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> | |
a43783ae | 17 | #include <errno.h> |
d0170af7 | 18 | #include <byteswap.h> |
fd20e811 | 19 | #include <inttypes.h> |
d0170af7 AH |
20 | #include <linux/kernel.h> |
21 | #include <linux/types.h> | |
22 | #include <linux/bitops.h> | |
23 | #include <linux/log2.h> | |
24 | ||
25 | #include "cpumap.h" | |
26 | #include "color.h" | |
27 | #include "evsel.h" | |
28 | #include "evlist.h" | |
29 | #include "machine.h" | |
1101f69a | 30 | #include "map.h" |
d0170af7 AH |
31 | #include "session.h" |
32 | #include "util.h" | |
33 | #include "thread.h" | |
34 | #include "thread-stack.h" | |
35 | #include "debug.h" | |
36 | #include "tsc.h" | |
37 | #include "auxtrace.h" | |
38 | #include "intel-pt-decoder/intel-pt-insn-decoder.h" | |
39 | #include "intel-bts.h" | |
40 | ||
41 | #define MAX_TIMESTAMP (~0ULL) | |
42 | ||
43 | #define INTEL_BTS_ERR_NOINSN 5 | |
44 | #define INTEL_BTS_ERR_LOST 9 | |
45 | ||
46 | #if __BYTE_ORDER == __BIG_ENDIAN | |
47 | #define le64_to_cpu bswap_64 | |
48 | #else | |
49 | #define le64_to_cpu | |
50 | #endif | |
51 | ||
52 | struct intel_bts { | |
53 | struct auxtrace auxtrace; | |
54 | struct auxtrace_queues queues; | |
55 | struct auxtrace_heap heap; | |
56 | u32 auxtrace_type; | |
57 | struct perf_session *session; | |
58 | struct machine *machine; | |
59 | bool sampling_mode; | |
60 | bool snapshot_mode; | |
61 | bool data_queued; | |
62 | u32 pmu_type; | |
63 | struct perf_tsc_conversion tc; | |
64 | bool cap_user_time_zero; | |
65 | struct itrace_synth_opts synth_opts; | |
66 | bool sample_branches; | |
67 | u32 branches_filter; | |
68 | u64 branches_sample_type; | |
69 | u64 branches_id; | |
70 | size_t branches_event_size; | |
d1706b39 | 71 | unsigned long num_events; |
d0170af7 AH |
72 | }; |
73 | ||
74 | struct intel_bts_queue { | |
75 | struct intel_bts *bts; | |
76 | unsigned int queue_nr; | |
77 | struct auxtrace_buffer *buffer; | |
78 | bool on_heap; | |
79 | bool done; | |
80 | pid_t pid; | |
81 | pid_t tid; | |
82 | int cpu; | |
83 | u64 time; | |
84 | struct intel_pt_insn intel_pt_insn; | |
85 | u32 sample_flags; | |
86 | }; | |
87 | ||
88 | struct branch { | |
89 | u64 from; | |
90 | u64 to; | |
91 | u64 misc; | |
92 | }; | |
93 | ||
94 | static void intel_bts_dump(struct intel_bts *bts __maybe_unused, | |
95 | unsigned char *buf, size_t len) | |
96 | { | |
97 | struct branch *branch; | |
98 | size_t i, pos = 0, br_sz = sizeof(struct branch), sz; | |
99 | const char *color = PERF_COLOR_BLUE; | |
100 | ||
101 | color_fprintf(stdout, color, | |
102 | ". ... Intel BTS data: size %zu bytes\n", | |
103 | len); | |
104 | ||
105 | while (len) { | |
106 | if (len >= br_sz) | |
107 | sz = br_sz; | |
108 | else | |
109 | sz = len; | |
110 | printf("."); | |
111 | color_fprintf(stdout, color, " %08x: ", pos); | |
112 | for (i = 0; i < sz; i++) | |
113 | color_fprintf(stdout, color, " %02x", buf[i]); | |
114 | for (; i < br_sz; i++) | |
115 | color_fprintf(stdout, color, " "); | |
116 | if (len >= br_sz) { | |
117 | branch = (struct branch *)buf; | |
118 | color_fprintf(stdout, color, " %"PRIx64" -> %"PRIx64" %s\n", | |
119 | le64_to_cpu(branch->from), | |
120 | le64_to_cpu(branch->to), | |
121 | le64_to_cpu(branch->misc) & 0x10 ? | |
122 | "pred" : "miss"); | |
123 | } else { | |
124 | color_fprintf(stdout, color, " Bad record!\n"); | |
125 | } | |
126 | pos += sz; | |
127 | buf += sz; | |
128 | len -= sz; | |
129 | } | |
130 | } | |
131 | ||
132 | static void intel_bts_dump_event(struct intel_bts *bts, unsigned char *buf, | |
133 | size_t len) | |
134 | { | |
135 | printf(".\n"); | |
136 | intel_bts_dump(bts, buf, len); | |
137 | } | |
138 | ||
139 | static int intel_bts_lost(struct intel_bts *bts, struct perf_sample *sample) | |
140 | { | |
141 | union perf_event event; | |
142 | int err; | |
143 | ||
