Commit | Line | Data |
---|---|---|
d0170af7 AH |
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; | |
d1706b39 | 69 | unsigned long num_events; |
d0170af7 AH |
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 | ||
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 | ||
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 | ||
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 | ||
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 | ||
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 | ||
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 | ||
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 | ||
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 | ||
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 | ||
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 | ||
d1706b39 AK |
279 | if (bts->synth_opts.initial_skip && |
280 | bts->num_events++ <= bts->synth_opts.initial_skip) | |
281 | return 0; | |
282 | ||
d0170af7 AH |
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 | ||
3ea223ad | 287 | sample.cpumode = PERF_RECORD_MISC_USER; |
d0170af7 AH |
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 | ||
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, machine->symbol_filter); | |
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 | ||
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 | ||
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 | ||
424 | static int intel_bts_process_buffer(struct intel_bts_queue *btsq, | |
425 | struct auxtrace_buffer *buffer) | |
426 | { | |
427 | struct branch *branch; | |
428 | size_t sz, bsz = sizeof(struct branch); | |
429 | u32 filter = btsq->bts->branches_filter; | |
430 | int err = 0; | |
431 | ||
432 | if (buffer->use_data) { | |
433 | sz = buffer->use_size; | |
434 | branch = buffer->use_data; | |
435 | } else { | |
436 | sz = buffer->size; | |
437 | branch = buffer->data; | |
438 | } | |
439 | ||
440 | if (!btsq->bts->sample_branches) | |
441 | return 0; | |
442 | ||
443 | for (; sz > bsz; branch += 1, sz -= bsz) { | |
444 | if (!branch->from && !branch->to) | |
445 | continue; | |
446 | intel_bts_get_branch_type(btsq, branch); | |
447 | if (filter && !(filter & btsq->sample_flags)) | |
448 | continue; | |
449 | err = intel_bts_synth_branch_sample(btsq, branch); | |
450 | if (err) | |
451 | break; | |
452 | } | |
453 | return err; | |
454 | } | |
455 | ||
456 | static int intel_bts_process_queue(struct intel_bts_queue *btsq, u64 *timestamp) | |
457 | { | |
458 | struct auxtrace_buffer *buffer = btsq->buffer, *old_buffer = buffer; | |
459 | struct auxtrace_queue *queue; | |
460 | struct thread *thread; | |
461 | int err; | |
462 | ||
463 | if (btsq->done) | |
464 | return 1; | |
465 | ||
466 | if (btsq->pid == -1) { | |
467 | thread = machine__find_thread(btsq->bts->machine, -1, | |
468 | btsq->tid); | |
469 | if (thread) | |
470 | btsq->pid = thread->pid_; | |
471 | } else { | |
472 | thread = machine__findnew_thread(btsq->bts->machine, btsq->pid, | |
473 | btsq->tid); | |
474 | } | |
475 | ||
476 | queue = &btsq->bts->queues.queue_array[btsq->queue_nr]; | |
477 | ||
478 | if (!buffer) | |
479 | buffer = auxtrace_buffer__next(queue, NULL); | |
480 | ||
481 | if (!buffer) { | |
482 | if (!btsq->bts->sampling_mode) | |
483 | btsq->done = 1; | |
484 | err = 1; | |
485 | goto out_put; | |
486 | } | |
487 | ||
488 | /* Currently there is no support for split buffers */ | |
489 | if (buffer->consecutive) { | |
490 | err = -EINVAL; | |
491 | goto out_put; | |
