Commit | Line | Data |
---|---|---|
90e457f7 AH |
1 | /* |
2 | * intel_pt.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 | ||
fd20e811 | 16 | #include <inttypes.h> |
90e457f7 AH |
17 | #include <stdio.h> |
18 | #include <stdbool.h> | |
19 | #include <errno.h> | |
20 | #include <linux/kernel.h> | |
21 | #include <linux/types.h> | |
22 | ||
23 | #include "../perf.h" | |
24 | #include "session.h" | |
25 | #include "machine.h" | |
98521b38 | 26 | #include "memswap.h" |
f14445ee | 27 | #include "sort.h" |
90e457f7 AH |
28 | #include "tool.h" |
29 | #include "event.h" | |
30 | #include "evlist.h" | |
31 | #include "evsel.h" | |
32 | #include "map.h" | |
33 | #include "color.h" | |
34 | #include "util.h" | |
35 | #include "thread.h" | |
36 | #include "thread-stack.h" | |
37 | #include "symbol.h" | |
38 | #include "callchain.h" | |
39 | #include "dso.h" | |
40 | #include "debug.h" | |
41 | #include "auxtrace.h" | |
42 | #include "tsc.h" | |
43 | #include "intel-pt.h" | |
41840d21 | 44 | #include "config.h" |
90e457f7 AH |
45 | |
46 | #include "intel-pt-decoder/intel-pt-log.h" | |
47 | #include "intel-pt-decoder/intel-pt-decoder.h" | |
48 | #include "intel-pt-decoder/intel-pt-insn-decoder.h" | |
49 | #include "intel-pt-decoder/intel-pt-pkt-decoder.h" | |
50 | ||
51 | #define MAX_TIMESTAMP (~0ULL) | |
52 | ||
53 | struct intel_pt { | |
54 | struct auxtrace auxtrace; | |
55 | struct auxtrace_queues queues; | |
56 | struct auxtrace_heap heap; | |
57 | u32 auxtrace_type; | |
58 | struct perf_session *session; | |
59 | struct machine *machine; | |
60 | struct perf_evsel *switch_evsel; | |
61 | struct thread *unknown_thread; | |
62 | bool timeless_decoding; | |
63 | bool sampling_mode; | |
64 | bool snapshot_mode; | |
65 | bool per_cpu_mmaps; | |
66 | bool have_tsc; | |
67 | bool data_queued; | |
68 | bool est_tsc; | |
69 | bool sync_switch; | |
ba11ba65 | 70 | bool mispred_all; |
90e457f7 AH |
71 | int have_sched_switch; |
72 | u32 pmu_type; | |
73 | u64 kernel_start; | |
74 | u64 switch_ip; | |
75 | u64 ptss_ip; | |
76 | ||
77 | struct perf_tsc_conversion tc; | |
78 | bool cap_user_time_zero; | |
79 | ||
80 | struct itrace_synth_opts synth_opts; | |
81 | ||
82 | bool sample_instructions; | |
83 | u64 instructions_sample_type; | |
90e457f7 AH |
84 | u64 instructions_id; |
85 | ||
86 | bool sample_branches; | |
87 | u32 branches_filter; | |
88 | u64 branches_sample_type; | |
89 | u64 branches_id; | |
90 | ||
91 | bool sample_transactions; | |
92 | u64 transactions_sample_type; | |
93 | u64 transactions_id; | |
94 | ||
37973075 AH |
95 | bool sample_ptwrites; |
96 | u64 ptwrites_sample_type; | |
97 | u64 ptwrites_id; | |
98 | ||
99 | bool sample_pwr_events; | |
100 | u64 pwr_events_sample_type; | |
101 | u64 mwait_id; | |
102 | u64 pwre_id; | |
103 | u64 exstop_id; | |
104 | u64 pwrx_id; | |
105 | u64 cbr_id; | |
106 | ||
90e457f7 AH |
107 | bool synth_needs_swap; |
108 | ||
109 | u64 tsc_bit; | |
11fa7cb8 AH |
110 | u64 mtc_bit; |
111 | u64 mtc_freq_bits; | |
112 | u32 tsc_ctc_ratio_n; | |
113 | u32 tsc_ctc_ratio_d; | |
114 | u64 cyc_bit; | |
90e457f7 AH |
115 | u64 noretcomp_bit; |
116 | unsigned max_non_turbo_ratio; | |
37973075 | 117 | unsigned cbr2khz; |
d1706b39 AK |
118 | |
119 | unsigned long num_events; | |
2b9e32c4 AH |
120 | |
121 | char *filter; | |
2acee108 | 122 | struct addr_filters filts; |
90e457f7 AH |
123 | }; |
124 | ||
125 | enum switch_state { | |
126 | INTEL_PT_SS_NOT_TRACING, | |
127 | INTEL_PT_SS_UNKNOWN, | |
128 | INTEL_PT_SS_TRACING, | |
129 | INTEL_PT_SS_EXPECTING_SWITCH_EVENT, | |
130 | INTEL_PT_SS_EXPECTING_SWITCH_IP, | |
131 | }; | |
132 | ||
133 | struct intel_pt_queue { | |
134 | struct intel_pt *pt; | |
135 | unsigned int queue_nr; | |
136 | struct auxtrace_buffer *buffer; | |
137 | void *decoder; | |
138 | const struct intel_pt_state *state; | |
139 | struct ip_callchain *chain; | |
f14445ee AH |
140 | struct branch_stack *last_branch; |
141 | struct branch_stack *last_branch_rb; | |
142 | size_t last_branch_pos; | |
90e457f7 AH |
143 | union perf_event *event_buf; |
144 | bool on_heap; | |
145 | bool stop; | |
146 | bool step_through_buffers; | |
147 | bool use_buffer_pid_tid; | |
148 | pid_t pid, tid; | |
149 | int cpu; | |
150 | int switch_state; | |
151 | pid_t next_tid; | |
152 | struct thread *thread; | |
153 | bool exclude_kernel; | |
154 | bool have_sample; | |
155 | u64 time; | |
156 | u64 timestamp; | |
157 | u32 flags; | |
158 | u16 insn_len; | |
2a21d036 | 159 | u64 last_insn_cnt; |
faaa8768 | 160 | char insn[INTEL_PT_INSN_BUF_SZ]; |
90e457f7 AH |
161 | }; |
162 | ||
163 | static void intel_pt_dump(struct intel_pt *pt __maybe_unused, | |
164 | unsigned char *buf, size_t len) | |
165 | { | |
166 | struct intel_pt_pkt packet; | |
167 | size_t pos = 0; | |
168 | int ret, pkt_len, i; | |
169 | char desc[INTEL_PT_PKT_DESC_MAX]; | |
170 | const char *color = PERF_COLOR_BLUE; | |
171 | ||
172 | color_fprintf(stdout, color, | |
173 | ". ... Intel Processor Trace data: size %zu bytes\n", | |
174 | len); | |
175 | ||
176 | while (len) { | |
177 | ret = intel_pt_get_packet(buf, len, &packet); | |
178 | if (ret > 0) | |
179 | pkt_len = ret; | |
180 | else | |
181 | pkt_len = 1; | |
182 | printf("."); | |
183 | color_fprintf(stdout, color, " %08x: ", pos); | |
184 | for (i = 0; i < pkt_len; i++) | |
185 | color_fprintf(stdout, color, " %02x", buf[i]); | |
186 | for (; i < 16; i++) | |
187 | color_fprintf(stdout, color, " "); | |
188 | if (ret > 0) { | |
189 | ret = intel_pt_pkt_desc(&packet, desc, | |
190 | INTEL_PT_PKT_DESC_MAX); | |
191 | if (ret > 0) | |
192 | color_fprintf(stdout, color, " %s\n", desc); | |
193 | } else { | |
194 | color_fprintf(stdout, color, " Bad packet!\n"); | |
195 | } | |
196 | pos += pkt_len; | |
197 | buf += pkt_len; | |
198 | len -= pkt_len; | |
199 | } | |
200 | } | |
201 | ||
202 | static void intel_pt_dump_event(struct intel_pt *pt, unsigned char *buf, | |
203 | size_t len) | |
204 | { | |
205 | printf(".\n"); | |
206 | intel_pt_dump(pt, buf, len); | |
207 | } | |
208 | ||
209 | static int intel_pt_do_fix_overlap(struct intel_pt *pt, struct auxtrace_buffer *a, | |
210 | struct auxtrace_buffer *b) | |
211 | { | |
212 | void *start; | |
213 | ||
214 | start = intel_pt_find_overlap(a->data, a->size, b->data, b->size, | |
215 | pt->have_tsc); | |
216 | if (!start) | |
217 | return -EINVAL; | |
218 | b->use_size = b->data + b->size - start; | |
219 | b->use_data = start; | |
220 | return 0; | |
221 | } | |
222 | ||
223 | static void intel_pt_use_buffer_pid_tid(struct intel_pt_queue *ptq, | |
224 | struct auxtrace_queue *queue, | |
225 | struct auxtrace_buffer *buffer) | |
226 | { | |
227 | if (queue->cpu == -1 && buffer->cpu != -1) | |
228 | ptq->cpu = buffer->cpu; | |
229 | ||
230 | ptq->pid = buffer->pid; | |
231 | ptq->tid = buffer->tid; | |
232 | ||
233 | intel_pt_log("queue %u cpu %d pid %d tid %d\n", | |
234 | ptq->queue_nr, ptq->cpu, ptq->pid, ptq->tid); | |
235 | ||
236 | thread__zput(ptq->thread); | |
237 | ||
238 | if (ptq->tid != -1) { | |
239 | if (ptq->pid != -1) | |
240 | ptq->thread = machine__findnew_thread(ptq->pt->machine, | |
241 | ptq->pid, | |
242 | ptq->tid); | |
243 | else | |
244 | ptq->thread = machine__find_thread(ptq->pt->machine, -1, | |
245 | ptq->tid); | |
246 | } | |
247 | } | |
248 | ||
249 | /* This function assumes data is processed sequentially only */ | |
250 | static int intel_pt_get_trace(struct intel_pt_buffer *b, void *data) | |
251 | { | |
252 | struct intel_pt_queue *ptq = data; | |
253 | struct auxtrace_buffer *buffer = ptq->buffer, *old_buffer = buffer; | |
254 | struct auxtrace_queue *queue; | |
255 | ||
256 | if (ptq->stop) { | |
257 | b->len = 0; | |
258 | return 0; | |
259 | } | |
260 | ||
261 | queue = &ptq->pt->queues.queue_array[ptq->queue_nr]; | |
810c398b | 262 | next: |
90e457f7 AH |
263 | buffer = auxtrace_buffer__next(queue, buffer); |
264 | if (!buffer) { | |
265 | if (old_buffer) | |
266 | auxtrace_buffer__drop_data(old_buffer); | |
267 | b->len = 0; | |
268 | return 0; | |
269 | } | |
270 | ||
271 | ptq->buffer = buffer; | |
272 | ||
273 | if (!buffer->data) { | |
8ceb41d7 | 274 | int fd = perf_data__fd(ptq->pt->session->data); |
90e457f7 AH |
275 | |
276 | buffer->data = auxtrace_buffer__get_data(buffer, fd); | |
277 | if (!buffer->data) | |
278 | return -ENOMEM; | |
279 | } | |
280 | ||
281 | if (ptq->pt->snapshot_mode && !buffer->consecutive && old_buffer && | |
282 | intel_pt_do_fix_overlap(ptq->pt, old_buffer, buffer)) | |
283 | return -ENOMEM; | |
284 | ||
90e457f7 AH |
285 | if (buffer->use_data) { |
286 | b->len = buffer->use_size; | |
287 | b->buf = buffer->use_data; | |
288 | } else { | |
289 | b->len = buffer->size; | |
290 | b->buf = buffer->data; | |
291 | } | |
292 | b->ref_timestamp = buffer->reference; | |
293 | ||
810c398b AH |
294 | /* |
295 | * If in snapshot mode and the buffer has no usable data, get next | |
296 | * buffer and again check overlap against old_buffer. | |
297 | */ | |
298 | if (ptq->pt->snapshot_mode && !b->len) | |
299 | goto next; | |
300 | ||
301 | if (old_buffer) | |
302 | auxtrace_buffer__drop_data(old_buffer); | |
303 | ||
90e457f7 AH |
304 | if (!old_buffer || ptq->pt->sampling_mode || (ptq->pt->snapshot_mode && |
305 | !buffer->consecutive)) { | |
306 | b->consecutive = false; | |
307 | b->trace_nr = buffer->buffer_nr + 1; | |
308 | } else { | |
309 | b->consecutive = true; | |
310 | } | |
311 | ||
312 | if (ptq->use_buffer_pid_tid && (ptq->pid != buffer->pid || | |
313 | ptq->tid != buffer->tid)) | |
314 | intel_pt_use_buffer_pid_tid(ptq, queue, buffer); | |
315 | ||
316 | if (ptq->step_through_buffers) | |
317 | ptq->stop = true; | |
318 | ||
319 | if (!b->len) | |
320 | return intel_pt_get_trace(b, data); | |
321 | ||
322 | return 0; | |
323 | } | |
324 | ||
325 | struct intel_pt_cache_entry { | |
326 | struct auxtrace_cache_entry entry; | |
327 | u64 insn_cnt; | |
328 | u64 byte_cnt; | |
329 | enum intel_pt_insn_op op; | |
330 | enum intel_pt_insn_branch branch; | |
331 | int length; | |
332 | int32_t rel; | |
faaa8768 | 333 | char insn[INTEL_PT_INSN_BUF_SZ]; |
90e457f7 AH |
334 | }; |
335 | ||
336 | static int intel_pt_config_div(const char *var, const char *value, void *data) | |
337 | { | |
338 | int *d = data; | |
339 | long val; | |
340 | ||
341 | if (!strcmp(var, "intel-pt.cache-divisor")) { | |
342 | val = strtol(value, NULL, 0); | |
343 | if (val > 0 && val <= INT_MAX) | |
344 | *d = val; | |
345 | } | |
346 | ||
347 | return 0; | |
348 | } | |
349 | ||