144 | auxtrace_synth_error(&event.auxtrace_error, PERF_AUXTRACE_ERROR_ITRACE, | |
145 | INTEL_BTS_ERR_LOST, sample->cpu, sample->pid, | |
146 | sample->tid, 0, "Lost trace data"); | |
147 | ||
148 | err = perf_session__deliver_synth_event(bts->session, &event, NULL); | |
149 | if (err) | |
150 | pr_err("Intel BTS: failed to deliver error event, error %d\n", | |
151 | err); | |
152 | ||
153 | return err; | |
154 | } | |
155 | ||
156 | static struct intel_bts_queue *intel_bts_alloc_queue(struct intel_bts *bts, | |
157 | unsigned int queue_nr) | |
158 | { | |
159 | struct intel_bts_queue *btsq; | |
160 | ||
161 | btsq = zalloc(sizeof(struct intel_bts_queue)); | |
162 | if (!btsq) | |
163 | return NULL; | |
164 | ||
165 | btsq->bts = bts; | |
166 | btsq->queue_nr = queue_nr; | |
167 | btsq->pid = -1; | |
168 | btsq->tid = -1; | |
169 | btsq->cpu = -1; | |
170 | ||
171 | return btsq; | |
172 | } | |
173 | ||
174 | static int intel_bts_setup_queue(struct intel_bts *bts, | |
175 | struct auxtrace_queue *queue, | |
176 | unsigned int queue_nr) | |
177 | { | |
178 | struct intel_bts_queue *btsq = queue->priv; | |
179 | ||
180 | if (list_empty(&queue->head)) | |
181 | return 0; | |
182 | ||
183 | if (!btsq) { | |
184 | btsq = intel_bts_alloc_queue(bts, queue_nr); | |
185 | if (!btsq) | |
186 | return -ENOMEM; | |
187 | queue->priv = btsq; | |
188 | ||
189 | if (queue->cpu != -1) | |
190 | btsq->cpu = queue->cpu; | |
191 | btsq->tid = queue->tid; | |
192 | } | |
193 | ||
194 | if (bts->sampling_mode) | |
195 | return 0; | |
196 | ||
197 | if (!btsq->on_heap && !btsq->buffer) { | |
198 | int ret; | |
199 | ||
200 | btsq->buffer = auxtrace_buffer__next(queue, NULL); | |
201 | if (!btsq->buffer) | |
202 | return 0; | |
203 | ||
204 | ret = auxtrace_heap__add(&bts->heap, queue_nr, | |
205 | btsq->buffer->reference); | |
206 | if (ret) | |
207 | return ret; | |
208 | btsq->on_heap = true; | |
209 | } | |
210 | ||
211 | return 0; | |
212 | } | |
213 | ||
214 | static int intel_bts_setup_queues(struct intel_bts *bts) | |
215 | { | |
216 | unsigned int i; | |
217 | int ret; | |
218 | ||
219 | for (i = 0; i < bts->queues.nr_queues; i++) { | |
220 | ret = intel_bts_setup_queue(bts, &bts->queues.queue_array[i], | |
221 | i); | |
222 | if (ret) | |
223 | return ret; | |
224 | } | |
225 | return 0; | |
226 | } | |
227 | ||
228 | static inline int intel_bts_update_queues(struct intel_bts *bts) | |
229 | { | |
230 | if (bts->queues.new_data) { | |
231 | bts->queues.new_data = false; | |
232 | return intel_bts_setup_queues(bts); | |
233 | } | |
234 | return 0; | |
235 | } | |
236 | ||
237 | static unsigned char *intel_bts_find_overlap(unsigned char *buf_a, size_t len_a, | |
238 | unsigned char *buf_b, size_t len_b) | |
239 | { | |
240 | size_t offs, len; | |
241 | ||
242 | if (len_a > len_b) | |
243 | offs = len_a - len_b; | |
244 | else | |
245 | offs = 0; | |
246 | ||
247 | for (; offs < len_a; offs += sizeof(struct branch)) { | |
248 | len = len_a - offs; | |
249 | if (!memcmp(buf_a + offs, buf_b, len)) | |
250 | return buf_b + len; | |
251 | } | |
252 | ||
253 | return buf_b; | |
254 | } | |
255 | ||
256 | static int intel_bts_do_fix_overlap(struct auxtrace_queue *queue, | |
257 | struct auxtrace_buffer *b) | |
258 | { | |
259 | struct auxtrace_buffer *a; | |
260 | void *start; | |
261 | ||
262 | if (b->list.prev == &queue->head) | |
263 | return 0; | |
264 | a = list_entry(b->list.prev, struct auxtrace_buffer, list); | |
265 | start = intel_bts_find_overlap(a->data, a->size, b->data, b->size); | |
266 | if (!start) | |
267 | return -EINVAL; | |
268 | b->use_size = b->data + b->size - start; | |
269 | b->use_data = start; | |
270 | return 0; | |
271 | } | |
272 | ||
5d4f0eda AH |
273 | static inline u8 intel_bts_cpumode(struct intel_bts *bts, uint64_t ip) |
274 | { | |
275 | return machine__kernel_ip(bts->machine, ip) ? | |