492 | } | |
493 | ||
494 | if (!buffer->data) { | |
495 | int fd = perf_data_file__fd(btsq->bts->session->file); | |
496 | ||
497 | buffer->data = auxtrace_buffer__get_data(buffer, fd); | |
498 | if (!buffer->data) { | |
499 | err = -ENOMEM; | |
500 | goto out_put; | |
501 | } | |
502 | } | |
503 | ||
504 | if (btsq->bts->snapshot_mode && !buffer->consecutive && | |
505 | intel_bts_do_fix_overlap(queue, buffer)) { | |
506 | err = -ENOMEM; | |
507 | goto out_put; | |
508 | } | |
509 | ||
510 | if (!btsq->bts->synth_opts.callchain && thread && | |
511 | (!old_buffer || btsq->bts->sampling_mode || | |
512 | (btsq->bts->snapshot_mode && !buffer->consecutive))) | |
513 | thread_stack__set_trace_nr(thread, buffer->buffer_nr + 1); | |
514 | ||
515 | err = intel_bts_process_buffer(btsq, buffer); | |
516 | ||
517 | auxtrace_buffer__drop_data(buffer); | |
518 | ||
519 | btsq->buffer = auxtrace_buffer__next(queue, buffer); | |
520 | if (btsq->buffer) { | |
521 | if (timestamp) | |
522 | *timestamp = btsq->buffer->reference; | |
523 | } else { | |
524 | if (!btsq->bts->sampling_mode) | |
525 | btsq->done = 1; | |
526 | } | |
527 | out_put: | |
528 | thread__put(thread); | |
529 | return err; | |
530 | } | |
531 | ||
532 | static int intel_bts_flush_queue(struct intel_bts_queue *btsq) | |
533 | { | |
534 | u64 ts = 0; | |
535 | int ret; | |
536 | ||
537 | while (1) { | |
538 | ret = intel_bts_process_queue(btsq, &ts); | |
539 | if (ret < 0) | |
540 | return ret; | |
541 | if (ret) | |
542 | break; | |
543 | } | |
544 | return 0; | |
545 | } | |
546 | ||
547 | static int intel_bts_process_tid_exit(struct intel_bts *bts, pid_t tid) | |
548 | { | |
549 | struct auxtrace_queues *queues = &bts->queues; | |
550 | unsigned int i; | |
551 | ||
552 | for (i = 0; i < queues->nr_queues; i++) { | |
553 | struct auxtrace_queue *queue = &bts->queues.queue_array[i]; | |
554 | struct intel_bts_queue *btsq = queue->priv; | |
555 | ||
556 | if (btsq && btsq->tid == tid) | |
557 | return intel_bts_flush_queue(btsq); | |
558 | } | |
559 | return 0; | |
560 | } | |
561 | ||
562 | static int intel_bts_process_queues(struct intel_bts *bts, u64 timestamp) | |
563 | { | |
564 | while (1) { | |
565 | unsigned int queue_nr; | |
566 | struct auxtrace_queue *queue; | |
567 | struct intel_bts_queue *btsq; | |
568 | u64 ts = 0; | |
569 | int ret; | |
570 | ||
571 | if (!bts->heap.heap_cnt) | |
572 | return 0; | |
573 | ||
574 | if (bts->heap.heap_array[0].ordinal > timestamp) | |
575 | return 0; | |
576 | ||
577 | queue_nr = bts->heap.heap_array[0].queue_nr; | |
578 | queue = &bts->queues.queue_array[queue_nr]; | |
579 | btsq = queue->priv; | |
580 | ||
581 | auxtrace_heap__pop(&bts->heap); | |
582 | ||
583 | ret = intel_bts_process_queue(btsq, &ts); | |
584 | if (ret < 0) { | |
585 | auxtrace_heap__add(&bts->heap, queue_nr, ts); | |
586 | return ret; | |
587 | } | |
588 | ||
589 | if (!ret) { | |
590 | ret = auxtrace_heap__add(&bts->heap, queue_nr, ts); | |
591 | if (ret < 0) | |
592 | return ret; | |
593 | } else { | |
594 | btsq->on_heap = false; | |
595 | } | |
596 | } | |
597 | ||
598 | return 0; | |
599 | } | |
600 | ||
601 | static int intel_bts_process_event(struct perf_session *session, | |
602 | union perf_event *event, | |
603 | struct perf_sample *sample, | |
604 | struct perf_tool *tool) | |
605 | { | |
606 | struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts, | |
607 | auxtrace); | |
608 | u64 timestamp; | |
609 | int err; | |
610 | ||
611 | if (dump_trace) | |
612 | return 0; | |
