(struct intel_bts_queue *btsq)
541 {
542 u64 ts = 0;
543 int ret;
544
545 while (1) {
546 ret = intel_bts_process_queue(btsq, &ts);
547 if (ret < 0)
548 return ret;
549 if (ret)
550 break;
551 }
552 return 0;
553 }
554
intel_bts_process_tid_exit(struct intel_bts * bts,pid_t tid)555 static int intel_bts_process_tid_exit(struct intel_bts *bts, pid_t tid)
556 {
557 struct auxtrace_queues *queues = &bts->queues;
558 unsigned int i;
559
560 for (i = 0; i < queues->nr_queues; i++) {
561 struct auxtrace_queue *queue = &bts->queues.queue_array[i];
562 struct intel_bts_queue *btsq = queue->priv;
563
564 if (btsq && btsq->tid == tid)
565 return intel_bts_flush_queue(btsq);
566 }
567 return 0;
568 }
569
intel_bts_process_queues(struct intel_bts * bts,u64 timestamp)570 static int intel_bts_process_queues(struct intel_bts *bts, u64 timestamp)
571 {
572 while (1) {
573 unsigned int queue_nr;
574 struct auxtrace_queue *queue;
575 struct intel_bts_queue *btsq;
576 u64 ts = 0;
577 int ret;
578
579 if (!bts->heap.heap_cnt)
580 return 0;
581
582 if (bts->heap.heap_array[0].ordinal > timestamp)
583 return 0;
584
585 queue_nr = bts->heap.heap_array[0].queue_nr;
586 queue = &bts->queues.queue_array[queue_nr];
587 btsq = queue->priv;
588
589 auxtrace_heap__pop(&bts->heap);
590
591 ret = intel_bts_process_queue(btsq, &ts);
592 if (ret < 0) {
593 auxtrace_heap__add(&bts->heap, queue_nr, ts);
594 return ret;
595 }
596
597 if (!ret) {
598 ret = auxtrace_heap__add(&bts->heap, queue_nr, ts);
599 if (ret < 0)
600 return ret;
601 } else {
602 btsq->on_heap = false;
603 }
604 }
605
606 return 0;
607 }
608
intel_bts_process_event(struct perf_session * session,union perf_event * event,struct perf_sample * sample,struct perf_tool * tool)609 static int intel_bts_process_event(struct perf_session *session,
610 union perf_event *event,
611 struct perf_sample *sample,
612 struct perf_tool *tool)
613 {
614 struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts,
615 auxtrace);
616 u64 timestamp;
617 int err;
618
619 if (dump_trace)
620 return 0;
621
622 if (!tool->ordered_events) {
623 pr_err("Intel BTS requires ordered events\n");
624 return -EINVAL;
625 }
626
627 if (sample->time && sample->time != (u64)-1)
628 timestamp = perf_time_to_tsc(sample->time, &bts->tc);
629 else
630 timestamp = 0;
631
632 err = intel_bts_update_queues(bts);
633 if (err)
634 return err;
635
636 err = intel_bts_process_queues(bts, timestamp);
637 if (err)
638 return err;
639 if (event->header.type == PERF_RECORD_EXIT) {
640 err = intel_bts_process_tid_exit(bts, event->fork.tid);
641 if (err)
642 return err;
643 }
644
645 if (event->header.type == PERF_RECORD_AUX &&
646 (event->aux.flags & PERF_AUX_FLAG_TRUNCATED) &&
647 bts->synth_opts.errors)
648 err = intel_bts_lost(bts, sample);
649
650 return err;
651 }
652