276 | PERF_RECORD_MISC_KERNEL : | |
277 | PERF_RECORD_MISC_USER; | |
278 | } | |
279 | ||
d0170af7 AH |
280 | static int intel_bts_synth_branch_sample(struct intel_bts_queue *btsq, |
281 | struct branch *branch) | |
282 | { | |
283 | int ret; | |
284 | struct intel_bts *bts = btsq->bts; | |
285 | union perf_event event; | |
286 | struct perf_sample sample = { .ip = 0, }; | |
287 | ||
d1706b39 AK |
288 | if (bts->synth_opts.initial_skip && |
289 | bts->num_events++ <= bts->synth_opts.initial_skip) | |
290 | return 0; | |
291 | ||
d0170af7 | 292 | sample.ip = le64_to_cpu(branch->from); |
5d4f0eda | 293 | sample.cpumode = intel_bts_cpumode(bts, sample.ip); |
d0170af7 AH |
294 | sample.pid = btsq->pid; |
295 | sample.tid = btsq->tid; | |
296 | sample.addr = le64_to_cpu(branch->to); | |
297 | sample.id = btsq->bts->branches_id; | |
298 | sample.stream_id = btsq->bts->branches_id; | |
299 | sample.period = 1; | |
300 | sample.cpu = btsq->cpu; | |
301 | sample.flags = btsq->sample_flags; | |
302 | sample.insn_len = btsq->intel_pt_insn.length; | |
faaa8768 | 303 | memcpy(sample.insn, btsq->intel_pt_insn.buf, INTEL_PT_INSN_BUF_SZ); |
d0170af7 | 304 | |
5d4f0eda AH |
305 | event.sample.header.type = PERF_RECORD_SAMPLE; |
306 | event.sample.header.misc = sample.cpumode; | |
307 | event.sample.header.size = sizeof(struct perf_event_header); | |
308 | ||
d0170af7 AH |
309 | if (bts->synth_opts.inject) { |
310 | event.sample.header.size = bts->branches_event_size; | |
311 | ret = perf_event__synthesize_sample(&event, | |
312 | bts->branches_sample_type, | |
936f1f30 | 313 | 0, &sample); |
d0170af7 AH |
314 | if (ret) |
315 | return ret; | |
316 | } | |
317 | ||
318 | ret = perf_session__deliver_synth_event(bts->session, &event, &sample); | |
319 | if (ret) | |
320 | pr_err("Intel BTS: failed to deliver branch event, error %d\n", | |
321 | ret); | |
322 | ||
323 | return ret; | |
324 | } | |
325 | ||
326 | static int intel_bts_get_next_insn(struct intel_bts_queue *btsq, u64 ip) | |
327 | { | |
328 | struct machine *machine = btsq->bts->machine; | |
329 | struct thread *thread; | |
330 | struct addr_location al; | |
32f98aab | 331 | unsigned char buf[INTEL_PT_INSN_BUF_SZ]; |
d0170af7 AH |
332 | ssize_t len; |
333 | int x86_64; | |
334 | uint8_t cpumode; | |
335 | int err = -1; | |
336 | ||
d0170af7 AH |
337 | if (machine__kernel_ip(machine, ip)) |
338 | cpumode = PERF_RECORD_MISC_KERNEL; | |
339 | else | |
340 | cpumode = PERF_RECORD_MISC_USER; | |
341 | ||
342 | thread = machine__find_thread(machine, -1, btsq->tid); | |
343 | if (!thread) | |
344 | return -1; | |
345 | ||
71a84b5a | 346 | if (!thread__find_map(thread, cpumode, ip, &al) || !al.map->dso) |
d0170af7 AH |
347 | goto out_put; |
348 | ||
32f98aab AH |
349 | len = dso__data_read_addr(al.map->dso, al.map, machine, ip, buf, |
350 | INTEL_PT_INSN_BUF_SZ); | |
d0170af7 AH |
351 | if (len <= 0) |
352 | goto out_put; | |
353 | ||
354 | /* Load maps to ensure dso->is_64_bit has been updated */ | |
be39db9f | 355 | map__load(al.map); |
d0170af7 AH |
356 | |
357 | x86_64 = al.map->dso->is_64_bit; | |
358 | ||
359 | if (intel_pt_get_insn(buf, len, x86_64, &btsq->intel_pt_insn)) | |
360 | goto out_put; | |
361 | ||
362 | err = 0; | |
363 | out_put: | |
364 | thread__put(thread); | |
365 | return err; | |
366 | } | |
367 | ||
368 | static int intel_bts_synth_error(struct intel_bts *bts, int cpu, pid_t pid, | |
369 | pid_t tid, u64 ip) | |
370 | { | |
371 | union perf_event event; | |
372 | int err; | |
373 | ||
374 | auxtrace_synth_error(&event.auxtrace_error, PERF_AUXTRACE_ERROR_ITRACE, | |
375 | INTEL_BTS_ERR_NOINSN, cpu, pid, tid, ip, | |
376 | "Failed to get instruction"); | |
377 | ||
378 | err = perf_session__deliver_synth_event(bts->session, &event, NULL); | |
379 | if (err) | |
380 | pr_err("Intel BTS: failed to deliver error event, error %d\n", | |