613 | ||
614 | if (!tool->ordered_events) { | |
615 | pr_err("Intel BTS requires ordered events\n"); | |
616 | return -EINVAL; | |
617 | } | |
618 | ||
619 | if (sample->time && sample->time != (u64)-1) | |
620 | timestamp = perf_time_to_tsc(sample->time, &bts->tc); | |
621 | else | |
622 | timestamp = 0; | |
623 | ||
624 | err = intel_bts_update_queues(bts); | |
625 | if (err) | |
626 | return err; | |
627 | ||
628 | err = intel_bts_process_queues(bts, timestamp); | |
629 | if (err) | |
630 | return err; | |
631 | if (event->header.type == PERF_RECORD_EXIT) { | |
53ff6bc3 | 632 | err = intel_bts_process_tid_exit(bts, event->fork.tid); |
d0170af7 AH |
633 | if (err) |
634 | return err; | |
635 | } | |
636 | ||
637 | if (event->header.type == PERF_RECORD_AUX && | |
638 | (event->aux.flags & PERF_AUX_FLAG_TRUNCATED) && | |
639 | bts->synth_opts.errors) | |
640 | err = intel_bts_lost(bts, sample); | |
641 | ||
642 | return err; | |
643 | } | |
644 | ||
645 | static int intel_bts_process_auxtrace_event(struct perf_session *session, | |
646 | union perf_event *event, | |
647 | struct perf_tool *tool __maybe_unused) | |
648 | { | |
649 | struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts, | |
650 | auxtrace); | |
651 | ||
652 | if (bts->sampling_mode) | |
653 | return 0; | |
654 | ||
655 | if (!bts->data_queued) { | |
656 | struct auxtrace_buffer *buffer; | |
657 | off_t data_offset; | |
658 | int fd = perf_data_file__fd(session->file); | |
659 | int err; | |
660 | ||
661 | if (perf_data_file__is_pipe(session->file)) { | |
662 | data_offset = 0; | |
663 | } else { | |
664 | data_offset = lseek(fd, 0, SEEK_CUR); | |
665 | if (data_offset == -1) | |
666 | return -errno; | |
667 | } | |
668 | ||
669 | err = auxtrace_queues__add_event(&bts->queues, session, event, | |
670 | data_offset, &buffer); | |
671 | if (err) | |
672 | return err; | |
673 | ||
674 | /* Dump here now we have copied a piped trace out of the pipe */ | |
675 | if (dump_trace) { | |
676 | if (auxtrace_buffer__get_data(buffer, fd)) { | |
677 | intel_bts_dump_event(bts, buffer->data, | |
678 | buffer->size); | |
679 | auxtrace_buffer__put_data(buffer); | |
680 | } | |
681 | } | |
682 | } | |
683 | ||
684 | return 0; | |
685 | } | |
686 | ||
b8f8eb84 | 687 | static int intel_bts_flush(struct perf_session *session, |
d0170af7 AH |
688 | struct perf_tool *tool __maybe_unused) |
689 | { | |
690 | struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts, | |
691 | auxtrace); | |
692 | int ret; | |
693 | ||
694 | if (dump_trace || bts->sampling_mode) | |
695 | return 0; | |
696 | ||
697 | if (!tool->ordered_events) | |
698 | return -EINVAL; | |
699 | ||
700 | ret = intel_bts_update_queues(bts); | |
701 | if (ret < 0) | |
702 | return ret; | |
703 | ||
704 | return intel_bts_process_queues(bts, MAX_TIMESTAMP); | |
705 | } | |
706 | ||
707 | static void intel_bts_free_queue(void *priv) | |
708 | { | |
709 | struct intel_bts_queue *btsq = priv; | |
710 | ||
711 | if (!btsq) | |
712 | return; | |
713 | free(btsq); | |
714 | } | |
715 | ||
716 | static void intel_bts_free_events(struct perf_session *session) | |
717 | { | |
718 | struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts, | |
719 | auxtrace); | |
720 | struct auxtrace_queues *queues = &bts->queues; | |
721 | unsigned int i; | |
722 | ||
723 | for (i = 0; i < queues->nr_queues; i++) { | |
724 | intel_bts_free_queue(queues->queue_array[i].priv); | |
725 | queues->queue_array[i].priv = NULL; | |
726 | } | |
727 | auxtrace_queues__free(queues); | |