intel_bts_process_auxtrace_event(struct perf_session * session,union perf_event * event,struct perf_tool * tool __maybe_unused)653 static int intel_bts_process_auxtrace_event(struct perf_session *session,
654 union perf_event *event,
655 struct perf_tool *tool __maybe_unused)
656 {
657 struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts,
658 auxtrace);
659
660 if (bts->sampling_mode)
661 return 0;
662
663 if (!bts->data_queued) {
664 struct auxtrace_buffer *buffer;
665 off_t data_offset;
666 int fd = perf_data_file__fd(session->file);
667 int err;
668
669 if (perf_data_file__is_pipe(session->file)) {
670 data_offset = 0;
671 } else {
672 data_offset = lseek(fd, 0, SEEK_CUR);
673 if (data_offset == -1)
674 return -errno;
675 }
676
677 err = auxtrace_queues__add_event(&bts->queues, session, event,
678 data_offset, &buffer);
679 if (err)
680 return err;
681
682 /* Dump here now we have copied a piped trace out of the pipe */
683 if (dump_trace) {
684 if (auxtrace_buffer__get_data(buffer, fd)) {
685 intel_bts_dump_event(bts, buffer->data,
686 buffer->size);
687 auxtrace_buffer__put_data(buffer);
688 }
689 }
690 }
691
692 return 0;
693 }
694
intel_bts_flush(struct perf_session * session,struct perf_tool * tool __maybe_unused)695 static int intel_bts_flush(struct perf_session *session,
696 struct perf_tool *tool __maybe_unused)
697 {
698 struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts,
699 auxtrace);
700 int ret;
701
702 if (dump_trace || bts->sampling_mode)
703 return 0;
704
705 if (!tool->ordered_events)
706 return -EINVAL;
707
708 ret = intel_bts_update_queues(bts);
709 if (ret < 0)
710 return ret;
711
712 return intel_bts_process_queues(bts, MAX_TIMESTAMP);
713 }
714
intel_bts_free_queue(void * priv)715 static void intel_bts_free_queue(void *priv)
716 {
717 struct intel_bts_queue *btsq = priv;
718
719 if (!btsq)
720 return;
721 free(btsq);
722 }
723
intel_bts_free_events(struct perf_session * session)724 static void intel_bts_free_events(struct perf_session *session)
725 {
726 struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts,
727 auxtrace);
728 struct auxtrace_queues *queues = &bts->queues;
729 unsigned int i;
730
731 for (i = 0; i < queues->nr_queues; i++) {
732 intel_bts_free_queue(queues->queue_array[i].priv);
733 queues->queue_array[i].priv = NULL;
734 }
735 auxtrace_queues__free(queues);
736 }
737
intel_bts_free(struct perf_session * session)738 static void intel_bts_free(struct perf_session *session)
739 {
740 struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts,
741 auxtrace);
742
743 auxtrace_heap__free(&bts->heap);
744 intel_bts_free_events(session);
745 session->auxtrace = NULL;
746 free(bts);
747 }
748
749 struct intel_bts_synth {
750 struct perf_tool dummy_tool;
751 struct perf_session *session;
752 };
753
intel_bts_event_synth(struct perf_tool * tool,union perf_event * event,struct perf_sample * sample __maybe_unused,struct machine * machine __maybe_unused)754 static int intel_bts_event_synth(struct perf_tool *tool,