381 | err); | |
382 | ||
383 | return err; | |
384 | } | |
385 | ||
386 | static int intel_bts_get_branch_type(struct intel_bts_queue *btsq, | |
387 | struct branch *branch) | |
388 | { | |
389 | int err; | |
390 | ||
391 | if (!branch->from) { | |
392 | if (branch->to) | |
393 | btsq->sample_flags = PERF_IP_FLAG_BRANCH | | |
394 | PERF_IP_FLAG_TRACE_BEGIN; | |
395 | else | |
396 | btsq->sample_flags = 0; | |
397 | btsq->intel_pt_insn.length = 0; | |
398 | } else if (!branch->to) { | |
399 | btsq->sample_flags = PERF_IP_FLAG_BRANCH | | |
400 | PERF_IP_FLAG_TRACE_END; | |
401 | btsq->intel_pt_insn.length = 0; | |
402 | } else { | |
403 | err = intel_bts_get_next_insn(btsq, branch->from); | |
404 | if (err) { | |
405 | btsq->sample_flags = 0; | |
406 | btsq->intel_pt_insn.length = 0; | |
407 | if (!btsq->bts->synth_opts.errors) | |
408 | return 0; | |
409 | err = intel_bts_synth_error(btsq->bts, btsq->cpu, | |
410 | btsq->pid, btsq->tid, | |
411 | branch->from); | |
412 | return err; | |
413 | } | |
414 | btsq->sample_flags = intel_pt_insn_type(btsq->intel_pt_insn.op); | |
415 | /* Check for an async branch into the kernel */ | |
416 | if (!machine__kernel_ip(btsq->bts->machine, branch->from) && | |
417 | machine__kernel_ip(btsq->bts->machine, branch->to) && | |
418 | btsq->sample_flags != (PERF_IP_FLAG_BRANCH | | |
419 | PERF_IP_FLAG_CALL | | |
420 | PERF_IP_FLAG_SYSCALLRET)) | |
421 | btsq->sample_flags = PERF_IP_FLAG_BRANCH | | |
422 | PERF_IP_FLAG_CALL | | |
423 | PERF_IP_FLAG_ASYNC | | |
424 | PERF_IP_FLAG_INTERRUPT; | |
425 | } | |
426 | ||
427 | return 0; | |
428 | } | |
429 | ||
430 | static int intel_bts_process_buffer(struct intel_bts_queue *btsq, | |
50f73637 AH |
431 | struct auxtrace_buffer *buffer, |
432 | struct thread *thread) | |
d0170af7 AH |
433 | { |
434 | struct branch *branch; | |
435 | size_t sz, bsz = sizeof(struct branch); | |
436 | u32 filter = btsq->bts->branches_filter; | |
437 | int err = 0; | |
438 | ||
439 | if (buffer->use_data) { | |
440 | sz = buffer->use_size; | |
441 | branch = buffer->use_data; | |
442 | } else { | |
443 | sz = buffer->size; | |
444 | branch = buffer->data; | |
445 | } | |
446 | ||
447 | if (!btsq->bts->sample_branches) | |
448 | return 0; | |
449 | ||
450 | for (; sz > bsz; branch += 1, sz -= bsz) { | |
451 | if (!branch->from && !branch->to) | |
452 | continue; | |
453 | intel_bts_get_branch_type(btsq, branch); | |
50f73637 | 454 | if (btsq->bts->synth_opts.thread_stack) |
256d92bc | 455 | thread_stack__event(thread, btsq->cpu, btsq->sample_flags, |
50f73637 AH |
456 | le64_to_cpu(branch->from), |
457 | le64_to_cpu(branch->to), | |
458 | btsq->intel_pt_insn.length, | |
459 | buffer->buffer_nr + 1); | |
d0170af7 AH |
460 | if (filter && !(filter & btsq->sample_flags)) |
461 | continue; | |
462 | err = intel_bts_synth_branch_sample(btsq, branch); | |
463 | if (err) | |
464 | break; | |
465 | } | |
466 | return err; | |
467 | } | |
468 | ||
469 | static int intel_bts_process_queue(struct intel_bts_queue *btsq, u64 *timestamp) | |
470 | { | |
471 | struct auxtrace_buffer *buffer = btsq->buffer, *old_buffer = buffer; | |
472 | struct auxtrace_queue *queue; | |
473 | struct thread *thread; | |
474 | int err; | |
475 | ||
476 | if (btsq->done) | |
477 | return 1; | |
478 | ||
479 | if (btsq->pid == -1) { | |
480 | thread = machine__find_thread(btsq->bts->machine, -1, | |
481 | btsq->tid); | |
482 | if (thread) | |
483 | btsq->pid = thread->pid_; | |
484 | } else { | |
485 | thread = machine__findnew_thread(btsq->bts->machine, btsq->pid, | |
486 | btsq->tid); | |
487 | } | |
488 | ||
489 | queue = &btsq->bts->queues.queue_array[btsq->queue_nr]; | |
490 | ||
491 | if (!buffer) | |
492 | buffer = auxtrace_buffer__next(queue, NULL); | |
493 | ||
494 | if (!buffer) { | |
495 | if (!btsq->bts->sampling_mode) | |
496 | btsq->done = 1; | |