350 | static int intel_pt_cache_divisor(void) | |
351 | { | |
352 | static int d; | |
353 | ||
354 | if (d) | |
355 | return d; | |
356 | ||
357 | perf_config(intel_pt_config_div, &d); | |
358 | ||
359 | if (!d) | |
360 | d = 64; | |
361 | ||
362 | return d; | |
363 | } | |
364 | ||
365 | static unsigned int intel_pt_cache_size(struct dso *dso, | |
366 | struct machine *machine) | |
367 | { | |
368 | off_t size; | |
369 | ||
370 | size = dso__data_size(dso, machine); | |
371 | size /= intel_pt_cache_divisor(); | |
372 | if (size < 1000) | |
373 | return 10; | |
374 | if (size > (1 << 21)) | |
375 | return 21; | |
376 | return 32 - __builtin_clz(size); | |
377 | } | |
378 | ||
379 | static struct auxtrace_cache *intel_pt_cache(struct dso *dso, | |
380 | struct machine *machine) | |
381 | { | |
382 | struct auxtrace_cache *c; | |
383 | unsigned int bits; | |
384 | ||
385 | if (dso->auxtrace_cache) | |
386 | return dso->auxtrace_cache; | |
387 | ||
388 | bits = intel_pt_cache_size(dso, machine); | |
389 | ||
390 | /* Ignoring cache creation failure */ | |
391 | c = auxtrace_cache__new(bits, sizeof(struct intel_pt_cache_entry), 200); | |
392 | ||
393 | dso->auxtrace_cache = c; | |
394 | ||
395 | return c; | |
396 | } | |
397 | ||
398 | static int intel_pt_cache_add(struct dso *dso, struct machine *machine, | |
399 | u64 offset, u64 insn_cnt, u64 byte_cnt, | |
400 | struct intel_pt_insn *intel_pt_insn) | |
401 | { | |
402 | struct auxtrace_cache *c = intel_pt_cache(dso, machine); | |
403 | struct intel_pt_cache_entry *e; | |
404 | int err; | |
405 | ||
406 | if (!c) | |
407 | return -ENOMEM; | |
408 | ||
409 | e = auxtrace_cache__alloc_entry(c); | |
410 | if (!e) | |
411 | return -ENOMEM; | |
412 | ||
413 | e->insn_cnt = insn_cnt; | |
414 | e->byte_cnt = byte_cnt; | |
415 | e->op = intel_pt_insn->op; | |
416 | e->branch = intel_pt_insn->branch; | |
417 | e->length = intel_pt_insn->length; | |
418 | e->rel = intel_pt_insn->rel; | |
faaa8768 | 419 | memcpy(e->insn, intel_pt_insn->buf, INTEL_PT_INSN_BUF_SZ); |
90e457f7 AH |
420 | |
421 | err = auxtrace_cache__add(c, offset, &e->entry); | |
422 | if (err) | |
423 | auxtrace_cache__free_entry(c, e); | |
424 | ||
425 | return err; | |
426 | } | |
427 | ||
428 | static struct intel_pt_cache_entry * | |
429 | intel_pt_cache_lookup(struct dso *dso, struct machine *machine, u64 offset) | |
430 | { | |
431 | struct auxtrace_cache *c = intel_pt_cache(dso, machine); | |
432 | ||
433 | if (!c) | |
434 | return NULL; | |
435 | ||
436 | return auxtrace_cache__lookup(dso->auxtrace_cache, offset); | |
437 | } | |
438 | ||
439 | static int intel_pt_walk_next_insn(struct intel_pt_insn *intel_pt_insn, | |
440 | uint64_t *insn_cnt_ptr, uint64_t *ip, | |
441 | uint64_t to_ip, uint64_t max_insn_cnt, | |
442 | void *data) | |
443 | { | |
444 | struct intel_pt_queue *ptq = data; | |
445 | struct machine *machine = ptq->pt->machine; | |
446 | struct thread *thread; | |
447 | struct addr_location al; | |
32f98aab | 448 | unsigned char buf[INTEL_PT_INSN_BUF_SZ]; |
90e457f7 AH |
449 | ssize_t len; |
450 | int x86_64; | |
451 | u8 cpumode; | |
452 | u64 offset, start_offset, start_ip; | |
453 | u64 insn_cnt = 0; | |
454 | bool one_map = true; | |
455 | ||
faaa8768 AK |
456 | intel_pt_insn->length = 0; |
457 | ||
90e457f7 AH |
458 | if (to_ip && *ip == to_ip) |
459 | goto out_no_cache; | |
460 | ||
90e457f7 AH |
461 | if (*ip >= ptq->pt->kernel_start) |
462 | cpumode = PERF_RECORD_MISC_KERNEL; | |
463 | else | |
464 | cpumode = PERF_RECORD_MISC_USER; | |
465 | ||
466 | thread = ptq->thread; | |
467 | if (!thread) { | |
468 | if (cpumode != PERF_RECORD_MISC_KERNEL) | |
469 | return -EINVAL; | |
470 | thread = ptq->pt->unknown_thread; | |
471 | } | |
472 | ||
473 | while (1) { | |
474 | thread__find_addr_map(thread, cpumode, MAP__FUNCTION, *ip, &al); | |
475 | if (!al.map || !al.map->dso) | |
476 | return -EINVAL; | |
477 | ||
478 | if (al.map->dso->data.status == DSO_DATA_STATUS_ERROR && | |
479 | dso__data_status_seen(al.map->dso, | |
480 | DSO_DATA_STATUS_SEEN_ITRACE)) | |
481 | return -ENOENT; | |
482 | ||
483 | offset = al.map->map_ip(al.map, *ip); | |
484 | ||
485 | if (!to_ip && one_map) { | |
486 | struct intel_pt_cache_entry *e; | |
487 | ||
488 | e = intel_pt_cache_lookup(al.map->dso, machine, offset); | |
489 | if (e && | |
490 | (!max_insn_cnt || e->insn_cnt <= max_insn_cnt)) { | |
491 | *insn_cnt_ptr = e->insn_cnt; | |
492 | *ip += e->byte_cnt; | |
493 | intel_pt_insn->op = e->op; | |
494 | intel_pt_insn->branch = e->branch; | |
495 | intel_pt_insn->length = e->length; | |
496 | intel_pt_insn->rel = e->rel; | |
faaa8768 AK |
497 | memcpy(intel_pt_insn->buf, e->insn, |
498 | INTEL_PT_INSN_BUF_SZ); | |
90e457f7 AH |
499 | intel_pt_log_insn_no_data(intel_pt_insn, *ip); |
500 | return 0; | |
501 | } | |
502 | } | |
503 | ||
504 | start_offset = offset; | |
505 | start_ip = *ip; | |
506 | ||
507 | /* Load maps to ensure dso->is_64_bit has been updated */ | |
be39db9f | 508 | map__load(al.map); |
90e457f7 AH |
509 | |
510 | x86_64 = al.map->dso->is_64_bit; | |
511 | ||
512 | while (1) { | |
513 | len = dso__data_read_offset(al.map->dso, machine, | |
32f98aab AH |
514 | offset, buf, |
515 | INTEL_PT_INSN_BUF_SZ); | |
90e457f7 AH |
516 | if (len <= 0) |
517 | return -EINVAL; | |
518 | ||
519 | if (intel_pt_get_insn(buf, len, x86_64, intel_pt_insn)) | |
520 | return -EINVAL; | |
521 | ||
522 | intel_pt_log_insn(intel_pt_insn, *ip); | |
523 | ||
524 | insn_cnt += 1; | |
525 | ||
526 | if (intel_pt_insn->branch != INTEL_PT_BR_NO_BRANCH) | |
527 | goto out; | |
528 | ||
529 | if (max_insn_cnt && insn_cnt >= max_insn_cnt) | |
530 | goto out_no_cache; | |
531 | ||
532 | *ip += intel_pt_insn->length; | |
533 | ||
534 | if (to_ip && *ip == to_ip) | |
535 | goto out_no_cache; | |
536 | ||
537 | if (*ip >= al.map->end) | |
538 | break; | |
539 | ||
540 | offset += intel_pt_insn->length; | |
541 | } | |
542 | one_map = false; | |
543 | } | |
544 | out: | |
545 | *insn_cnt_ptr = insn_cnt; | |
546 | ||
547 | if (!one_map) | |
548 | goto out_no_cache; | |
549 | ||
550 | /* | |
551 | * Didn't lookup in the 'to_ip' case, so do it now to prevent duplicate | |
552 | * entries. | |
553 | */ | |
554 | if (to_ip) { | |
555 | struct intel_pt_cache_entry *e; | |
556 | ||
557 | e = intel_pt_cache_lookup(al.map->dso, machine, start_offset); | |
558 | if (e) | |
559 | return 0; | |
560 | } | |
561 | ||
562 | /* Ignore cache errors */ | |
563 | intel_pt_cache_add(al.map->dso, machine, start_offset, insn_cnt, | |
564 | *ip - start_ip, intel_pt_insn); | |
565 | ||
566 | return 0; | |
567 | ||
568 | out_no_cache: | |
569 | *insn_cnt_ptr = insn_cnt; | |
570 | return 0; | |
571 | } | |
572 | ||
2acee108 AH |
573 | static bool intel_pt_match_pgd_ip(struct intel_pt *pt, uint64_t ip, |
574 | uint64_t offset, const char *filename) | |
575 | { | |
576 | struct addr_filter *filt; | |
577 | bool have_filter = false; | |
578 | bool hit_tracestop = false; | |
579 | bool hit_filter = false; | |
580 | ||
581 | list_for_each_entry(filt, &pt->filts.head, list) { | |
582 | if (filt->start) | |
583 | have_filter = true; | |
584 | ||
585 | if ((filename && !filt->filename) || | |
586 | (!filename && filt->filename) || | |
587 | (filename && strcmp(filename, filt->filename))) | |
588 | continue; | |
589 | ||
590 | if (!(offset >= filt->addr && offset < filt->addr + filt->size)) | |
591 | continue; | |
592 | ||
593 | intel_pt_log("TIP.PGD ip %#"PRIx64" offset %#"PRIx64" in %s hit filter: %s offset %#"PRIx64" size %#"PRIx64"\n", | |
594 | ip, offset, filename ? filename : "[kernel]", | |
595 | filt->start ? "filter" : "stop", | |
596 | filt->addr, filt->size); | |
597 | ||
598 | if (filt->start) | |
599 | hit_filter = true; | |
600 | else | |
601 | hit_tracestop = true; | |
602 | } | |
603 | ||
604 | if (!hit_tracestop && !hit_filter) | |
605 | intel_pt_log("TIP.PGD ip %#"PRIx64" offset %#"PRIx64" in %s is not in a filter region\n", | |
606 | ip, offset, filename ? filename : "[kernel]"); | |
607 | ||
608 | return hit_tracestop || (have_filter && !hit_filter); | |
609 | } | |
610 | ||
611 | static int __intel_pt_pgd_ip(uint64_t ip, void *data) | |
612 | { | |
613 | struct intel_pt_queue *ptq = data; | |
614 | struct thread *thread; | |
615 | struct addr_location al; | |
616 | u8 cpumode; | |
617 | u64 offset; | |
618 | ||
619 | if (ip >= ptq->pt->kernel_start) | |
620 | return intel_pt_match_pgd_ip(ptq->pt, ip, ip, NULL); | |
621 | ||
622 | cpumode = PERF_RECORD_MISC_USER; | |
623 | ||
624 | thread = ptq->thread; | |
625 | if (!thread) | |
626 | return -EINVAL; | |
627 | ||
628 | thread__find_addr_map(thread, cpumode, MAP__FUNCTION, ip, &al); | |
629 | if (!al.map || !al.map->dso) | |
630 | return -EINVAL; | |
631 | ||
632 | offset = al.map->map_ip(al.map, ip); | |
633 | ||
634 | return intel_pt_match_pgd_ip(ptq->pt, ip, offset, | |
635 | al.map->dso->long_name); | |
636 | } | |
637 | ||
638 | static bool intel_pt_pgd_ip(uint64_t ip, void *data) | |
639 | { | |
640 | return __intel_pt_pgd_ip(ip, data) > 0; | |
641 | } | |
642 | ||
90e457f7 AH |
643 | static bool intel_pt_get_config(struct intel_pt *pt, |
644 | struct perf_event_attr *attr, u64 *config) | |
645 | { | |
646 | if (attr->type == pt->pmu_type) { | |
647 | if (config) | |
648 | *config = attr->config; | |
649 | return true; | |
650 | } | |
651 | ||
652 | return false; | |
653 | } | |
654 | ||
655 | static bool intel_pt_exclude_kernel(struct intel_pt *pt) | |
656 | { | |
657 | struct perf_evsel *evsel; | |
658 | ||
e5cadb93 | 659 | evlist__for_each_entry(pt->session->evlist, evsel) { |
90e457f7 AH |
660 | if (intel_pt_get_config(pt, &evsel->attr, NULL) && |
661 | !evsel->attr.exclude_kernel) | |
662 | return false; | |
663 | } | |
664 | return true; | |
665 | } | |
666 | ||
667 | static bool intel_pt_return_compression(struct intel_pt *pt) | |
668 | { | |
669 | struct perf_evsel *evsel; | |
670 | u64 config; | |
671 | ||
672 | if (!pt->noretcomp_bit) | |
673 | return true; | |
674 | ||
e5cadb93 | 675 | evlist__for_each_entry(pt->session->evlist, evsel) { |
90e457f7 AH |
676 | if (intel_pt_get_config(pt, &evsel->attr, &config) && |
677 | (config & pt->noretcomp_bit)) | |
678 | return false; | |
679 | } | |
680 | return true; | |
681 | } | |
682 | ||
83959817 AH |
683 | static bool intel_pt_branch_enable(struct intel_pt *pt) |
684 | { | |
685 | struct perf_evsel *evsel; | |
686 | u64 config; | |
687 | ||
688 | evlist__for_each_entry(pt->session->evlist, evsel) { | |
689 | if (intel_pt_get_config(pt, &evsel->attr, &config) && | |