755 union perf_event *event,
756 struct perf_sample *sample __maybe_unused,
757 struct machine *machine __maybe_unused)
758 {
759 struct intel_bts_synth *intel_bts_synth =
760 container_of(tool, struct intel_bts_synth, dummy_tool);
761
762 return perf_session__deliver_synth_event(intel_bts_synth->session,
763 event, NULL);
764 }
765
intel_bts_synth_event(struct perf_session * session,struct perf_event_attr * attr,u64 id)766 static int intel_bts_synth_event(struct perf_session *session,
767 struct perf_event_attr *attr, u64 id)
768 {
769 struct intel_bts_synth intel_bts_synth;
770
771 memset(&intel_bts_synth, 0, sizeof(struct intel_bts_synth));
772 intel_bts_synth.session = session;
773
774 return perf_event__synthesize_attr(&intel_bts_synth.dummy_tool, attr, 1,
775 &id, intel_bts_event_synth);
776 }
777
intel_bts_synth_events(struct intel_bts * bts,struct perf_session * session)778 static int intel_bts_synth_events(struct intel_bts *bts,
779 struct perf_session *session)
780 {
781 struct perf_evlist *evlist = session->evlist;
782 struct perf_evsel *evsel;
783 struct perf_event_attr attr;
784 bool found = false;
785 u64 id;
786 int err;
787
788 evlist__for_each_entry(evlist, evsel) {
789 if (evsel->attr.type == bts->pmu_type && evsel->ids) {
790 found = true;
791 break;
792 }
793 }
794
795 if (!found) {
796 pr_debug("There are no selected events with Intel BTS data\n");
797 return 0;
798 }
799
800 memset(&attr, 0, sizeof(struct perf_event_attr));
801 attr.size = sizeof(struct perf_event_attr);
802 attr.type = PERF_TYPE_HARDWARE;
803 attr.sample_type = evsel->attr.sample_type & PERF_SAMPLE_MASK;
804 attr.sample_type |= PERF_SAMPLE_IP | PERF_SAMPLE_TID |
805 PERF_SAMPLE_PERIOD;
806 attr.sample_type &= ~(u64)PERF_SAMPLE_TIME;
807 attr.sample_type &= ~(u64)PERF_SAMPLE_CPU;
808 attr.exclude_user = evsel->attr.exclude_user;
809 attr.exclude_kernel = evsel->attr.exclude_kernel;
810 attr.exclude_hv = evsel->attr.exclude_hv;
811 attr.exclude_host = evsel->attr.exclude_host;
812 attr.exclude_guest = evsel->attr.exclude_guest;
813 attr.sample_id_all = evsel->attr.sample_id_all;
814 attr.read_format = evsel->attr.read_format;
815
816 id = evsel->id[0] + 1000000000;
817 if (!id)
818 id = 1;
819
820 if (bts->synth_opts.branches) {
821 attr.config = PERF_COUNT_HW_BRANCH_INSTRUCTIONS;
822 attr.sample_period = 1;
823 attr.sample_type |= PERF_SAMPLE_ADDR;
824 pr_debug("Synthesizing 'branches' event with id %" PRIu64 " sample type %#" PRIx64 "\n",
825 id, (u64)attr.sample_type);
826 err = intel_bts_synth_event(session, &attr, id);
827 if (err) {
828 pr_err("%s: failed to synthesize 'branches' event type\n",
829 __func__);
830 return err;
831 }
832 bts->sample_branches = true;
833 bts->branches_sample_type = attr.sample_type;
834 bts->branches_id = id;
835 /*
836 * We only use sample types from PERF_SAMPLE_MASK so we can use
837 * __perf_evsel__sample_size() here.