497 | err = 1; | |
498 | goto out_put; | |
499 | } | |
500 | ||
501 | /* Currently there is no support for split buffers */ | |
502 | if (buffer->consecutive) { | |
503 | err = -EINVAL; | |
504 | goto out_put; | |
505 | } | |
506 | ||
507 | if (!buffer->data) { | |
8ceb41d7 | 508 | int fd = perf_data__fd(btsq->bts->session->data); |
d0170af7 AH |
509 | |
510 | buffer->data = auxtrace_buffer__get_data(buffer, fd); | |
511 | if (!buffer->data) { | |
512 | err = -ENOMEM; | |
513 | goto out_put; | |
514 | } | |
515 | } | |
516 | ||
517 | if (btsq->bts->snapshot_mode && !buffer->consecutive && | |
518 | intel_bts_do_fix_overlap(queue, buffer)) { | |
519 | err = -ENOMEM; | |
520 | goto out_put; | |
521 | } | |
522 | ||
50f73637 AH |
523 | if (!btsq->bts->synth_opts.callchain && |
524 | !btsq->bts->synth_opts.thread_stack && thread && | |
d0170af7 AH |
525 | (!old_buffer || btsq->bts->sampling_mode || |
526 | (btsq->bts->snapshot_mode && !buffer->consecutive))) | |
256d92bc | 527 | thread_stack__set_trace_nr(thread, btsq->cpu, buffer->buffer_nr + 1); |
d0170af7 | 528 | |
50f73637 | 529 | err = intel_bts_process_buffer(btsq, buffer, thread); |
d0170af7 AH |
530 | |
531 | auxtrace_buffer__drop_data(buffer); | |
532 | ||
533 | btsq->buffer = auxtrace_buffer__next(queue, buffer); | |
534 | if (btsq->buffer) { | |
535 | if (timestamp) | |
536 | *timestamp = btsq->buffer->reference; | |
537 | } else { | |
538 | if (!btsq->bts->sampling_mode) | |
539 | btsq->done = 1; | |
540 | } | |
541 | out_put: | |
542 | thread__put(thread); | |
543 | return err; | |
544 | } | |
545 | ||
546 | static int intel_bts_flush_queue(struct intel_bts_queue *btsq) | |
547 | { | |
548 | u64 ts = 0; | |
549 | int ret; | |
550 | ||
551 | while (1) { | |
552 | ret = intel_bts_process_queue(btsq, &ts); | |
553 | if (ret < 0) | |
554 | return ret; | |
555 | if (ret) | |
556 | break; | |
557 | } | |
558 | return 0; | |
559 | } | |
560 | ||
561 | static int intel_bts_process_tid_exit(struct intel_bts *bts, pid_t tid) | |
562 | { | |
563 | struct auxtrace_queues *queues = &bts->queues; | |
564 | unsigned int i; | |
565 | ||
566 | for (i = 0; i < queues->nr_queues; i++) { | |
567 | struct auxtrace_queue *queue = &bts->queues.queue_array[i]; | |
568 | struct intel_bts_queue *btsq = queue->priv; | |
569 | ||
570 | if (btsq && btsq->tid == tid) | |
571 | return intel_bts_flush_queue(btsq); | |
572 | } | |
573 | return 0; | |
574 | } | |
575 | ||
576 | static int intel_bts_process_queues(struct intel_bts *bts, u64 timestamp) | |
577 | { | |
578 | while (1) { | |
579 | unsigned int queue_nr; | |
580 | struct auxtrace_queue *queue; | |
581 | struct intel_bts_queue *btsq; | |
582 | u64 ts = 0; | |
583 | int ret; | |
584 | ||
585 | if (!bts->heap.heap_cnt) | |
586 | return 0; | |
587 | ||
588 | if (bts->heap.heap_array[0].ordinal > timestamp) | |
589 | return 0; | |
590 | ||
591 | queue_nr = bts->heap.heap_array[0].queue_nr; | |
592 | queue = &bts->queues.queue_array[queue_nr]; | |
593 | btsq = queue->priv; | |
594 | ||
595 | auxtrace_heap__pop(&bts->heap); | |
596 | ||
597 | ret = intel_bts_process_queue(btsq, &ts); | |
598 | if (ret < 0) { | |
599 | auxtrace_heap__add(&bts->heap, queue_nr, ts); | |
600 | return ret; | |
601 | } | |
602 | ||
603 | if (!ret) { | |
604 | ret = auxtrace_heap__add(&bts->heap, queue_nr, ts); | |
605 | if (ret < 0) | |
606 | return ret; | |
607 | } else { | |
608 | btsq->on_heap = false; | |
609 | } | |
610 | } | |
611 | ||
612 | return 0; | |
613 | } | |
614 | ||
615 | static int intel_bts_process_event(struct perf_session *session, | |
616 | union perf_event *event, | |
617 | struct perf_sample *sample, | |
618 | struct perf_tool *tool) | |
619 | { | |
620 | struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts, | |
621 | auxtrace); | |
622 | u64 timestamp; | |
623 | int err; | |
624 | ||
625 | if (dump_trace) | |