690 | (config & 1) && !(config & 0x2000)) | |
691 | return false; | |
692 | } | |
693 | return true; | |
694 | } | |
695 | ||
11fa7cb8 AH |
696 | static unsigned int intel_pt_mtc_period(struct intel_pt *pt) |
697 | { | |
698 | struct perf_evsel *evsel; | |
699 | unsigned int shift; | |
700 | u64 config; | |
701 | ||
702 | if (!pt->mtc_freq_bits) | |
703 | return 0; | |
704 | ||
705 | for (shift = 0, config = pt->mtc_freq_bits; !(config & 1); shift++) | |
706 | config >>= 1; | |
707 | ||
e5cadb93 | 708 | evlist__for_each_entry(pt->session->evlist, evsel) { |
11fa7cb8 AH |
709 | if (intel_pt_get_config(pt, &evsel->attr, &config)) |
710 | return (config & pt->mtc_freq_bits) >> shift; | |
711 | } | |
712 | return 0; | |
713 | } | |
714 | ||
90e457f7 AH |
715 | static bool intel_pt_timeless_decoding(struct intel_pt *pt) |
716 | { | |
717 | struct perf_evsel *evsel; | |
718 | bool timeless_decoding = true; | |
719 | u64 config; | |
720 | ||
721 | if (!pt->tsc_bit || !pt->cap_user_time_zero) | |
722 | return true; | |
723 | ||
e5cadb93 | 724 | evlist__for_each_entry(pt->session->evlist, evsel) { |
90e457f7 AH |
725 | if (!(evsel->attr.sample_type & PERF_SAMPLE_TIME)) |
726 | return true; | |
727 | if (intel_pt_get_config(pt, &evsel->attr, &config)) { | |
728 | if (config & pt->tsc_bit) | |
729 | timeless_decoding = false; | |
730 | else | |
731 | return true; | |
732 | } | |
733 | } | |
734 | return timeless_decoding; | |
735 | } | |
736 | ||
737 | static bool intel_pt_tracing_kernel(struct intel_pt *pt) | |
738 | { | |
739 | struct perf_evsel *evsel; | |
740 | ||
e5cadb93 | 741 | evlist__for_each_entry(pt->session->evlist, evsel) { |
90e457f7 AH |
742 | if (intel_pt_get_config(pt, &evsel->attr, NULL) && |
743 | !evsel->attr.exclude_kernel) | |
744 | return true; | |
745 | } | |
746 | return false; | |
747 | } | |
748 | ||
749 | static bool intel_pt_have_tsc(struct intel_pt *pt) | |
750 | { | |
751 | struct perf_evsel *evsel; | |
752 | bool have_tsc = false; | |
753 | u64 config; | |
754 | ||
755 | if (!pt->tsc_bit) | |
756 | return false; | |
757 | ||
e5cadb93 | 758 | evlist__for_each_entry(pt->session->evlist, evsel) { |
90e457f7 AH |
759 | if (intel_pt_get_config(pt, &evsel->attr, &config)) { |
760 | if (config & pt->tsc_bit) | |
761 | have_tsc = true; | |
762 | else | |
763 | return false; | |
764 | } | |
765 | } | |
766 | return have_tsc; | |
767 | } | |
768 | ||
769 | static u64 intel_pt_ns_to_ticks(const struct intel_pt *pt, u64 ns) | |
770 | { | |
771 | u64 quot, rem; | |
772 | ||
773 | quot = ns / pt->tc.time_mult; | |
774 | rem = ns % pt->tc.time_mult; | |
775 | return (quot << pt->tc.time_shift) + (rem << pt->tc.time_shift) / | |
776 | pt->tc.time_mult; | |
777 | } | |
778 | ||
779 | static struct intel_pt_queue *intel_pt_alloc_queue(struct intel_pt *pt, | |
780 | unsigned int queue_nr) | |
781 | { | |
782 | struct intel_pt_params params = { .get_trace = 0, }; | |
783 | struct intel_pt_queue *ptq; | |
784 | ||
785 | ptq = zalloc(sizeof(struct intel_pt_queue)); | |
786 | if (!ptq) | |
787 | return NULL; | |
788 | ||
789 | if (pt->synth_opts.callchain) { | |
790 | size_t sz = sizeof(struct ip_callchain); | |
791 | ||
792 | sz += pt->synth_opts.callchain_sz * sizeof(u64); | |
793 | ptq->chain = zalloc(sz); | |
794 | if (!ptq->chain) | |
795 | goto out_free; | |
796 | } | |
797 | ||
f14445ee AH |
798 | if (pt->synth_opts.last_branch) { |
799 | size_t sz = sizeof(struct branch_stack); | |
800 | ||
801 | sz += pt->synth_opts.last_branch_sz * | |
802 | sizeof(struct branch_entry); | |
803 | ptq->last_branch = zalloc(sz); | |
804 | if (!ptq->last_branch) | |
805 | goto out_free; | |
806 | ptq->last_branch_rb = zalloc(sz); | |
807 | if (!ptq->last_branch_rb) | |
808 | goto out_free; | |
809 | } | |
810 | ||
90e457f7 AH |
811 | ptq->event_buf = malloc(PERF_SAMPLE_MAX_SIZE); |
812 | if (!ptq->event_buf) | |
813 | goto out_free; | |
814 | ||
815 | ptq->pt = pt; | |
816 | ptq->queue_nr = queue_nr; | |
817 | ptq->exclude_kernel = intel_pt_exclude_kernel(pt); | |
818 | ptq->pid = -1; | |
819 | ptq->tid = -1; | |
820 | ptq->cpu = -1; | |
821 | ptq->next_tid = -1; | |
822 | ||
823 | params.get_trace = intel_pt_get_trace; | |
824 | params.walk_insn = intel_pt_walk_next_insn; | |
825 | params.data = ptq; | |
826 | params.return_compression = intel_pt_return_compression(pt); | |
83959817 | 827 | params.branch_enable = intel_pt_branch_enable(pt); |
90e457f7 | 828 | params.max_non_turbo_ratio = pt->max_non_turbo_ratio; |
11fa7cb8 AH |
829 | params.mtc_period = intel_pt_mtc_period(pt); |
830 | params.tsc_ctc_ratio_n = pt->tsc_ctc_ratio_n; | |
831 | params.tsc_ctc_ratio_d = pt->tsc_ctc_ratio_d; | |
90e457f7 | 832 | |
2acee108 AH |
833 | if (pt->filts.cnt > 0) |
834 | params.pgd_ip = intel_pt_pgd_ip; | |
835 | ||
90e457f7 AH |
836 | if (pt->synth_opts.instructions) { |
837 | if (pt->synth_opts.period) { | |
838 | switch (pt->synth_opts.period_type) { | |
839 | case PERF_ITRACE_PERIOD_INSTRUCTIONS: | |
840 | params.period_type = | |
841 | INTEL_PT_PERIOD_INSTRUCTIONS; | |
842 | params.period = pt->synth_opts.period; | |
843 | break; | |
844 | case PERF_ITRACE_PERIOD_TICKS: | |
845 | params.period_type = INTEL_PT_PERIOD_TICKS; | |
846 | params.period = pt->synth_opts.period; | |
847 | break; | |
848 | case PERF_ITRACE_PERIOD_NANOSECS: | |
849 | params.period_type = INTEL_PT_PERIOD_TICKS; | |
850 | params.period = intel_pt_ns_to_ticks(pt, | |
851 | pt->synth_opts.period); | |
852 | break; | |
853 | default: | |
854 | break; | |
855 | } | |
856 | } | |
857 | ||
858 | if (!params.period) { | |
859 | params.period_type = INTEL_PT_PERIOD_INSTRUCTIONS; | |
e1791347 | 860 | params.period = 1; |
90e457f7 AH |
861 | } |
862 | } | |
863 | ||
864 | ptq->decoder = intel_pt_decoder_new(¶ms); | |
865 | if (!ptq->decoder) | |
866 | goto out_free; | |
867 | ||
868 | return ptq; | |
869 | ||
870 | out_free: | |
871 | zfree(&ptq->event_buf); | |
f14445ee AH |
872 | zfree(&ptq->last_branch); |
873 | zfree(&ptq->last_branch_rb); | |
90e457f7 AH |
874 | zfree(&ptq->chain); |
875 | free(ptq); | |
876 | return NULL; | |
877 | } | |
878 | ||
879 | static void intel_pt_free_queue(void *priv) | |
880 | { | |
881 | struct intel_pt_queue *ptq = priv; | |
882 | ||
883 | if (!ptq) | |
884 | return; | |
885 | thread__zput(ptq->thread); | |
886 | intel_pt_decoder_free(ptq->decoder); | |
887 | zfree(&ptq->event_buf); | |
f14445ee AH |
888 | zfree(&ptq->last_branch); |
889 | zfree(&ptq->last_branch_rb); | |
90e457f7 AH |
890 | zfree(&ptq->chain); |
891 | free(ptq); | |
892 | } | |
893 | ||
894 | static void intel_pt_set_pid_tid_cpu(struct intel_pt *pt, | |
895 | struct auxtrace_queue *queue) | |
896 | { | |
897 | struct intel_pt_queue *ptq = queue->priv; | |
898 | ||
899 | if (queue->tid == -1 || pt->have_sched_switch) { | |
900 | ptq->tid = machine__get_current_tid(pt->machine, ptq->cpu); | |
901 | thread__zput(ptq->thread); | |
902 | } | |
903 | ||
904 | if (!ptq->thread && ptq->tid != -1) | |
905 | ptq->thread = machine__find_thread(pt->machine, -1, ptq->tid); | |
906 | ||
907 | if (ptq->thread) { | |
908 | ptq->pid = ptq->thread->pid_; | |
909 | if (queue->cpu == -1) | |
910 | ptq->cpu = ptq->thread->cpu; | |
911 | } | |
912 | } | |
913 | ||
914 | static void intel_pt_sample_flags(struct intel_pt_queue *ptq) | |
915 | { | |
916 | if (ptq->state->flags & INTEL_PT_ABORT_TX) { | |
917 | ptq->flags = PERF_IP_FLAG_BRANCH | PERF_IP_FLAG_TX_ABORT; | |
918 | } else if (ptq->state->flags & INTEL_PT_ASYNC) { | |
919 | if (ptq->state->to_ip) | |
920 | ptq->flags = PERF_IP_FLAG_BRANCH | PERF_IP_FLAG_CALL | | |
921 | PERF_IP_FLAG_ASYNC | | |
922 | PERF_IP_FLAG_INTERRUPT; | |
923 | else | |
924 | ptq->flags = PERF_IP_FLAG_BRANCH | | |
925 | PERF_IP_FLAG_TRACE_END; | |
926 | ptq->insn_len = 0; | |
927 | } else { | |
928 | if (ptq->state->from_ip) | |
929 | ptq->flags = intel_pt_insn_type(ptq->state->insn_op); | |
930 | else | |
931 | ptq->flags = PERF_IP_FLAG_BRANCH | | |
932 | PERF_IP_FLAG_TRACE_BEGIN; | |
933 | if (ptq->state->flags & INTEL_PT_IN_TX) | |
934 | ptq->flags |= PERF_IP_FLAG_IN_TX; | |
935 | ptq->insn_len = ptq->state->insn_len; | |
faaa8768 | 936 | memcpy(ptq->insn, ptq->state->insn, INTEL_PT_INSN_BUF_SZ); |
90e457f7 AH |
937 | } |
938 | } | |
939 | ||
940 | static int intel_pt_setup_queue(struct intel_pt *pt, | |
941 | struct auxtrace_queue *queue, | |
942 | unsigned int queue_nr) | |
943 | { | |
944 | struct intel_pt_queue *ptq = queue->priv; | |
945 | ||
946 | if (list_empty(&queue->head)) | |
947 | return 0; | |
948 | ||
949 | if (!ptq) { | |
950 | ptq = intel_pt_alloc_queue(pt, queue_nr); | |
951 | if (!ptq) | |
952 | return -ENOMEM; | |
953 | queue->priv = ptq; | |
954 | ||
955 | if (queue->cpu != -1) | |
956 | ptq->cpu = queue->cpu; | |
957 | ptq->tid = queue->tid; | |
958 | ||
959 | if (pt->sampling_mode) { | |
960 | if (pt->timeless_decoding) | |
961 | ptq->step_through_buffers = true; | |
962 | if (pt->timeless_decoding || !pt->have_sched_switch) | |
963 | ptq->use_buffer_pid_tid = true; | |
964 | } | |
965 | } | |
966 | ||
967 | if (!ptq->on_heap && | |
968 | (!pt->sync_switch || | |
969 | ptq->switch_state != INTEL_PT_SS_EXPECTING_SWITCH_EVENT)) { | |
970 | const struct intel_pt_state *state; | |
971 | int ret; | |
972 | ||
973 | if (pt->timeless_decoding) | |
974 | return 0; | |
975 | ||
976 | intel_pt_log("queue %u getting timestamp\n", queue_nr); | |
977 | intel_pt_log("queue %u decoding cpu %d pid %d tid %d\n", | |
978 | queue_nr, ptq->cpu, ptq->pid, ptq->tid); | |
979 | while (1) { | |
980 | state = intel_pt_decode(ptq->decoder); | |
981 | if (state->err) { | |
982 | if (state->err == INTEL_PT_ERR_NODATA) { | |
983 | intel_pt_log("queue %u has no timestamp\n", | |
984 | queue_nr); | |
985 | return 0; | |
986 | } | |
987 | continue; | |
988 | } | |
989 | if (state->timestamp) | |
990 | break; | |
991 | } | |
992 | ||
993 | ptq->timestamp = state->timestamp; | |
994 | intel_pt_log("queue %u timestamp 0x%" PRIx64 "\n", | |
995 | queue_nr, ptq->timestamp); | |
996 | ptq->state = state; | |
997 | ptq->have_sample = true; | |
998 | intel_pt_sample_flags(ptq); | |
999 | ret = auxtrace_heap__add(&pt->heap, queue_nr, ptq->timestamp); | |
1000 | if (ret) | |
1001 | return ret; | |
1002 | ptq->on_heap = true; | |
1003 | } | |
1004 | ||
1005 | return 0; | |
1006 | } | |
1007 | ||
1008 | static int intel_pt_setup_queues(struct intel_pt *pt) | |
1009 | { | |
1010 | unsigned int i; | |
1011 | int ret; | |
1012 | ||
1013 | for (i = 0; i < pt->queues.nr_queues; i++) { | |