838 */
839 bts->branches_event_size = sizeof(struct sample_event) +
840 __perf_evsel__sample_size(attr.sample_type);
841 }
842
843 bts->synth_needs_swap = evsel->needs_swap;
844
845 return 0;
846 }
847
848 static const char * const intel_bts_info_fmts[] = {
849 [INTEL_BTS_PMU_TYPE] = " PMU Type %"PRId64"\n",
850 [INTEL_BTS_TIME_SHIFT] = " Time Shift %"PRIu64"\n",
851 [INTEL_BTS_TIME_MULT] = " Time Muliplier %"PRIu64"\n",
852 [INTEL_BTS_TIME_ZERO] = " Time Zero %"PRIu64"\n",
853 [INTEL_BTS_CAP_USER_TIME_ZERO] = " Cap Time Zero %"PRId64"\n",
854 [INTEL_BTS_SNAPSHOT_MODE] = " Snapshot mode %"PRId64"\n",
855 };
856
intel_bts_print_info(u64 * arr,int start,int finish)857 static void intel_bts_print_info(u64 *arr, int start, int finish)
858 {
859 int i;
860
861 if (!dump_trace)
862 return;
863
864 for (i = start; i <= finish; i++)
865 fprintf(stdout, intel_bts_info_fmts[i], arr[i]);
866 }
867
868 u64 intel_bts_auxtrace_info_priv[INTEL_BTS_AUXTRACE_PRIV_SIZE];
869
intel_bts_process_auxtrace_info(union perf_event * event,struct perf_session * session)870 int intel_bts_process_auxtrace_info(union perf_event *event,
871 struct perf_session *session)
872 {
873 struct auxtrace_info_event *auxtrace_info = &event->auxtrace_info;
874 size_t min_sz = sizeof(u64) * INTEL_BTS_SNAPSHOT_MODE;
875 struct intel_bts *bts;
876 int err;
877
878 if (auxtrace_info->header.size < sizeof(struct auxtrace_info_event) +
879 min_sz)
880 return -EINVAL;
881
882 bts = zalloc(sizeof(struct intel_bts));
883 if (!bts)
884 return -ENOMEM;
885
886 err = auxtrace_queues__init(&bts->queues);
887 if (err)
888 goto err_free;
889
890 bts->session = session;
891 bts->machine = &session->machines.host; /* No kvm support */
892 bts->auxtrace_type = auxtrace_info->type;
893 bts->pmu_type = auxtrace_info->priv[INTEL_BTS_PMU_TYPE];
894 bts->tc.time_shift = auxtrace_info->priv[INTEL_BTS_TIME_SHIFT];
895 bts->tc.time_mult = auxtrace_info->priv[INTEL_BTS_TIME_MULT];
896 bts->tc.time_zero = auxtrace_info->priv[INTEL_BTS_TIME_ZERO];
897 bts->cap_user_time_zero =
898 auxtrace_info->priv[INTEL_BTS_CAP_USER_TIME_ZERO];
899 bts->snapshot_mode = auxtrace_info->priv[INTEL_BTS_SNAPSHOT_MODE];
900
901 bts->sampling_mode = false;
902
903 bts->auxtrace.process_event = intel_bts_process_event;
904 bts->auxtrace.process_auxtrace_event = intel_bts_process_auxtrace_event;
905 bts->auxtrace.flush_events = intel_bts_flush;
906 bts->auxtrace.free_events = intel_bts_free_events;
907 bts->auxtrace.free = intel_bts_free;
908 session->auxtrace = &bts->auxtrace;
909
910 intel_bts_print_info(&auxtrace_info->priv[0], INTEL_BTS_PMU_TYPE,
911 INTEL_BTS_SNAPSHOT_MODE);
912
913 if (dump_trace)
914 return 0;
915
916 if (session->itrace_synth_opts && session->itrace_synth_opts->set) {
917 bts->synth_opts = *session->itrace_synth_opts;
918 } else {
919 itrace_synth_opts__set_default(&bts->synth_opts);
920 if (session->itrace_synth_opts)
921 bts->synth_opts.thread_stack =
922 session->itrace_synth_opts->thread_stack;
923 }
924
925 if (bts->synth_opts.calls)
926 bts->branches_filter |= PERF_IP_FLAG_CALL | PERF_IP_FLAG_ASYNC |
927 PERF_IP_FLAG_TRACE_END;
928 if (bts->synth_opts.returns)
929 bts->branches_filter |= PERF_IP_FLAG_RETURN |
930 PERF_IP_FLAG_TRACE_BEGIN;
931
932 err = intel_bts_synth_events(bts, session);
933 if (err)
934 goto err_free_queues;
935
936 err = auxtrace_queues__process_index(&bts->queues, session);
937 if (err)
938 goto err_free_queues;
939
940 if (bts->queues.populated)
941 bts->data_queued = true;
942
943 return 0;
944
945 err_free_queues:
946 auxtrace_queues__free(&bts->queues);
947 session->auxtrace = NULL;
948 err_free:
949 free(bts);
950 return err;
951 }
952