626 | return 0; | |
627 | ||
628 | if (!tool->ordered_events) { | |
629 | pr_err("Intel BTS requires ordered events\n"); | |
630 | return -EINVAL; | |
631 | } | |
632 | ||
633 | if (sample->time && sample->time != (u64)-1) | |
634 | timestamp = perf_time_to_tsc(sample->time, &bts->tc); | |
635 | else | |
636 | timestamp = 0; | |
637 | ||
638 | err = intel_bts_update_queues(bts); | |
639 | if (err) | |
640 | return err; | |
641 | ||
642 | err = intel_bts_process_queues(bts, timestamp); | |
643 | if (err) | |
644 | return err; | |
645 | if (event->header.type == PERF_RECORD_EXIT) { | |
53ff6bc3 | 646 | err = intel_bts_process_tid_exit(bts, event->fork.tid); |
d0170af7 AH |
647 | if (err) |
648 | return err; | |
649 | } | |
650 | ||
651 | if (event->header.type == PERF_RECORD_AUX && | |
652 | (event->aux.flags & PERF_AUX_FLAG_TRUNCATED) && | |
653 | bts->synth_opts.errors) | |
654 | err = intel_bts_lost(bts, sample); | |
655 | ||
656 | return err; | |
657 | } | |
658 | ||
659 | static int intel_bts_process_auxtrace_event(struct perf_session *session, | |
660 | union perf_event *event, | |
661 | struct perf_tool *tool __maybe_unused) | |
662 | { | |
663 | struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts, | |
664 | auxtrace); | |
665 | ||
666 | if (bts->sampling_mode) | |
667 | return 0; | |
668 | ||
669 | if (!bts->data_queued) { | |
670 | struct auxtrace_buffer *buffer; | |
671 | off_t data_offset; | |
8ceb41d7 | 672 | int fd = perf_data__fd(session->data); |
d0170af7 AH |
673 | int err; |
674 | ||
8ceb41d7 | 675 | if (perf_data__is_pipe(session->data)) { |
d0170af7 AH |
676 | data_offset = 0; |
677 | } else { | |
678 | data_offset = lseek(fd, 0, SEEK_CUR); | |
679 | if (data_offset == -1) | |
680 | return -errno; | |
681 | } | |
682 | ||
683 | err = auxtrace_queues__add_event(&bts->queues, session, event, | |
684 | data_offset, &buffer); | |
685 | if (err) | |
686 | return err; | |
687 | ||
688 | /* Dump here now we have copied a piped trace out of the pipe */ | |
689 | if (dump_trace) { | |
690 | if (auxtrace_buffer__get_data(buffer, fd)) { | |
691 | intel_bts_dump_event(bts, buffer->data, | |
692 | buffer->size); | |
693 | auxtrace_buffer__put_data(buffer); | |
694 | } | |
695 | } | |
696 | } | |
697 | ||
698 | return 0; | |
699 | } | |
700 | ||
b8f8eb84 | 701 | static int intel_bts_flush(struct perf_session *session, |
d0170af7 AH |
702 | struct perf_tool *tool __maybe_unused) |
703 | { | |
704 | struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts, | |
705 | auxtrace); | |
706 | int ret; | |
707 | ||
708 | if (dump_trace || bts->sampling_mode) | |
709 | return 0; | |
710 | ||
711 | if (!tool->ordered_events) | |
712 | return -EINVAL; | |
713 | ||
714 | ret = intel_bts_update_queues(bts); | |
715 | if (ret < 0) | |
716 | return ret; | |
717 | ||
718 | return intel_bts_process_queues(bts, MAX_TIMESTAMP); | |
719 | } | |
720 | ||
721 | static void intel_bts_free_queue(void *priv) | |
722 | { | |
723 | struct intel_bts_queue *btsq = priv; | |
724 | ||
725 | if (!btsq) | |
726 | return; | |
727 | free(btsq); | |
728 | } | |
729 | ||
730 | static void intel_bts_free_events(struct perf_session *session) | |
731 | { | |
732 | struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts, | |
733 | auxtrace); | |
734 | struct auxtrace_queues *queues = &bts->queues; | |
735 | unsigned int i; | |
736 | ||
737 | for (i = 0; i < queues->nr_queues; i++) { | |
738 | intel_bts_free_queue(queues->queue_array[i].priv); | |
739 | queues->queue_array[i].priv = NULL; | |
740 | } | |
741 | auxtrace_queues__free(queues); | |
742 | } | |
743 | ||
744 | static void intel_bts_free(struct perf_session *session) | |
745 | { | |
746 | struct intel_bts *bts = container_of(session->auxtrace, struct intel_bts, | |
747 | auxtrace); | |
748 | ||
749 | auxtrace_heap__free(&bts->heap); | |
750 | intel_bts_free_events(session); | |