1014 | ret = intel_pt_setup_queue(pt, &pt->queues.queue_array[i], i); | |
1015 | if (ret) | |
1016 | return ret; | |
1017 | } | |
1018 | return 0; | |
1019 | } | |
1020 | ||
f14445ee AH |
1021 | static inline void intel_pt_copy_last_branch_rb(struct intel_pt_queue *ptq) |
1022 | { | |
1023 | struct branch_stack *bs_src = ptq->last_branch_rb; | |
1024 | struct branch_stack *bs_dst = ptq->last_branch; | |
1025 | size_t nr = 0; | |
1026 | ||
1027 | bs_dst->nr = bs_src->nr; | |
1028 | ||
1029 | if (!bs_src->nr) | |
1030 | return; | |
1031 | ||
1032 | nr = ptq->pt->synth_opts.last_branch_sz - ptq->last_branch_pos; | |
1033 | memcpy(&bs_dst->entries[0], | |
1034 | &bs_src->entries[ptq->last_branch_pos], | |
1035 | sizeof(struct branch_entry) * nr); | |
1036 | ||
1037 | if (bs_src->nr >= ptq->pt->synth_opts.last_branch_sz) { | |
1038 | memcpy(&bs_dst->entries[nr], | |
1039 | &bs_src->entries[0], | |
1040 | sizeof(struct branch_entry) * ptq->last_branch_pos); | |
1041 | } | |
1042 | } | |
1043 | ||
1044 | static inline void intel_pt_reset_last_branch_rb(struct intel_pt_queue *ptq) | |
1045 | { | |
1046 | ptq->last_branch_pos = 0; | |
1047 | ptq->last_branch_rb->nr = 0; | |
1048 | } | |
1049 | ||
1050 | static void intel_pt_update_last_branch_rb(struct intel_pt_queue *ptq) | |
1051 | { | |
1052 | const struct intel_pt_state *state = ptq->state; | |
1053 | struct branch_stack *bs = ptq->last_branch_rb; | |
1054 | struct branch_entry *be; | |
1055 | ||
1056 | if (!ptq->last_branch_pos) | |
1057 | ptq->last_branch_pos = ptq->pt->synth_opts.last_branch_sz; | |
1058 | ||
1059 | ptq->last_branch_pos -= 1; | |
1060 | ||
1061 | be = &bs->entries[ptq->last_branch_pos]; | |
1062 | be->from = state->from_ip; | |
1063 | be->to = state->to_ip; | |
1064 | be->flags.abort = !!(state->flags & INTEL_PT_ABORT_TX); | |
1065 | be->flags.in_tx = !!(state->flags & INTEL_PT_IN_TX); | |
1066 | /* No support for mispredict */ | |
ba11ba65 | 1067 | be->flags.mispred = ptq->pt->mispred_all; |
f14445ee AH |
1068 | |
1069 | if (bs->nr < ptq->pt->synth_opts.last_branch_sz) | |
1070 | bs->nr += 1; | |
1071 | } | |
1072 | ||
0f3e5379 AH |
1073 | static inline bool intel_pt_skip_event(struct intel_pt *pt) |
1074 | { | |
1075 | return pt->synth_opts.initial_skip && | |
1076 | pt->num_events++ < pt->synth_opts.initial_skip; | |
1077 | } | |
1078 | ||
1079 | static void intel_pt_prep_b_sample(struct intel_pt *pt, | |
1080 | struct intel_pt_queue *ptq, | |
1081 | union perf_event *event, | |
1082 | struct perf_sample *sample) | |
1083 | { | |
1084 | event->sample.header.type = PERF_RECORD_SAMPLE; | |
1085 | event->sample.header.misc = PERF_RECORD_MISC_USER; | |
1086 | event->sample.header.size = sizeof(struct perf_event_header); | |
1087 | ||
1088 | if (!pt->timeless_decoding) | |
1089 | sample->time = tsc_to_perf_time(ptq->timestamp, &pt->tc); | |
1090 | ||
1091 | sample->cpumode = PERF_RECORD_MISC_USER; | |
1092 | sample->ip = ptq->state->from_ip; | |
1093 | sample->pid = ptq->pid; | |
1094 | sample->tid = ptq->tid; | |
1095 | sample->addr = ptq->state->to_ip; | |
1096 | sample->period = 1; | |
1097 | sample->cpu = ptq->cpu; | |
1098 | sample->flags = ptq->flags; | |
1099 | sample->insn_len = ptq->insn_len; | |
1100 | memcpy(sample->insn, ptq->insn, INTEL_PT_INSN_BUF_SZ); | |
1101 | } | |
1102 | ||
90e457f7 AH |
1103 | static int intel_pt_inject_event(union perf_event *event, |
1104 | struct perf_sample *sample, u64 type, | |
1105 | bool swapped) | |
1106 | { | |
1107 | event->header.size = perf_event__sample_event_size(sample, type, 0); | |
1108 | return perf_event__synthesize_sample(event, type, 0, sample, swapped); | |
1109 | } | |
1110 | ||
0f3e5379 AH |
1111 | static inline int intel_pt_opt_inject(struct intel_pt *pt, |
1112 | union perf_event *event, | |
1113 | struct perf_sample *sample, u64 type) | |
1114 | { | |
1115 | if (!pt->synth_opts.inject) | |
1116 | return 0; | |
1117 | ||
1118 | return intel_pt_inject_event(event, sample, type, pt->synth_needs_swap); | |
1119 | } | |
1120 | ||
1121 | static int intel_pt_deliver_synth_b_event(struct intel_pt *pt, | |
1122 | union perf_event *event, | |
1123 | struct perf_sample *sample, u64 type) | |
90e457f7 AH |
1124 | { |
1125 | int ret; | |
0f3e5379 AH |
1126 | |
1127 | ret = intel_pt_opt_inject(pt, event, sample, type); | |
1128 | if (ret) | |
1129 | return ret; | |
1130 | ||
1131 | ret = perf_session__deliver_synth_event(pt->session, event, sample); | |
1132 | if (ret) | |
1133 | pr_err("Intel PT: failed to deliver event, error %d\n", ret); | |
1134 | ||
1135 | return ret; | |
1136 | } | |
1137 | ||
1138 | static int intel_pt_synth_branch_sample(struct intel_pt_queue *ptq) | |
1139 | { | |
90e457f7 AH |
1140 | struct intel_pt *pt = ptq->pt; |
1141 | union perf_event *event = ptq->event_buf; | |
1142 | struct perf_sample sample = { .ip = 0, }; | |
f14445ee AH |
1143 | struct dummy_branch_stack { |
1144 | u64 nr; | |
1145 | struct branch_entry entries; | |
1146 | } dummy_bs; | |
90e457f7 | 1147 | |
385e3306 AH |
1148 | if (pt->branches_filter && !(pt->branches_filter & ptq->flags)) |
1149 | return 0; | |
1150 | ||
0f3e5379 | 1151 | if (intel_pt_skip_event(pt)) |
d1706b39 AK |
1152 | return 0; |
1153 | ||
0f3e5379 | 1154 | intel_pt_prep_b_sample(pt, ptq, event, &sample); |
90e457f7 | 1155 | |
90e457f7 AH |
1156 | sample.id = ptq->pt->branches_id; |
1157 | sample.stream_id = ptq->pt->branches_id; | |
90e457f7 | 1158 | |
f14445ee AH |
1159 | /* |
1160 | * perf report cannot handle events without a branch stack when using | |
1161 | * SORT_MODE__BRANCH so make a dummy one. | |
1162 | */ | |
1163 | if (pt->synth_opts.last_branch && sort__mode == SORT_MODE__BRANCH) { | |
1164 | dummy_bs = (struct dummy_branch_stack){ | |
1165 | .nr = 1, | |
1166 | .entries = { | |
1167 | .from = sample.ip, | |
1168 | .to = sample.addr, | |
1169 | }, | |
1170 | }; | |
1171 | sample.branch_stack = (struct branch_stack *)&dummy_bs; | |
1172 | } | |
1173 | ||
0f3e5379 AH |
1174 | return intel_pt_deliver_synth_b_event(pt, event, &sample, |
1175 | pt->branches_sample_type); | |
90e457f7 AH |
1176 | } |
1177 | ||
0f3e5379 AH |
1178 | static void intel_pt_prep_sample(struct intel_pt *pt, |
1179 | struct intel_pt_queue *ptq, | |
1180 | union perf_event *event, | |
1181 | struct perf_sample *sample) | |
90e457f7 | 1182 | { |
0f3e5379 | 1183 | intel_pt_prep_b_sample(pt, ptq, event, sample); |
2a21d036 | 1184 | |
90e457f7 AH |
1185 | if (pt->synth_opts.callchain) { |
1186 | thread_stack__sample(ptq->thread, ptq->chain, | |
0f3e5379 AH |
1187 | pt->synth_opts.callchain_sz, sample->ip); |
1188 | sample->callchain = ptq->chain; | |
90e457f7 AH |
1189 | } |
1190 | ||
f14445ee AH |
1191 | if (pt->synth_opts.last_branch) { |
1192 | intel_pt_copy_last_branch_rb(ptq); | |
0f3e5379 | 1193 | sample->branch_stack = ptq->last_branch; |
f14445ee | 1194 | } |
0f3e5379 | 1195 | } |
f14445ee | 1196 | |
0f3e5379 AH |
1197 | static inline int intel_pt_deliver_synth_event(struct intel_pt *pt, |
1198 | struct intel_pt_queue *ptq, | |
1199 | union perf_event *event, | |
1200 | struct perf_sample *sample, | |
1201 | u64 type) | |
1202 | { | |
1203 | int ret; | |
90e457f7 | 1204 | |
0f3e5379 | 1205 | ret = intel_pt_deliver_synth_b_event(pt, event, sample, type); |
90e457f7 | 1206 | |
f14445ee AH |
1207 | if (pt->synth_opts.last_branch) |
1208 | intel_pt_reset_last_branch_rb(ptq); | |
1209 | ||
90e457f7 AH |
1210 | return ret; |
1211 | } | |
1212 | ||
0f3e5379 | 1213 | static int intel_pt_synth_instruction_sample(struct intel_pt_queue *ptq) |
90e457f7 | 1214 | { |
90e457f7 AH |
1215 | struct intel_pt *pt = ptq->pt; |
1216 | union perf_event *event = ptq->event_buf; | |
1217 | struct perf_sample sample = { .ip = 0, }; | |
1218 | ||
0f3e5379 | 1219 | if (intel_pt_skip_event(pt)) |
d1706b39 AK |
1220 | return 0; |
1221 | ||
0f3e5379 | 1222 | intel_pt_prep_sample(pt, ptq, event, &sample); |
90e457f7 | 1223 | |
0f3e5379 AH |
1224 | sample.id = ptq->pt->instructions_id; |
1225 | sample.stream_id = ptq->pt->instructions_id; | |
1226 | sample.period = ptq->state->tot_insn_cnt - ptq->last_insn_cnt; | |
90e457f7 | 1227 | |
0f3e5379 | 1228 | ptq->last_insn_cnt = ptq->state->tot_insn_cnt; |
90e457f7 | 1229 | |
0f3e5379 AH |
1230 | return intel_pt_deliver_synth_event(pt, ptq, event, &sample, |
1231 | pt->instructions_sample_type); | |
1232 | } | |
90e457f7 | 1233 | |
0f3e5379 AH |
1234 | static int intel_pt_synth_transaction_sample(struct intel_pt_queue *ptq) |
1235 | { | |
1236 | struct intel_pt *pt = ptq->pt; | |
1237 | union perf_event *event = ptq->event_buf; | |
1238 | struct perf_sample sample = { .ip = 0, }; | |
f14445ee | 1239 | |
0f3e5379 AH |
1240 | if (intel_pt_skip_event(pt)) |
1241 | return 0; | |
90e457f7 | 1242 | |
0f3e5379 | 1243 | intel_pt_prep_sample(pt, ptq, event, &sample); |
90e457f7 | 1244 | |
0f3e5379 AH |
1245 | sample.id = ptq->pt->transactions_id; |
1246 | sample.stream_id = ptq->pt->transactions_id; | |
f14445ee | 1247 | |
0f3e5379 AH |
1248 | return intel_pt_deliver_synth_event(pt, ptq, event, &sample, |
1249 | pt->transactions_sample_type); | |
90e457f7 AH |
1250 | } |
1251 | ||
37973075 AH |
1252 | static void intel_pt_prep_p_sample(struct intel_pt *pt, |
1253 | struct intel_pt_queue *ptq, | |
1254 | union perf_event *event, | |
1255 | struct perf_sample *sample) | |
1256 | { | |
1257 | intel_pt_prep_sample(pt, ptq, event, sample); | |
1258 | ||
1259 | /* | |
1260 | * Zero IP is used to mean "trace start" but that is not the case for | |
1261 | * power or PTWRITE events with no IP, so clear the flags. | |
1262 | */ | |
1263 | if (!sample->ip) | |
1264 | sample->flags = 0; | |
1265 | } | |
1266 | ||
1267 | static int intel_pt_synth_ptwrite_sample(struct intel_pt_queue *ptq) | |
1268 | { | |
1269 | struct intel_pt *pt = ptq->pt; | |
1270 | union perf_event *event = ptq->event_buf; | |
1271 | struct perf_sample sample = { .ip = 0, }; | |
1272 | struct perf_synth_intel_ptwrite raw; | |
1273 | ||
1274 | if (intel_pt_skip_event(pt)) | |
1275 | return 0; | |
1276 | ||
1277 | intel_pt_prep_p_sample(pt, ptq, event, &sample); | |
1278 | ||
1279 | sample.id = ptq->pt->ptwrites_id; | |
1280 | sample.stream_id = ptq->pt->ptwrites_id; | |
1281 | ||
1282 | raw.flags = 0; | |
1283 | raw.ip = !!(ptq->state->flags & INTEL_PT_FUP_IP); | |
1284 | raw.payload = cpu_to_le64(ptq->state->ptw_payload); | |
1285 | ||
1286 | sample.raw_size = perf_synth__raw_size(raw); | |
1287 | sample.raw_data = perf_synth__raw_data(&raw); | |
1288 | ||
1289 | return intel_pt_deliver_synth_event(pt, ptq, event, &sample, | |