728 | } | |
729 | ||
730 | static void intel_bts_free(struct perf_session *session) | |
731 | { | |
732 | struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts, | |
733 | auxtrace); | |
734 | ||
735 | auxtrace_heap__free(&bts->heap); | |
736 | intel_bts_free_events(session); | |
737 | session->auxtrace = NULL; | |
738 | free(bts); | |
739 | } | |
740 | ||
741 | struct intel_bts_synth { | |
742 | struct perf_tool dummy_tool; | |
743 | struct perf_session *session; | |
744 | }; | |
745 | ||
746 | static int intel_bts_event_synth(struct perf_tool *tool, | |
747 | union perf_event *event, | |
748 | struct perf_sample *sample __maybe_unused, | |
749 | struct machine *machine __maybe_unused) | |
750 | { | |
751 | struct intel_bts_synth *intel_bts_synth = | |
752 | container_of(tool, struct intel_bts_synth, dummy_tool); | |
753 | ||
754 | return perf_session__deliver_synth_event(intel_bts_synth->session, | |
755 | event, NULL); | |
756 | } | |
757 | ||
758 | static int intel_bts_synth_event(struct perf_session *session, | |
759 | struct perf_event_attr *attr, u64 id) | |
760 | { | |
761 | struct intel_bts_synth intel_bts_synth; | |
762 | ||
763 | memset(&intel_bts_synth, 0, sizeof(struct intel_bts_synth)); | |
764 | intel_bts_synth.session = session; | |
765 | ||
766 | return perf_event__synthesize_attr(&intel_bts_synth.dummy_tool, attr, 1, | |
767 | &id, intel_bts_event_synth); | |
768 | } | |
769 | ||
770 | static int intel_bts_synth_events(struct intel_bts *bts, | |
771 | struct perf_session *session) | |
772 | { | |
773 | struct perf_evlist *evlist = session->evlist; | |
774 | struct perf_evsel *evsel; | |
775 | struct perf_event_attr attr; | |
776 | bool found = false; | |
777 | u64 id; | |
778 | int err; | |
779 | ||
780 | evlist__for_each(evlist, evsel) { | |
781 | if (evsel->attr.type == bts->pmu_type && evsel->ids) { | |
782 | found = true; | |
783 | break; | |
784 | } | |
785 | } | |
786 | ||
787 | if (!found) { | |
788 | pr_debug("There are no selected events with Intel BTS data\n"); | |
789 | return 0; | |
790 | } | |
791 | ||
792 | memset(&attr, 0, sizeof(struct perf_event_attr)); | |
793 | attr.size = sizeof(struct perf_event_attr); | |
794 | attr.type = PERF_TYPE_HARDWARE; | |
795 | attr.sample_type = evsel->attr.sample_type & PERF_SAMPLE_MASK; | |
796 | attr.sample_type |= PERF_SAMPLE_IP | PERF_SAMPLE_TID | | |
797 | PERF_SAMPLE_PERIOD; | |
798 | attr.sample_type &= ~(u64)PERF_SAMPLE_TIME; | |
799 | attr.sample_type &= ~(u64)PERF_SAMPLE_CPU; | |
800 | attr.exclude_user = evsel->attr.exclude_user; | |
801 | attr.exclude_kernel = evsel->attr.exclude_kernel; | |
802 | attr.exclude_hv = evsel->attr.exclude_hv; | |
803 | attr.exclude_host = evsel->attr.exclude_host; | |
804 | attr.exclude_guest = evsel->attr.exclude_guest; | |
805 | attr.sample_id_all = evsel->attr.sample_id_all; | |
806 | attr.read_format = evsel->attr.read_format; | |
807 | ||
808 | id = evsel->id[0] + 1000000000; | |
809 | if (!id) | |
810 | id = 1; | |
811 | ||
812 | if (bts->synth_opts.branches) { | |
813 | attr.config = PERF_COUNT_HW_BRANCH_INSTRUCTIONS; | |
814 | attr.sample_period = 1; | |
815 | attr.sample_type |= PERF_SAMPLE_ADDR; | |
816 | pr_debug("Synthesizing 'branches' event with id %" PRIu64 " sample type %#" PRIx64 "\n", | |
817 | id, (u64)attr.sample_type); | |
818 | err = intel_bts_synth_event(session, &attr, id); | |
819 | if (err) { | |
820 | pr_err("%s: failed to synthesize 'branches' event type\n", | |
821 | __func__); | |
822 | return err; | |