751 | session->auxtrace = NULL; | |
752 | free(bts); | |
753 | } | |
754 | ||
755 | struct intel_bts_synth { | |
756 | struct perf_tool dummy_tool; | |
757 | struct perf_session *session; | |
758 | }; | |
759 | ||
760 | static int intel_bts_event_synth(struct perf_tool *tool, | |
761 | union perf_event *event, | |
762 | struct perf_sample *sample __maybe_unused, | |
763 | struct machine *machine __maybe_unused) | |
764 | { | |
765 | struct intel_bts_synth *intel_bts_synth = | |
766 | container_of(tool, struct intel_bts_synth, dummy_tool); | |
767 | ||
768 | return perf_session__deliver_synth_event(intel_bts_synth->session, | |
769 | event, NULL); | |
770 | } | |
771 | ||
772 | static int intel_bts_synth_event(struct perf_session *session, | |
773 | struct perf_event_attr *attr, u64 id) | |
774 | { | |
775 | struct intel_bts_synth intel_bts_synth; | |
776 | ||
777 | memset(&intel_bts_synth, 0, sizeof(struct intel_bts_synth)); | |
778 | intel_bts_synth.session = session; | |
779 | ||
780 | return perf_event__synthesize_attr(&intel_bts_synth.dummy_tool, attr, 1, | |
781 | &id, intel_bts_event_synth); | |
782 | } | |
783 | ||
784 | static int intel_bts_synth_events(struct intel_bts *bts, | |
785 | struct perf_session *session) | |
786 | { | |
787 | struct perf_evlist *evlist = session->evlist; | |
788 | struct perf_evsel *evsel; | |
789 | struct perf_event_attr attr; | |
790 | bool found = false; | |
791 | u64 id; | |
792 | int err; | |
793 | ||
e5cadb93 | 794 | evlist__for_each_entry(evlist, evsel) { |
d0170af7 AH |
795 | if (evsel->attr.type == bts->pmu_type && evsel->ids) { |
796 | found = true; | |
797 | break; | |
798 | } | |
799 | } | |
800 | ||
801 | if (!found) { | |
802 | pr_debug("There are no selected events with Intel BTS data\n"); | |
803 | return 0; | |
804 | } | |
805 | ||
806 | memset(&attr, 0, sizeof(struct perf_event_attr)); | |
807 | attr.size = sizeof(struct perf_event_attr); | |
808 | attr.type = PERF_TYPE_HARDWARE; | |
809 | attr.sample_type = evsel->attr.sample_type & PERF_SAMPLE_MASK; | |
810 | attr.sample_type |= PERF_SAMPLE_IP | PERF_SAMPLE_TID | | |
811 | PERF_SAMPLE_PERIOD; | |
812 | attr.sample_type &= ~(u64)PERF_SAMPLE_TIME; | |
813 | attr.sample_type &= ~(u64)PERF_SAMPLE_CPU; | |
814 | attr.exclude_user = evsel->attr.exclude_user; | |
815 | attr.exclude_kernel = evsel->attr.exclude_kernel; | |
816 | attr.exclude_hv = evsel->attr.exclude_hv; | |
817 | attr.exclude_host = evsel->attr.exclude_host; | |
818 | attr.exclude_guest = evsel->attr.exclude_guest; | |
819 | attr.sample_id_all = evsel->attr.sample_id_all; | |
820 | attr.read_format = evsel->attr.read_format; | |
821 | ||
822 | id = evsel->id[0] + 1000000000; | |
823 | if (!id) | |
824 | id = 1; | |
825 | ||
826 | if (bts->synth_opts.branches) { | |
827 | attr.config = PERF_COUNT_HW_BRANCH_INSTRUCTIONS; | |
828 | attr.sample_period = 1; | |
829 | attr.sample_type |= PERF_SAMPLE_ADDR; | |
830 | pr_debug("Synthesizing 'branches' event with id %" PRIu64 " sample type %#" PRIx64 "\n", | |
831 | id, (u64)attr.sample_type); | |
832 | err = intel_bts_synth_event(session, &attr, id); | |
833 | if (err) { | |
834 | pr_err("%s: failed to synthesize 'branches' event type\n", | |
835 | __func__); | |
836 | return err; | |
837 | } | |
838 | bts->sample_branches = true; | |
839 | bts->branches_sample_type = attr.sample_type; | |
840 | bts->branches_id = id; | |
841 | /* | |
842 | * We only use sample types from PERF_SAMPLE_MASK so we can use | |
843 | * __perf_evsel__sample_size() here. | |
844 | */ | |
845 | bts->branches_event_size = sizeof(struct sample_event) + | |
846 | __perf_evsel__sample_size(attr.sample_type); | |
847 | } | |
848 | ||
d0170af7 AH |
849 | return 0; |
850 | } | |
851 | ||
852 | static const char * const intel_bts_info_fmts[] = { | |
853 | [INTEL_BTS_PMU_TYPE] = " PMU Type %"PRId64"\n", | |