1290 | pt->ptwrites_sample_type); | |
1291 | } | |
1292 | ||
1293 | static int intel_pt_synth_cbr_sample(struct intel_pt_queue *ptq) | |
1294 | { | |
1295 | struct intel_pt *pt = ptq->pt; | |
1296 | union perf_event *event = ptq->event_buf; | |
1297 | struct perf_sample sample = { .ip = 0, }; | |
1298 | struct perf_synth_intel_cbr raw; | |
1299 | u32 flags; | |
1300 | ||
1301 | if (intel_pt_skip_event(pt)) | |
1302 | return 0; | |
1303 | ||
1304 | intel_pt_prep_p_sample(pt, ptq, event, &sample); | |
1305 | ||
1306 | sample.id = ptq->pt->cbr_id; | |
1307 | sample.stream_id = ptq->pt->cbr_id; | |
1308 | ||
1309 | flags = (u16)ptq->state->cbr_payload | (pt->max_non_turbo_ratio << 16); | |
1310 | raw.flags = cpu_to_le32(flags); | |
1311 | raw.freq = cpu_to_le32(raw.cbr * pt->cbr2khz); | |
1312 | raw.reserved3 = 0; | |
1313 | ||
1314 | sample.raw_size = perf_synth__raw_size(raw); | |
1315 | sample.raw_data = perf_synth__raw_data(&raw); | |
1316 | ||
1317 | return intel_pt_deliver_synth_event(pt, ptq, event, &sample, | |
1318 | pt->pwr_events_sample_type); | |
1319 | } | |
1320 | ||
1321 | static int intel_pt_synth_mwait_sample(struct intel_pt_queue *ptq) | |
1322 | { | |
1323 | struct intel_pt *pt = ptq->pt; | |
1324 | union perf_event *event = ptq->event_buf; | |
1325 | struct perf_sample sample = { .ip = 0, }; | |
1326 | struct perf_synth_intel_mwait raw; | |
1327 | ||
1328 | if (intel_pt_skip_event(pt)) | |
1329 | return 0; | |
1330 | ||
1331 | intel_pt_prep_p_sample(pt, ptq, event, &sample); | |
1332 | ||
1333 | sample.id = ptq->pt->mwait_id; | |
1334 | sample.stream_id = ptq->pt->mwait_id; | |
1335 | ||
1336 | raw.reserved = 0; | |
1337 | raw.payload = cpu_to_le64(ptq->state->mwait_payload); | |
1338 | ||
1339 | sample.raw_size = perf_synth__raw_size(raw); | |
1340 | sample.raw_data = perf_synth__raw_data(&raw); | |
1341 | ||
1342 | return intel_pt_deliver_synth_event(pt, ptq, event, &sample, | |
1343 | pt->pwr_events_sample_type); | |
1344 | } | |
1345 | ||
1346 | static int intel_pt_synth_pwre_sample(struct intel_pt_queue *ptq) | |
1347 | { | |
1348 | struct intel_pt *pt = ptq->pt; | |
1349 | union perf_event *event = ptq->event_buf; | |
1350 | struct perf_sample sample = { .ip = 0, }; | |
1351 | struct perf_synth_intel_pwre raw; | |
1352 | ||
1353 | if (intel_pt_skip_event(pt)) | |
1354 | return 0; | |
1355 | ||
1356 | intel_pt_prep_p_sample(pt, ptq, event, &sample); | |
1357 | ||
1358 | sample.id = ptq->pt->pwre_id; | |
1359 | sample.stream_id = ptq->pt->pwre_id; | |
1360 | ||
1361 | raw.reserved = 0; | |
1362 | raw.payload = cpu_to_le64(ptq->state->pwre_payload); | |
1363 | ||
1364 | sample.raw_size = perf_synth__raw_size(raw); | |
1365 | sample.raw_data = perf_synth__raw_data(&raw); | |
1366 | ||
1367 | return intel_pt_deliver_synth_event(pt, ptq, event, &sample, | |
1368 | pt->pwr_events_sample_type); | |
1369 | } | |
1370 | ||
1371 | static int intel_pt_synth_exstop_sample(struct intel_pt_queue *ptq) | |
1372 | { | |
1373 | struct intel_pt *pt = ptq->pt; | |
1374 | union perf_event *event = ptq->event_buf; | |
1375 | struct perf_sample sample = { .ip = 0, }; | |
1376 | struct perf_synth_intel_exstop raw; | |
1377 | ||
1378 | if (intel_pt_skip_event(pt)) | |
1379 | return 0; | |
1380 | ||
1381 | intel_pt_prep_p_sample(pt, ptq, event, &sample); | |
1382 | ||
1383 | sample.id = ptq->pt->exstop_id; | |
1384 | sample.stream_id = ptq->pt->exstop_id; | |
1385 | ||
1386 | raw.flags = 0; | |
1387 | raw.ip = !!(ptq->state->flags & INTEL_PT_FUP_IP); | |
1388 | ||
1389 | sample.raw_size = perf_synth__raw_size(raw); | |
1390 | sample.raw_data = perf_synth__raw_data(&raw); | |
1391 | ||
1392 | return intel_pt_deliver_synth_event(pt, ptq, event, &sample, | |
1393 | pt->pwr_events_sample_type); | |
1394 | } | |
1395 | ||
1396 | static int intel_pt_synth_pwrx_sample(struct intel_pt_queue *ptq) | |
1397 | { | |
1398 | struct intel_pt *pt = ptq->pt; | |
1399 | union perf_event *event = ptq->event_buf; | |
1400 | struct perf_sample sample = { .ip = 0, }; | |
1401 | struct perf_synth_intel_pwrx raw; | |
1402 | ||
1403 | if (intel_pt_skip_event(pt)) | |
1404 | return 0; | |
1405 | ||
1406 | intel_pt_prep_p_sample(pt, ptq, event, &sample); | |
1407 | ||
1408 | sample.id = ptq->pt->pwrx_id; | |
1409 | sample.stream_id = ptq->pt->pwrx_id; | |
1410 | ||
1411 | raw.reserved = 0; | |
1412 | raw.payload = cpu_to_le64(ptq->state->pwrx_payload); | |
1413 | ||
1414 | sample.raw_size = perf_synth__raw_size(raw); | |
1415 | sample.raw_data = perf_synth__raw_data(&raw); | |
1416 | ||
1417 | return intel_pt_deliver_synth_event(pt, ptq, event, &sample, | |
1418 | pt->pwr_events_sample_type); | |
1419 | } | |
1420 | ||
90e457f7 AH |
1421 | static int intel_pt_synth_error(struct intel_pt *pt, int code, int cpu, |
1422 | pid_t pid, pid_t tid, u64 ip) | |
1423 | { | |
1424 | union perf_event event; | |
1425 | char msg[MAX_AUXTRACE_ERROR_MSG]; | |
1426 | int err; | |
1427 | ||
1428 | intel_pt__strerror(code, msg, MAX_AUXTRACE_ERROR_MSG); | |
1429 | ||
1430 | auxtrace_synth_error(&event.auxtrace_error, PERF_AUXTRACE_ERROR_ITRACE, | |
1431 | code, cpu, pid, tid, ip, msg); | |
1432 | ||
1433 | err = perf_session__deliver_synth_event(pt->session, &event, NULL); | |
1434 | if (err) | |
1435 | pr_err("Intel Processor Trace: failed to deliver error event, error %d\n", | |
1436 | err); | |
1437 | ||
1438 | return err; | |
1439 | } | |
1440 | ||
1441 | static int intel_pt_next_tid(struct intel_pt *pt, struct intel_pt_queue *ptq) | |
1442 | { | |
1443 | struct auxtrace_queue *queue; | |
1444 | pid_t tid = ptq->next_tid; | |
1445 | int err; | |
1446 | ||
1447 | if (tid == -1) | |
1448 | return 0; | |
1449 | ||
1450 | intel_pt_log("switch: cpu %d tid %d\n", ptq->cpu, tid); | |
1451 | ||
1452 | err = machine__set_current_tid(pt->machine, ptq->cpu, -1, tid); | |
1453 | ||
1454 | queue = &pt->queues.queue_array[ptq->queue_nr]; | |
1455 | intel_pt_set_pid_tid_cpu(pt, queue); | |
1456 | ||
1457 | ptq->next_tid = -1; | |
1458 | ||
1459 | return err; | |
1460 | } | |
1461 | ||
1462 | static inline bool intel_pt_is_switch_ip(struct intel_pt_queue *ptq, u64 ip) | |
1463 | { | |
1464 | struct intel_pt *pt = ptq->pt; | |
1465 | ||
1466 | return ip == pt->switch_ip && | |
1467 | (ptq->flags & PERF_IP_FLAG_BRANCH) && | |
1468 | !(ptq->flags & (PERF_IP_FLAG_CONDITIONAL | PERF_IP_FLAG_ASYNC | | |
1469 | PERF_IP_FLAG_INTERRUPT | PERF_IP_FLAG_TX_ABORT)); | |
1470 | } | |
1471 | ||
37973075 AH |
1472 | #define INTEL_PT_PWR_EVT (INTEL_PT_MWAIT_OP | INTEL_PT_PWR_ENTRY | \ |
1473 | INTEL_PT_EX_STOP | INTEL_PT_PWR_EXIT | \ | |
1474 | INTEL_PT_CBR_CHG) | |
1475 | ||
90e457f7 AH |
1476 | static int intel_pt_sample(struct intel_pt_queue *ptq) |
1477 | { | |
1478 | const struct intel_pt_state *state = ptq->state; | |
1479 | struct intel_pt *pt = ptq->pt; | |
1480 | int err; | |
1481 | ||
1482 | if (!ptq->have_sample) | |
1483 | return 0; | |
1484 | ||
1485 | ptq->have_sample = false; | |
1486 | ||
37973075 AH |
1487 | if (pt->sample_pwr_events && (state->type & INTEL_PT_PWR_EVT)) { |
1488 | if (state->type & INTEL_PT_CBR_CHG) { | |
1489 | err = intel_pt_synth_cbr_sample(ptq); | |
1490 | if (err) | |
1491 | return err; | |
1492 | } | |
1493 | if (state->type & INTEL_PT_MWAIT_OP) { | |
1494 | err = intel_pt_synth_mwait_sample(ptq); | |
1495 | if (err) | |
1496 | return err; | |
1497 | } | |
1498 | if (state->type & INTEL_PT_PWR_ENTRY) { | |
1499 | err = intel_pt_synth_pwre_sample(ptq); | |
1500 | if (err) | |
1501 | return err; | |
1502 | } | |
1503 | if (state->type & INTEL_PT_EX_STOP) { | |
1504 | err = intel_pt_synth_exstop_sample(ptq); | |
1505 | if (err) | |
1506 | return err; | |
1507 | } | |
1508 | if (state->type & INTEL_PT_PWR_EXIT) { | |
1509 | err = intel_pt_synth_pwrx_sample(ptq); | |
1510 | if (err) | |
1511 | return err; | |
1512 | } | |
1513 | } | |
1514 | ||
406a1805 | 1515 | if (pt->sample_instructions && (state->type & INTEL_PT_INSTRUCTION)) { |
90e457f7 AH |
1516 | err = intel_pt_synth_instruction_sample(ptq); |
1517 | if (err) | |
1518 | return err; | |
1519 | } | |
1520 | ||
406a1805 | 1521 | if (pt->sample_transactions && (state->type & INTEL_PT_TRANSACTION)) { |
90e457f7 AH |
1522 | err = intel_pt_synth_transaction_sample(ptq); |
1523 | if (err) | |
1524 | return err; | |
1525 | } | |
1526 | ||
37973075 AH |
1527 | if (pt->sample_ptwrites && (state->type & INTEL_PT_PTW)) { |
1528 | err = intel_pt_synth_ptwrite_sample(ptq); | |
1529 | if (err) | |
1530 | return err; | |
1531 | } | |
1532 | ||
90e457f7 AH |
1533 | if (!(state->type & INTEL_PT_BRANCH)) |
1534 | return 0; | |
1535 | ||
50f73637 | 1536 | if (pt->synth_opts.callchain || pt->synth_opts.thread_stack) |
90e457f7 AH |
1537 | thread_stack__event(ptq->thread, ptq->flags, state->from_ip, |
1538 | state->to_ip, ptq->insn_len, | |
1539 | state->trace_nr); | |
1540 | else | |
1541 | thread_stack__set_trace_nr(ptq->thread, state->trace_nr); | |
1542 | ||
1543 | if (pt->sample_branches) { | |
1544 | err = intel_pt_synth_branch_sample(ptq); | |
1545 | if (err) | |
1546 | return err; | |
1547 | } | |
1548 | ||
f14445ee AH |
1549 | if (pt->synth_opts.last_branch) |
1550 | intel_pt_update_last_branch_rb(ptq); | |
1551 | ||
90e457f7 AH |
1552 | if (!pt->sync_switch) |
1553 | return 0; | |
1554 | ||
1555 | if (intel_pt_is_switch_ip(ptq, state->to_ip)) { | |
1556 | switch (ptq->switch_state) { | |
1557 | case INTEL_PT_SS_UNKNOWN: | |
1558 | case INTEL_PT_SS_EXPECTING_SWITCH_IP: | |
1559 | err = intel_pt_next_tid(pt, ptq); | |
1560 | if (err) | |
1561 | return err; | |
1562 | ptq->switch_state = INTEL_PT_SS_TRACING; | |
1563 | break; | |
1564 | default: | |
1565 | ptq->switch_state = INTEL_PT_SS_EXPECTING_SWITCH_EVENT; | |
1566 | return 1; | |
1567 | } | |
1568 | } else if (!state->to_ip) { | |
1569 | ptq->switch_state = INTEL_PT_SS_NOT_TRACING; | |
1570 | } else if (ptq->switch_state == INTEL_PT_SS_NOT_TRACING) { | |
1571 | ptq->switch_state = INTEL_PT_SS_UNKNOWN; | |
1572 | } else if (ptq->switch_state == INTEL_PT_SS_UNKNOWN && | |
1573 | state->to_ip == pt->ptss_ip && | |
1574 | (ptq->flags & PERF_IP_FLAG_CALL)) { | |
1575 | ptq->switch_state = INTEL_PT_SS_TRACING; | |
1576 | } | |
1577 | ||
1578 | return 0; | |
1579 | } | |
1580 | ||
86c27869 | 1581 | static u64 intel_pt_switch_ip(struct intel_pt *pt, u64 *ptss_ip) |
90e457f7 | 1582 | { |
86c27869 | 1583 | struct machine *machine = pt->machine; |
90e457f7 AH |
1584 | struct map *map; |
1585 | struct symbol *sym, *start; | |
1586 | u64 ip, switch_ip = 0; | |