823 | } | |
824 | bts->sample_branches = true; | |
825 | bts->branches_sample_type = attr.sample_type; | |
826 | bts->branches_id = id; | |
827 | /* | |
828 | * We only use sample types from PERF_SAMPLE_MASK so we can use | |
829 | * __perf_evsel__sample_size() here. | |
830 | */ | |
831 | bts->branches_event_size = sizeof(struct sample_event) + | |
832 | __perf_evsel__sample_size(attr.sample_type); | |
833 | } | |
834 | ||
835 | bts->synth_needs_swap = evsel->needs_swap; | |
836 | ||
837 | return 0; | |
838 | } | |
839 | ||
840 | static const char * const intel_bts_info_fmts[] = { | |
841 | [INTEL_BTS_PMU_TYPE] = " PMU Type %"PRId64"\n", | |
842 | [INTEL_BTS_TIME_SHIFT] = " Time Shift %"PRIu64"\n", | |
843 | [INTEL_BTS_TIME_MULT] = " Time Muliplier %"PRIu64"\n", | |
844 | [INTEL_BTS_TIME_ZERO] = " Time Zero %"PRIu64"\n", | |
845 | [INTEL_BTS_CAP_USER_TIME_ZERO] = " Cap Time Zero %"PRId64"\n", | |
846 | [INTEL_BTS_SNAPSHOT_MODE] = " Snapshot mode %"PRId64"\n", | |
847 | }; | |
848 | ||
849 | static void intel_bts_print_info(u64 *arr, int start, int finish) | |
850 | { | |
851 | int i; | |
852 | ||
853 | if (!dump_trace) | |
854 | return; | |
855 | ||
856 | for (i = start; i <= finish; i++) | |
857 | fprintf(stdout, intel_bts_info_fmts[i], arr[i]); | |
858 | } | |
859 | ||
860 | u64 intel_bts_auxtrace_info_priv[INTEL_BTS_AUXTRACE_PRIV_SIZE]; | |
861 | ||
862 | int intel_bts_process_auxtrace_info(union perf_event *event, | |
863 | struct perf_session *session) | |
864 | { | |
865 | struct auxtrace_info_event *auxtrace_info = &event->auxtrace_info; | |
866 | size_t min_sz = sizeof(u64) * INTEL_BTS_SNAPSHOT_MODE; | |
867 | struct intel_bts *bts; | |
868 | int err; | |
869 | ||
870 | if (auxtrace_info->header.size < sizeof(struct auxtrace_info_event) + | |
871 | min_sz) | |
872 | return -EINVAL; | |
873 | ||
874 | bts = zalloc(sizeof(struct intel_bts)); | |
875 | if (!bts) | |
876 | return -ENOMEM; | |
877 | ||
878 | err = auxtrace_queues__init(&bts->queues); | |
879 | if (err) | |
880 | goto err_free; | |
881 | ||
882 | bts->session = session; | |
883 | bts->machine = &session->machines.host; /* No kvm support */ | |
884 | bts->auxtrace_type = auxtrace_info->type; | |
885 | bts->pmu_type = auxtrace_info->priv[INTEL_BTS_PMU_TYPE]; | |
886 | bts->tc.time_shift = auxtrace_info->priv[INTEL_BTS_TIME_SHIFT]; | |
887 | bts->tc.time_mult = auxtrace_info->priv[INTEL_BTS_TIME_MULT]; | |
888 | bts->tc.time_zero = auxtrace_info->priv[INTEL_BTS_TIME_ZERO]; | |
889 | bts->cap_user_time_zero = | |
890 | auxtrace_info->priv[INTEL_BTS_CAP_USER_TIME_ZERO]; | |
891 | bts->snapshot_mode = auxtrace_info->priv[INTEL_BTS_SNAPSHOT_MODE]; | |
892 | ||
893 | bts->sampling_mode = false; | |
894 | ||
895 | bts->auxtrace.process_event = intel_bts_process_event; | |
896 | bts->auxtrace.process_auxtrace_event = intel_bts_process_auxtrace_event; | |
897 | bts->auxtrace.flush_events = intel_bts_flush; | |
898 | bts->auxtrace.free_events = intel_bts_free_events; | |
899 | bts->auxtrace.free = intel_bts_free; | |
900 | session->auxtrace = &bts->auxtrace; | |
901 | ||
902 | intel_bts_print_info(&auxtrace_info->priv[0], INTEL_BTS_PMU_TYPE, | |
903 | INTEL_BTS_SNAPSHOT_MODE); | |
904 | ||
905 | if (dump_trace) | |
906 | return 0; | |
907 | ||
908 | if (session->itrace_synth_opts && session->itrace_synth_opts->set) | |
909 | bts->synth_opts = *session->itrace_synth_opts; | |
910 | else | |
911 | itrace_synth_opts__set_default(&bts->synth_opts); | |
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 | } |