854 | [INTEL_BTS_TIME_SHIFT] = " Time Shift %"PRIu64"\n", | |
855 | [INTEL_BTS_TIME_MULT] = " Time Muliplier %"PRIu64"\n", | |
856 | [INTEL_BTS_TIME_ZERO] = " Time Zero %"PRIu64"\n", | |
857 | [INTEL_BTS_CAP_USER_TIME_ZERO] = " Cap Time Zero %"PRId64"\n", | |
858 | [INTEL_BTS_SNAPSHOT_MODE] = " Snapshot mode %"PRId64"\n", | |
859 | }; | |
860 | ||
861 | static void intel_bts_print_info(u64 *arr, int start, int finish) | |
862 | { | |
863 | int i; | |
864 | ||
865 | if (!dump_trace) | |
866 | return; | |
867 | ||
868 | for (i = start; i <= finish; i++) | |
869 | fprintf(stdout, intel_bts_info_fmts[i], arr[i]); | |
870 | } | |
871 | ||
d0170af7 AH |
872 | int intel_bts_process_auxtrace_info(union perf_event *event, |
873 | struct perf_session *session) | |
874 | { | |
875 | struct auxtrace_info_event *auxtrace_info = &event->auxtrace_info; | |
876 | size_t min_sz = sizeof(u64) * INTEL_BTS_SNAPSHOT_MODE; | |
877 | struct intel_bts *bts; | |
878 | int err; | |
879 | ||
880 | if (auxtrace_info->header.size < sizeof(struct auxtrace_info_event) + | |
881 | min_sz) | |
882 | return -EINVAL; | |
883 | ||
884 | bts = zalloc(sizeof(struct intel_bts)); | |
885 | if (!bts) | |
886 | return -ENOMEM; | |
887 | ||
888 | err = auxtrace_queues__init(&bts->queues); | |
889 | if (err) | |
890 | goto err_free; | |
891 | ||
892 | bts->session = session; | |
893 | bts->machine = &session->machines.host; /* No kvm support */ | |
894 | bts->auxtrace_type = auxtrace_info->type; | |
895 | bts->pmu_type = auxtrace_info->priv[INTEL_BTS_PMU_TYPE]; | |
896 | bts->tc.time_shift = auxtrace_info->priv[INTEL_BTS_TIME_SHIFT]; | |
897 | bts->tc.time_mult = auxtrace_info->priv[INTEL_BTS_TIME_MULT]; | |
898 | bts->tc.time_zero = auxtrace_info->priv[INTEL_BTS_TIME_ZERO]; | |
899 | bts->cap_user_time_zero = | |
900 | auxtrace_info->priv[INTEL_BTS_CAP_USER_TIME_ZERO]; | |
901 | bts->snapshot_mode = auxtrace_info->priv[INTEL_BTS_SNAPSHOT_MODE]; | |
902 | ||
903 | bts->sampling_mode = false; | |
904 | ||
905 | bts->auxtrace.process_event = intel_bts_process_event; | |
906 | bts->auxtrace.process_auxtrace_event = intel_bts_process_auxtrace_event; | |
907 | bts->auxtrace.flush_events = intel_bts_flush; | |
908 | bts->auxtrace.free_events = intel_bts_free_events; | |
909 | bts->auxtrace.free = intel_bts_free; | |
910 | session->auxtrace = &bts->auxtrace; | |
911 | ||
912 | intel_bts_print_info(&auxtrace_info->priv[0], INTEL_BTS_PMU_TYPE, | |
913 | INTEL_BTS_SNAPSHOT_MODE); | |
914 | ||
915 | if (dump_trace) | |
916 | return 0; | |
917 | ||
50f73637 | 918 | if (session->itrace_synth_opts && session->itrace_synth_opts->set) { |
d0170af7 | 919 | bts->synth_opts = *session->itrace_synth_opts; |
50f73637 | 920 | } else { |
4eb06815 AK |
921 | itrace_synth_opts__set_default(&bts->synth_opts, |
922 | session->itrace_synth_opts->default_no_sample); | |
50f73637 AH |
923 | if (session->itrace_synth_opts) |
924 | bts->synth_opts.thread_stack = | |
925 | session->itrace_synth_opts->thread_stack; | |
926 | } | |
d0170af7 AH |
927 | |
928 | if (bts->synth_opts.calls) | |
929 | bts->branches_filter |= PERF_IP_FLAG_CALL | PERF_IP_FLAG_ASYNC | | |
930 | PERF_IP_FLAG_TRACE_END; | |
931 | if (bts->synth_opts.returns) | |
932 | bts->branches_filter |= PERF_IP_FLAG_RETURN | | |
933 | PERF_IP_FLAG_TRACE_BEGIN; | |
934 | ||
935 | err = intel_bts_synth_events(bts, session); | |
936 | if (err) | |
937 | goto err_free_queues; | |
938 | ||
939 | err = auxtrace_queues__process_index(&bts->queues, session); | |
940 | if (err) | |
941 | goto err_free_queues; | |
942 | ||
943 | if (bts->queues.populated) | |
944 | bts->data_queued = true; | |
945 | ||
946 | return 0; | |
947 | ||
948 | err_free_queues: | |
949 | auxtrace_queues__free(&bts->queues); | |
950 | session->auxtrace = NULL; | |
951 | err_free: | |
952 | free(bts); | |
953 | return err; | |
954 | } |