86c27869 | 1587 | const char *ptss; |
90e457f7 AH |
1588 | |
1589 | if (ptss_ip) | |
1590 | *ptss_ip = 0; | |
1591 | ||
a5e813c6 | 1592 | map = machine__kernel_map(machine); |
90e457f7 AH |
1593 | if (!map) |
1594 | return 0; | |
1595 | ||
be39db9f | 1596 | if (map__load(map)) |
90e457f7 AH |
1597 | return 0; |
1598 | ||
1599 | start = dso__first_symbol(map->dso, MAP__FUNCTION); | |
1600 | ||
1601 | for (sym = start; sym; sym = dso__next_symbol(sym)) { | |
1602 | if (sym->binding == STB_GLOBAL && | |
1603 | !strcmp(sym->name, "__switch_to")) { | |
1604 | ip = map->unmap_ip(map, sym->start); | |
1605 | if (ip >= map->start && ip < map->end) { | |
1606 | switch_ip = ip; | |
1607 | break; | |
1608 | } | |
1609 | } | |
1610 | } | |
1611 | ||
1612 | if (!switch_ip || !ptss_ip) | |
1613 | return 0; | |
1614 | ||
86c27869 AH |
1615 | if (pt->have_sched_switch == 1) |
1616 | ptss = "perf_trace_sched_switch"; | |
1617 | else | |
1618 | ptss = "__perf_event_task_sched_out"; | |
1619 | ||
90e457f7 | 1620 | for (sym = start; sym; sym = dso__next_symbol(sym)) { |
86c27869 | 1621 | if (!strcmp(sym->name, ptss)) { |
90e457f7 AH |
1622 | ip = map->unmap_ip(map, sym->start); |
1623 | if (ip >= map->start && ip < map->end) { | |
1624 | *ptss_ip = ip; | |
1625 | break; | |
1626 | } | |
1627 | } | |
1628 | } | |
1629 | ||
1630 | return switch_ip; | |
1631 | } | |
1632 | ||
1633 | static int intel_pt_run_decoder(struct intel_pt_queue *ptq, u64 *timestamp) | |
1634 | { | |
1635 | const struct intel_pt_state *state = ptq->state; | |
1636 | struct intel_pt *pt = ptq->pt; | |
1637 | int err; | |
1638 | ||
1639 | if (!pt->kernel_start) { | |
1640 | pt->kernel_start = machine__kernel_start(pt->machine); | |
86c27869 AH |
1641 | if (pt->per_cpu_mmaps && |
1642 | (pt->have_sched_switch == 1 || pt->have_sched_switch == 3) && | |
90e457f7 AH |
1643 | !pt->timeless_decoding && intel_pt_tracing_kernel(pt) && |
1644 | !pt->sampling_mode) { | |
86c27869 | 1645 | pt->switch_ip = intel_pt_switch_ip(pt, &pt->ptss_ip); |
90e457f7 AH |
1646 | if (pt->switch_ip) { |
1647 | intel_pt_log("switch_ip: %"PRIx64" ptss_ip: %"PRIx64"\n", | |
1648 | pt->switch_ip, pt->ptss_ip); | |
1649 | pt->sync_switch = true; | |
1650 | } | |
1651 | } | |
1652 | } | |
1653 | ||
1654 | intel_pt_log("queue %u decoding cpu %d pid %d tid %d\n", | |
1655 | ptq->queue_nr, ptq->cpu, ptq->pid, ptq->tid); | |
1656 | while (1) { | |
1657 | err = intel_pt_sample(ptq); | |
1658 | if (err) | |
1659 | return err; | |
1660 | ||
1661 | state = intel_pt_decode(ptq->decoder); | |
1662 | if (state->err) { | |
1663 | if (state->err == INTEL_PT_ERR_NODATA) | |
1664 | return 1; | |
1665 | if (pt->sync_switch && | |
1666 | state->from_ip >= pt->kernel_start) { | |
1667 | pt->sync_switch = false; | |
1668 | intel_pt_next_tid(pt, ptq); | |
1669 | } | |
1670 | if (pt->synth_opts.errors) { | |
1671 | err = intel_pt_synth_error(pt, state->err, | |
1672 | ptq->cpu, ptq->pid, | |
1673 | ptq->tid, | |
1674 | state->from_ip); | |
1675 | if (err) | |
1676 | return err; | |
1677 | } | |
1678 | continue; | |
1679 | } | |
1680 | ||
1681 | ptq->state = state; | |
1682 | ptq->have_sample = true; | |
1683 | intel_pt_sample_flags(ptq); | |
1684 | ||
1685 | /* Use estimated TSC upon return to user space */ | |
1686 | if (pt->est_tsc && | |
1687 | (state->from_ip >= pt->kernel_start || !state->from_ip) && | |
1688 | state->to_ip && state->to_ip < pt->kernel_start) { | |
1689 | intel_pt_log("TSC %"PRIx64" est. TSC %"PRIx64"\n", | |
1690 | state->timestamp, state->est_timestamp); | |
1691 | ptq->timestamp = state->est_timestamp; | |
1692 | /* Use estimated TSC in unknown switch state */ | |
1693 | } else if (pt->sync_switch && | |
1694 | ptq->switch_state == INTEL_PT_SS_UNKNOWN && | |
1695 | intel_pt_is_switch_ip(ptq, state->to_ip) && | |
1696 | ptq->next_tid == -1) { | |
1697 | intel_pt_log("TSC %"PRIx64" est. TSC %"PRIx64"\n", | |
1698 | state->timestamp, state->est_timestamp); | |
1699 | ptq->timestamp = state->est_timestamp; | |
1700 | } else if (state->timestamp > ptq->timestamp) { | |
1701 | ptq->timestamp = state->timestamp; | |
1702 | } | |
1703 | ||
1704 | if (!pt->timeless_decoding && ptq->timestamp >= *timestamp) { | |
1705 | *timestamp = ptq->timestamp; | |
1706 | return 0; | |
1707 | } | |
1708 | } | |
1709 | return 0; | |
1710 | } | |
1711 | ||
1712 | static inline int intel_pt_update_queues(struct intel_pt *pt) | |
1713 | { | |
1714 | if (pt->queues.new_data) { | |
1715 | pt->queues.new_data = false; | |
1716 | return intel_pt_setup_queues(pt); | |
1717 | } | |
1718 | return 0; | |
1719 | } | |
1720 | ||
1721 | static int intel_pt_process_queues(struct intel_pt *pt, u64 timestamp) | |
1722 | { | |
1723 | unsigned int queue_nr; | |
1724 | u64 ts; | |
1725 | int ret; | |
1726 | ||
1727 | while (1) { | |
1728 | struct auxtrace_queue *queue; | |
1729 | struct intel_pt_queue *ptq; | |
1730 | ||
1731 | if (!pt->heap.heap_cnt) | |
1732 | return 0; | |
1733 | ||
1734 | if (pt->heap.heap_array[0].ordinal >= timestamp) | |
1735 | return 0; | |
1736 | ||
1737 | queue_nr = pt->heap.heap_array[0].queue_nr; | |
1738 | queue = &pt->queues.queue_array[queue_nr]; | |
1739 | ptq = queue->priv; | |
1740 | ||
1741 | intel_pt_log("queue %u processing 0x%" PRIx64 " to 0x%" PRIx64 "\n", | |
1742 | queue_nr, pt->heap.heap_array[0].ordinal, | |
1743 | timestamp); | |
1744 | ||
1745 | auxtrace_heap__pop(&pt->heap); | |
1746 | ||
1747 | if (pt->heap.heap_cnt) { | |
1748 | ts = pt->heap.heap_array[0].ordinal + 1; | |
1749 | if (ts > timestamp) | |
1750 | ts = timestamp; | |
1751 | } else { | |
1752 | ts = timestamp; | |
1753 | } | |
1754 | ||
1755 | intel_pt_set_pid_tid_cpu(pt, queue); | |
1756 | ||
1757 | ret = intel_pt_run_decoder(ptq, &ts); | |
1758 | ||
1759 | if (ret < 0) { | |
1760 | auxtrace_heap__add(&pt->heap, queue_nr, ts); | |
1761 | return ret; | |
1762 | } | |
1763 | ||
1764 | if (!ret) { | |
1765 | ret = auxtrace_heap__add(&pt->heap, queue_nr, ts); | |
1766 | if (ret < 0) | |
1767 | return ret; | |
1768 | } else { | |
1769 | ptq->on_heap = false; | |
1770 | } | |
1771 | } | |
1772 | ||
1773 | return 0; | |
1774 | } | |
1775 | ||
1776 | static int intel_pt_process_timeless_queues(struct intel_pt *pt, pid_t tid, | |
1777 | u64 time_) | |
1778 | { | |
1779 | struct auxtrace_queues *queues = &pt->queues; | |
1780 | unsigned int i; | |
1781 | u64 ts = 0; | |
1782 | ||
1783 | for (i = 0; i < queues->nr_queues; i++) { | |
1784 | struct auxtrace_queue *queue = &pt->queues.queue_array[i]; | |
1785 | struct intel_pt_queue *ptq = queue->priv; | |
1786 | ||
1787 | if (ptq && (tid == -1 || ptq->tid == tid)) { | |
1788 | ptq->time = time_; | |
1789 | intel_pt_set_pid_tid_cpu(pt, queue); | |
1790 | intel_pt_run_decoder(ptq, &ts); | |
1791 | } | |
1792 | } | |
1793 | return 0; | |
1794 | } | |
1795 | ||
1796 | static int intel_pt_lost(struct intel_pt *pt, struct perf_sample *sample) | |
1797 | { | |
1798 | return intel_pt_synth_error(pt, INTEL_PT_ERR_LOST, sample->cpu, | |
1799 | sample->pid, sample->tid, 0); | |
1800 | } | |
1801 | ||
1802 | static struct intel_pt_queue *intel_pt_cpu_to_ptq(struct intel_pt *pt, int cpu) | |
1803 | { | |
1804 | unsigned i, j; | |
1805 | ||
1806 | if (cpu < 0 || !pt->queues.nr_queues) | |
1807 | return NULL; | |
1808 | ||
1809 | if ((unsigned)cpu >= pt->queues.nr_queues) | |
1810 | i = pt->queues.nr_queues - 1; | |
1811 | else | |
1812 | i = cpu; | |
1813 | ||
1814 | if (pt->queues.queue_array[i].cpu == cpu) | |
1815 | return pt->queues.queue_array[i].priv; | |
1816 | ||
1817 | for (j = 0; i > 0; j++) { | |
1818 | if (pt->queues.queue_array[--i].cpu == cpu) | |
1819 | return pt->queues.queue_array[i].priv; | |
1820 | } | |
1821 | ||
1822 | for (; j < pt->queues.nr_queues; j++) { | |
1823 | if (pt->queues.queue_array[j].cpu == cpu) | |
1824 | return pt->queues.queue_array[j].priv; | |
1825 | } | |
1826 | ||
1827 | return NULL; | |
1828 | } | |
1829 | ||
86c27869 AH |
1830 | static int intel_pt_sync_switch(struct intel_pt *pt, int cpu, pid_t tid, |
1831 | u64 timestamp) | |
90e457f7 AH |
1832 | { |
1833 | struct intel_pt_queue *ptq; | |
86c27869 | 1834 | int err; |
90e457f7 AH |
1835 | |
1836 | if (!pt->sync_switch) | |
86c27869 | 1837 | return 1; |
90e457f7 AH |
1838 | |
1839 | ptq = intel_pt_cpu_to_ptq(pt, cpu); | |
1840 | if (!ptq) | |
86c27869 | 1841 | return 1; |
90e457f7 AH |
1842 | |
1843 | switch (ptq->switch_state) { | |
1844 | case INTEL_PT_SS_NOT_TRACING: | |
1845 | ptq->next_tid = -1; | |
1846 | break; | |
1847 | case INTEL_PT_SS_UNKNOWN: | |
1848 | case INTEL_PT_SS_TRACING: | |
1849 | ptq->next_tid = tid; | |
1850 | ptq->switch_state = INTEL_PT_SS_EXPECTING_SWITCH_IP; | |
1851 | return 0; | |
1852 | case INTEL_PT_SS_EXPECTING_SWITCH_EVENT: | |
1853 | if (!ptq->on_heap) { | |
86c27869 | 1854 | ptq->timestamp = perf_time_to_tsc(timestamp, |
90e457f7 AH |
1855 | &pt->tc); |
1856 | err = auxtrace_heap__add(&pt->heap, ptq->queue_nr, | |
1857 | ptq->timestamp); | |
1858 | if (err) | |
1859 | return err; | |
1860 | ptq->on_heap = true; | |
1861 | } | |
1862 | ptq->switch_state = INTEL_PT_SS_TRACING; | |
1863 | break; | |
1864 | case INTEL_PT_SS_EXPECTING_SWITCH_IP: | |
1865 | ptq->next_tid = tid; | |
1866 | intel_pt_log("ERROR: cpu %d expecting switch ip\n", cpu); | |
1867 | break; | |
1868 | default: | |
1869 | break; | |
1870 | } | |
86c27869 AH |
1871 | |
1872 | return 1; | |
1873 | } | |
1874 | ||
1875 | static int intel_pt_process_switch(struct intel_pt *pt, | |
1876 | struct perf_sample *sample) | |
1877 | { | |
1878 | struct perf_evsel *evsel; | |
1879 | pid_t tid; | |
1880 | int cpu, ret; | |
1881 | ||
1882 | evsel = perf_evlist__id2evsel(pt->session->evlist, sample->id); | |
1883 | if (evsel != pt->switch_evsel) | |
1884 | return 0; | |
1885 | ||
1886 | tid = perf_evsel__intval(evsel, sample, "next_pid"); | |
1887 | cpu = sample->cpu; | |
1888 | ||
1889 | intel_pt_log("sched_switch: cpu %d tid %d time %"PRIu64" tsc %#"PRIx64"\n", | |
1890 | cpu, tid, sample->time, perf_time_to_tsc(sample->time, | |
1891 | &pt->tc)); | |
1892 | ||
1893 | ret = intel_pt_sync_switch(pt, cpu, tid, sample->time); | |
1894 | if (ret <= 0) | |
1895 | return ret; | |
1896 | ||
90e457f7 AH |
1897 | return machine__set_current_tid(pt->machine, cpu, -1, tid); |
1898 | } | |
1899 | ||
86c27869 AH |
1900 | static int intel_pt_context_switch(struct intel_pt *pt, union perf_event *event, |
1901 | struct perf_sample *sample) | |
1902 | { | |
1903 | bool out = event->header.misc & PERF_RECORD_MISC_SWITCH_OUT; | |
1904 | pid_t pid, tid; | |
1905 | int cpu, ret; | |
1906 | ||
1907 | cpu = sample->cpu; | |
1908 | ||
1909 | if (pt->have_sched_switch == 3) { | |
1910 | if (!out) | |
1911 | return 0; | |
1912 | if (event->header.type != PERF_RECORD_SWITCH_CPU_WIDE) { | |
1913 | pr_err("Expecting CPU-wide context switch event\n"); | |
1914 | return -EINVAL; | |
1915 | } | |
1916 | pid = event->context_switch.next_prev_pid; | |
1917 | tid = event->context_switch.next_prev_tid; | |
1918 | } else { | |
1919 | if (out) | |
1920 | return 0; | |
1921 | pid = sample->pid; | |
1922 | tid = sample->tid; | |
1923 | } | |
1924 | ||
1925 | if (tid == -1) { | |
1926 | pr_err("context_switch event has no tid\n"); | |
1927 | return -EINVAL; | |
1928 | } | |
1929 | ||
1930 | intel_pt_log("context_switch: cpu %d pid %d tid %d time %"PRIu64" tsc %#"PRIx64"\n", | |
1931 | cpu, pid, tid, sample->time, perf_time_to_tsc(sample->time, | |
1932 | &pt->tc)); | |
1933 | ||
1934 | ret = intel_pt_sync_switch(pt, cpu, tid, sample->time); | |
1935 | if (ret <= 0) | |
1936 | return ret; | |
1937 | ||
1938 | return machine__set_current_tid(pt->machine, cpu, pid, tid); | |
1939 | } | |
1940 | ||
90e457f7 AH |
1941 | static int intel_pt_process_itrace_start(struct intel_pt *pt, |
1942 | union perf_event *event, | |
1943 | struct perf_sample *sample) | |
1944 | { | |
1945 | if (!pt->per_cpu_mmaps) | |
1946 | return 0; | |
1947 | ||
1948 | intel_pt_log("itrace_start: cpu %d pid %d tid %d time %"PRIu64" tsc %#"PRIx64"\n", | |
1949 | sample->cpu, event->itrace_start.pid, | |
1950 | event->itrace_start.tid, sample->time, | |
1951 | perf_time_to_tsc(sample->time, &pt->tc)); | |
1952 | ||
1953 | return machine__set_current_tid(pt->machine, sample->cpu, | |
1954 | event->itrace_start.pid, | |
1955 | event->itrace_start.tid); | |
1956 | } | |
1957 | ||
1958 | static int intel_pt_process_event(struct perf_session *session, | |
1959 | union perf_event *event, | |
1960 | struct perf_sample *sample, | |
1961 | struct perf_tool *tool) | |
1962 | { | |
1963 | struct intel_pt *pt = container_of(session->auxtrace, struct intel_pt, | |
1964 | auxtrace); | |
1965 | u64 timestamp; | |
1966 | int err = 0; | |
1967 | ||
1968 | if (dump_trace) | |
1969 | return 0; | |
1970 | ||
1971 | if (!tool->ordered_events) { | |
1972 | pr_err("Intel Processor Trace requires ordered events\n"); | |
1973 | return -EINVAL; | |
1974 | } | |
1975 | ||
81cd60cc | 1976 | if (sample->time && sample->time != (u64)-1) |
90e457f7 AH |
1977 | timestamp = perf_time_to_tsc(sample->time, &pt->tc); |
1978 | else | |
1979 | timestamp = 0; | |
1980 | ||
1981 | if (timestamp || pt->timeless_decoding) { | |
1982 | err = intel_pt_update_queues(pt); | |
1983 | if (err) | |
1984 | return err; | |
1985 | } | |
1986 | ||
1987 | if (pt->timeless_decoding) { | |
1988 | if (event->header.type == PERF_RECORD_EXIT) { | |
1989 | err = intel_pt_process_timeless_queues(pt, | |
53ff6bc3 | 1990 | event->fork.tid, |
90e457f7 AH |
1991 | sample->time); |
1992 | } | |
1993 | } else if (timestamp) { | |
1994 | err = intel_pt_process_queues(pt, timestamp); | |
1995 | } | |
1996 | if (err) | |
1997 | return err; | |
1998 | ||
1999 | if (event->header.type == PERF_RECORD_AUX && | |
2000 | (event->aux.flags & PERF_AUX_FLAG_TRUNCATED) && | |
2001 | pt->synth_opts.errors) { | |
2002 | err = intel_pt_lost(pt, sample); | |
2003 | if (err) | |
2004 | return err; | |
2005 | } | |
2006 | ||
2007 | if (pt->switch_evsel && event->header.type == PERF_RECORD_SAMPLE) | |
2008 | err = intel_pt_process_switch(pt, sample); | |
2009 | else if (event->header.type == PERF_RECORD_ITRACE_START) | |
2010 | err = intel_pt_process_itrace_start(pt, event, sample); | |
86c27869 AH |
2011 | else if (event->header.type == PERF_RECORD_SWITCH || |
2012 | event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) | |
2013 | err = intel_pt_context_switch(pt, event, sample); | |
90e457f7 AH |
2014 | |
2015 | intel_pt_log("event %s (%u): cpu %d time %"PRIu64" tsc %#"PRIx64"\n", | |
2016 | perf_event__name(event->header.type), event->header.type, | |
2017 | sample->cpu, sample->time, timestamp); | |
2018 | ||
2019 | return err; | |
2020 | } | |
2021 | ||
2022 | static int intel_pt_flush(struct perf_session *session, struct perf_tool *tool) | |
2023 | { | |
2024 | struct intel_pt *pt = container_of(session->auxtrace, struct intel_pt, | |
2025 | auxtrace); | |
2026 | int ret; | |
2027 | ||
2028 | if (dump_trace) | |
2029 | return 0; | |
2030 | ||
2031 | if (!tool->ordered_events) | |
2032 | return -EINVAL; | |
2033 | ||
2034 | ret = intel_pt_update_queues(pt); | |
2035 | if (ret < 0) | |
2036 | return ret; | |
2037 | ||
2038 | if (pt->timeless_decoding) | |
2039 | return intel_pt_process_timeless_queues(pt, -1, | |
2040 | MAX_TIMESTAMP - 1); | |
2041 | ||
2042 | return intel_pt_process_queues(pt, MAX_TIMESTAMP); | |
2043 | } | |
2044 | ||
2045 | static void intel_pt_free_events(struct perf_session *session) | |
2046 | { | |
2047 | struct intel_pt *pt = container_of(session->auxtrace, struct intel_pt, | |
2048 | auxtrace); | |
2049 | struct auxtrace_queues *queues = &pt->queues; | |
2050 | unsigned int i; | |
2051 | ||
2052 | for (i = 0; i < queues->nr_queues; i++) { | |
2053 | intel_pt_free_queue(queues->queue_array[i].priv); | |
2054 | queues->queue_array[i].priv = NULL; | |
2055 | } | |
2056 | intel_pt_log_disable(); | |
2057 | auxtrace_queues__free(queues); | |
2058 | } | |
2059 | ||
2060 | static void intel_pt_free(struct perf_session *session) | |
2061 | { | |
2062 | struct intel_pt *pt = container_of(session->auxtrace, struct intel_pt, | |
2063 | auxtrace); | |
2064 | ||
2065 | auxtrace_heap__free(&pt->heap); | |
2066 | intel_pt_free_events(session); | |
2067 | session->auxtrace = NULL; | |
abd82868 | 2068 | thread__put(pt->unknown_thread); |
2acee108 | 2069 | addr_filters__exit(&pt->filts); |
2b9e32c4 | 2070 | zfree(&pt->filter); |
90e457f7 AH |
2071 | free(pt); |
2072 | } | |
2073 | ||
2074 | static int intel_pt_process_auxtrace_event(struct perf_session *session, | |
2075 | union perf_event *event, | |
2076 | struct perf_tool *tool __maybe_unused) | |
2077 | { | |
2078 | struct intel_pt *pt = container_of(session->auxtrace, struct intel_pt, | |
2079 | auxtrace); | |
2080 | ||
2081 | if (pt->sampling_mode) | |
2082 | return 0; | |
2083 | ||
2084 | if (!pt->data_queued) { | |
2085 | struct auxtrace_buffer *buffer; | |
2086 | off_t data_offset; | |
8ceb41d7 | 2087 | int fd = perf_data__fd(session->data); |
90e457f7 AH |
2088 | int err; |
2089 | ||
8ceb41d7 | 2090 | if (perf_data__is_pipe(session->data)) { |
90e457f7 AH |
2091 | data_offset = 0; |
2092 | } else { | |
2093 | data_offset = lseek(fd, 0, SEEK_CUR); | |
2094 | if (data_offset == -1) | |
2095 | return -errno; | |
2096 | } | |
2097 | ||
2098 | err = auxtrace_queues__add_event(&pt->queues, session, event, | |
2099 | data_offset, &buffer); | |
2100 | if (err) | |
2101 | return err; | |
2102 | ||
2103 | /* Dump here now we have copied a piped trace out of the pipe */ | |
2104 | if (dump_trace) { | |
2105 | if (auxtrace_buffer__get_data(buffer, fd)) { | |
2106 | intel_pt_dump_event(pt, buffer->data, | |
2107 | buffer->size); | |
2108 | auxtrace_buffer__put_data(buffer); | |
2109 | } | |
2110 | } | |
2111 | } | |
2112 | ||
2113 | return 0; | |
2114 | } | |
2115 | ||
2116 | struct intel_pt_synth { | |
2117 | struct perf_tool dummy_tool; | |
2118 | struct perf_session *session; | |
2119 | }; | |
2120 | ||
2121 | static int intel_pt_event_synth(struct perf_tool *tool, | |
2122 | union perf_event *event, | |
2123 | struct perf_sample *sample __maybe_unused, | |
2124 | struct machine *machine __maybe_unused) | |
2125 | { | |
2126 | struct intel_pt_synth *intel_pt_synth = | |
2127 | container_of(tool, struct intel_pt_synth, dummy_tool); | |
2128 | ||
2129 | return perf_session__deliver_synth_event(intel_pt_synth->session, event, | |
2130 | NULL); | |
2131 | } | |
2132 | ||
63a22cd9 | 2133 | static int intel_pt_synth_event(struct perf_session *session, const char *name, |
90e457f7 AH |
2134 | struct perf_event_attr *attr, u64 id) |
2135 | { | |
2136 | struct intel_pt_synth intel_pt_synth; | |
63a22cd9 AH |
2137 | int err; |
2138 | ||
2139 | pr_debug("Synthesizing '%s' event with id %" PRIu64 " sample type %#" PRIx64 "\n", | |
2140 | name, id, (u64)attr->sample_type); | |
90e457f7 AH |
2141 | |
2142 | memset(&intel_pt_synth, 0, sizeof(struct intel_pt_synth)); | |
2143 | intel_pt_synth.session = session; | |
2144 | ||
63a22cd9 AH |
2145 | err = perf_event__synthesize_attr(&intel_pt_synth.dummy_tool, attr, 1, |
2146 | &id, intel_pt_event_synth); | |
2147 | if (err) | |
2148 | pr_err("%s: failed to synthesize '%s' event type\n", | |
2149 | __func__, name); | |
2150 | ||
2151 | return err; | |
90e457f7 AH |
2152 | } |
2153 | ||
bbac88ed AH |
2154 | static void intel_pt_set_event_name(struct perf_evlist *evlist, u64 id, |
2155 | const char *name) | |
2156 | { | |
2157 | struct perf_evsel *evsel; | |
2158 | ||
2159 | evlist__for_each_entry(evlist, evsel) { | |
2160 | if (evsel->id && evsel->id[0] == id) { | |
2161 | if (evsel->name) | |
2162 | zfree(&evsel->name); | |
2163 | evsel->name = strdup(name); | |
2164 | break; | |
2165 | } | |
2166 | } | |
2167 | } | |
2168 | ||
85a564d2 AH |
2169 | static struct perf_evsel *intel_pt_evsel(struct intel_pt *pt, |
2170 | struct perf_evlist *evlist) | |
2171 | { | |
2172 | struct perf_evsel *evsel; | |
2173 | ||
2174 | evlist__for_each_entry(evlist, evsel) { | |
2175 | if (evsel->attr.type == pt->pmu_type && evsel->ids) | |
2176 | return evsel; | |
2177 | } | |
2178 | ||
2179 | return NULL; | |
2180 | } | |
2181 | ||
90e457f7 AH |
2182 | static int intel_pt_synth_events(struct intel_pt *pt, |
2183 | struct perf_session *session) | |
2184 | { | |
2185 | struct perf_evlist *evlist = session->evlist; | |
85a564d2 | 2186 | struct perf_evsel *evsel = intel_pt_evsel(pt, evlist); |
90e457f7 | 2187 | struct perf_event_attr attr; |
90e457f7 AH |
2188 | u64 id; |
2189 | int err; | |
2190 | ||
85a564d2 | 2191 | if (!evsel) { |
90e457f7 AH |
2192 | pr_debug("There are no selected events with Intel Processor Trace data\n"); |
2193 | return 0; | |
2194 | } | |
2195 | ||
2196 | memset(&attr, 0, sizeof(struct perf_event_attr)); | |
2197 | attr.size = sizeof(struct perf_event_attr); | |
2198 | attr.type = PERF_TYPE_HARDWARE; | |
2199 | attr.sample_type = evsel->attr.sample_type & PERF_SAMPLE_MASK; | |
2200 | attr.sample_type |= PERF_SAMPLE_IP | PERF_SAMPLE_TID | | |
2201 | PERF_SAMPLE_PERIOD; | |
2202 | if (pt->timeless_decoding) | |
2203 | attr.sample_type &= ~(u64)PERF_SAMPLE_TIME; | |
2204 | else | |
2205 | attr.sample_type |= PERF_SAMPLE_TIME; | |
2206 | if (!pt->per_cpu_mmaps) | |
2207 | attr.sample_type &= ~(u64)PERF_SAMPLE_CPU; | |
2208 | attr.exclude_user = evsel->attr.exclude_user; | |
2209 | attr.exclude_kernel = evsel->attr.exclude_kernel; | |
2210 | attr.exclude_hv = evsel->attr.exclude_hv; | |
2211 | attr.exclude_host = evsel->attr.exclude_host; | |
2212 | attr.exclude_guest = evsel->attr.exclude_guest; | |
2213 | attr.sample_id_all = evsel->attr.sample_id_all; | |
2214 | attr.read_format = evsel->attr.read_format; | |
2215 | ||
2216 | id = evsel->id[0] + 1000000000; | |
2217 | if (!id) | |
2218 | id = 1; | |
2219 | ||
4a9fd4e0 AH |
2220 | if (pt->synth_opts.branches) { |
2221 | attr.config = PERF_COUNT_HW_BRANCH_INSTRUCTIONS; | |
2222 | attr.sample_period = 1; | |
2223 | attr.sample_type |= PERF_SAMPLE_ADDR; | |
2224 | err = intel_pt_synth_event(session, "branches", &attr, id); | |
2225 | if (err) | |
2226 | return err; | |
2227 | pt->sample_branches = true; | |
2228 | pt->branches_sample_type = attr.sample_type; | |
2229 | pt->branches_id = id; | |
2230 | id += 1; | |
2231 | attr.sample_type &= ~(u64)PERF_SAMPLE_ADDR; | |
2232 | } | |
2233 | ||
2234 | if (pt->synth_opts.callchain) | |
2235 | attr.sample_type |= PERF_SAMPLE_CALLCHAIN; | |
2236 | if (pt->synth_opts.last_branch) | |
2237 | attr.sample_type |= PERF_SAMPLE_BRANCH_STACK; | |
2238 | ||
90e457f7 AH |
2239 | if (pt->synth_opts.instructions) { |
2240 | attr.config = PERF_COUNT_HW_INSTRUCTIONS; | |
2241 | if (pt->synth_opts.period_type == PERF_ITRACE_PERIOD_NANOSECS) | |
2242 | attr.sample_period = | |
2243 | intel_pt_ns_to_ticks(pt, pt->synth_opts.period); | |
2244 | else | |
2245 | attr.sample_period = pt->synth_opts.period; | |
63a22cd9 AH |
2246 | err = intel_pt_synth_event(session, "instructions", &attr, id); |
2247 | if (err) | |
90e457f7 | 2248 | return err; |
90e457f7 AH |
2249 | pt->sample_instructions = true; |
2250 | pt->instructions_sample_type = attr.sample_type; | |
2251 | pt->instructions_id = id; | |
2252 | id += 1; | |
2253 | } | |
2254 | ||
4a9fd4e0 AH |
2255 | attr.sample_type &= ~(u64)PERF_SAMPLE_PERIOD; |
2256 | attr.sample_period = 1; | |
2257 | ||
90e457f7 AH |
2258 | if (pt->synth_opts.transactions) { |
2259 | attr.config = PERF_COUNT_HW_INSTRUCTIONS; | |
63a22cd9 AH |
2260 | err = intel_pt_synth_event(session, "transactions", &attr, id); |
2261 | if (err) | |
90e457f7 | 2262 | return err; |
90e457f7 | 2263 | pt->sample_transactions = true; |
21160748 | 2264 | pt->transactions_sample_type = attr.sample_type; |
90e457f7 | 2265 | pt->transactions_id = id; |
bbac88ed | 2266 | intel_pt_set_event_name(evlist, id, "transactions"); |
90e457f7 | 2267 | id += 1; |
90e457f7 AH |
2268 | } |
2269 | ||
37973075 AH |
2270 | attr.type = PERF_TYPE_SYNTH; |
2271 | attr.sample_type |= PERF_SAMPLE_RAW; | |
2272 | ||
2273 | if (pt->synth_opts.ptwrites) { | |
2274 | attr.config = PERF_SYNTH_INTEL_PTWRITE; | |
2275 | err = intel_pt_synth_event(session, "ptwrite", &attr, id); | |
2276 | if (err) | |
2277 | return err; | |
2278 | pt->sample_ptwrites = true; | |
2279 | pt->ptwrites_sample_type = attr.sample_type; | |
2280 | pt->ptwrites_id = id; | |
2281 | intel_pt_set_event_name(evlist, id, "ptwrite"); | |
2282 | id += 1; | |
2283 | } | |
2284 | ||
2285 | if (pt->synth_opts.pwr_events) { | |
2286 | pt->sample_pwr_events = true; | |
2287 | pt->pwr_events_sample_type = attr.sample_type; | |
2288 | ||
2289 | attr.config = PERF_SYNTH_INTEL_CBR; | |
2290 | err = intel_pt_synth_event(session, "cbr", &attr, id); | |
2291 | if (err) | |
2292 | return err; | |
2293 | pt->cbr_id = id; | |
2294 | intel_pt_set_event_name(evlist, id, "cbr"); | |
2295 | id += 1; | |
2296 | } | |
2297 | ||
2298 | if (pt->synth_opts.pwr_events && (evsel->attr.config & 0x10)) { | |
2299 | attr.config = PERF_SYNTH_INTEL_MWAIT; | |
2300 | err = intel_pt_synth_event(session, "mwait", &attr, id); | |
2301 | if (err) | |
2302 | return err; | |
2303 | pt->mwait_id = id; | |
2304 | intel_pt_set_event_name(evlist, id, "mwait"); | |
2305 | id += 1; | |
2306 | ||
2307 | attr.config = PERF_SYNTH_INTEL_PWRE; | |
2308 | err = intel_pt_synth_event(session, "pwre", &attr, id); | |
2309 | if (err) | |
2310 | return err; | |
2311 | pt->pwre_id = id; | |
2312 | intel_pt_set_event_name(evlist, id, "pwre"); | |
2313 | id += 1; | |
2314 | ||
2315 | attr.config = PERF_SYNTH_INTEL_EXSTOP; | |
2316 | err = intel_pt_synth_event(session, "exstop", &attr, id); | |
2317 | if (err) | |
2318 | return err; | |
2319 | pt->exstop_id = id; | |
2320 | intel_pt_set_event_name(evlist, id, "exstop"); | |
2321 | id += 1; | |
2322 | ||
2323 | attr.config = PERF_SYNTH_INTEL_PWRX; | |
2324 | err = intel_pt_synth_event(session, "pwrx", &attr, id); | |
2325 | if (err) | |
2326 | return err; | |
2327 | pt->pwrx_id = id; | |
2328 | intel_pt_set_event_name(evlist, id, "pwrx"); | |
2329 | id += 1; | |
2330 | } | |
2331 | ||
90e457f7 AH |
2332 | pt->synth_needs_swap = evsel->needs_swap; |
2333 | ||
2334 | return 0; | |
2335 | } | |
2336 | ||
2337 | static struct perf_evsel *intel_pt_find_sched_switch(struct perf_evlist *evlist) | |
2338 | { | |
2339 | struct perf_evsel *evsel; | |
2340 | ||
e5cadb93 | 2341 | evlist__for_each_entry_reverse(evlist, evsel) { |
90e457f7 AH |
2342 | const char *name = perf_evsel__name(evsel); |
2343 | ||
2344 | if (!strcmp(name, "sched:sched_switch")) | |
2345 | return evsel; | |
2346 | } | |
2347 | ||
2348 | return NULL; | |
2349 | } | |
2350 | ||
86c27869 AH |
2351 | static bool intel_pt_find_switch(struct perf_evlist *evlist) |
2352 | { | |
2353 | struct perf_evsel *evsel; | |
2354 | ||
e5cadb93 | 2355 | evlist__for_each_entry(evlist, evsel) { |
86c27869 AH |
2356 | if (evsel->attr.context_switch) |
2357 | return true; | |
2358 | } | |
2359 | ||
2360 | return false; | |
2361 | } | |
2362 | ||
ba11ba65 AH |
2363 | static int intel_pt_perf_config(const char *var, const char *value, void *data) |
2364 | { | |
2365 | struct intel_pt *pt = data; | |
2366 | ||
2367 | if (!strcmp(var, "intel-pt.mispred-all")) | |
2368 | pt->mispred_all = perf_config_bool(var, value); | |
2369 | ||
2370 | return 0; | |
2371 | } | |
2372 | ||
90e457f7 | 2373 | static const char * const intel_pt_info_fmts[] = { |
11fa7cb8 AH |
2374 | [INTEL_PT_PMU_TYPE] = " PMU Type %"PRId64"\n", |
2375 | [INTEL_PT_TIME_SHIFT] = " Time Shift %"PRIu64"\n", | |
2376 | [INTEL_PT_TIME_MULT] = " Time Muliplier %"PRIu64"\n", | |
2377 | [INTEL_PT_TIME_ZERO] = " Time Zero %"PRIu64"\n", | |
2378 | [INTEL_PT_CAP_USER_TIME_ZERO] = " Cap Time Zero %"PRId64"\n", | |
2379 | [INTEL_PT_TSC_BIT] = " TSC bit %#"PRIx64"\n", | |
2380 | [INTEL_PT_NORETCOMP_BIT] = " NoRETComp bit %#"PRIx64"\n", | |
2381 | [INTEL_PT_HAVE_SCHED_SWITCH] = " Have sched_switch %"PRId64"\n", | |
2382 | [INTEL_PT_SNAPSHOT_MODE] = " Snapshot mode %"PRId64"\n", | |
2383 | [INTEL_PT_PER_CPU_MMAPS] = " Per-cpu maps %"PRId64"\n", | |
2384 | [INTEL_PT_MTC_BIT] = " MTC bit %#"PRIx64"\n", | |
2385 | [INTEL_PT_TSC_CTC_N] = " TSC:CTC numerator %"PRIu64"\n", | |
2386 | [INTEL_PT_TSC_CTC_D] = " TSC:CTC denominator %"PRIu64"\n", | |
2387 | [INTEL_PT_CYC_BIT] = " CYC bit %#"PRIx64"\n", | |
fa8025c3 | 2388 | [INTEL_PT_MAX_NONTURBO_RATIO] = " Max non-turbo ratio %"PRIu64"\n", |
2b9e32c4 | 2389 | [INTEL_PT_FILTER_STR_LEN] = " Filter string len. %"PRIu64"\n", |
90e457f7 AH |
2390 | }; |
2391 | ||
2392 | static void intel_pt_print_info(u64 *arr, int start, int finish) | |
2393 | { | |
2394 | int i; | |
2395 | ||
2396 | if (!dump_trace) | |
2397 | return; | |
2398 | ||
2399 | for (i = start; i <= finish; i++) | |
2400 | fprintf(stdout, intel_pt_info_fmts[i], arr[i]); | |
2401 | } | |
2402 | ||
2b9e32c4 AH |
2403 | static void intel_pt_print_info_str(const char *name, const char *str) |
2404 | { | |
2405 | if (!dump_trace) | |
2406 | return; | |
2407 | ||
2408 | fprintf(stdout, " %-20s%s\n", name, str ? str : ""); | |
2409 | } | |
2410 | ||
40b746a0 AH |
2411 | static bool intel_pt_has(struct auxtrace_info_event *auxtrace_info, int pos) |
2412 | { | |
2413 | return auxtrace_info->header.size >= | |
2414 | sizeof(struct auxtrace_info_event) + (sizeof(u64) * (pos + 1)); | |
2415 | } | |
2416 | ||
90e457f7 AH |
2417 | int intel_pt_process_auxtrace_info(union perf_event *event, |
2418 | struct perf_session *session) | |
2419 | { | |
2420 | struct auxtrace_info_event *auxtrace_info = &event->auxtrace_info; | |
2421 | size_t min_sz = sizeof(u64) * INTEL_PT_PER_CPU_MMAPS; | |
2422 | struct intel_pt *pt; | |
2b9e32c4 AH |
2423 | void *info_end; |
2424 | u64 *info; | |
90e457f7 AH |
2425 | int err; |
2426 | ||
2427 | if (auxtrace_info->header.size < sizeof(struct auxtrace_info_event) + | |
2428 | min_sz) | |
2429 | return -EINVAL; | |
2430 | ||
2431 | pt = zalloc(sizeof(struct intel_pt)); | |
2432 | if (!pt) | |
2433 | return -ENOMEM; | |
2434 | ||
2acee108 AH |
2435 | addr_filters__init(&pt->filts); |
2436 | ||
ecc4c561 ACM |
2437 | err = perf_config(intel_pt_perf_config, pt); |
2438 | if (err) | |
2439 | goto err_free; | |
ba11ba65 | 2440 | |
90e457f7 AH |
2441 | err = auxtrace_queues__init(&pt->queues); |
2442 | if (err) | |
2443 | goto err_free; | |
2444 | ||
2445 | intel_pt_log_set_name(INTEL_PT_PMU_NAME); | |
2446 | ||
2447 | pt->session = session; | |
2448 | pt->machine = &session->machines.host; /* No kvm support */ | |
2449 | pt->auxtrace_type = auxtrace_info->type; | |
2450 | pt->pmu_type = auxtrace_info->priv[INTEL_PT_PMU_TYPE]; | |
2451 | pt->tc.time_shift = auxtrace_info->priv[INTEL_PT_TIME_SHIFT]; | |
2452 | pt->tc.time_mult = auxtrace_info->priv[INTEL_PT_TIME_MULT]; | |
2453 | pt->tc.time_zero = auxtrace_info->priv[INTEL_PT_TIME_ZERO]; | |
2454 | pt->cap_user_time_zero = auxtrace_info->priv[INTEL_PT_CAP_USER_TIME_ZERO]; | |
2455 | pt->tsc_bit = auxtrace_info->priv[INTEL_PT_TSC_BIT]; | |
2456 | pt->noretcomp_bit = auxtrace_info->priv[INTEL_PT_NORETCOMP_BIT]; | |
2457 | pt->have_sched_switch = auxtrace_info->priv[INTEL_PT_HAVE_SCHED_SWITCH]; | |
2458 | pt->snapshot_mode = auxtrace_info->priv[INTEL_PT_SNAPSHOT_MODE]; | |
2459 | pt->per_cpu_mmaps = auxtrace_info->priv[INTEL_PT_PER_CPU_MMAPS]; | |
2460 | intel_pt_print_info(&auxtrace_info->priv[0], INTEL_PT_PMU_TYPE, | |
2461 | INTEL_PT_PER_CPU_MMAPS); | |
2462 | ||
40b746a0 | 2463 | if (intel_pt_has(auxtrace_info, INTEL_PT_CYC_BIT)) { |
11fa7cb8 AH |
2464 | pt->mtc_bit = auxtrace_info->priv[INTEL_PT_MTC_BIT]; |
2465 | pt->mtc_freq_bits = auxtrace_info->priv[INTEL_PT_MTC_FREQ_BITS]; | |
2466 | pt->tsc_ctc_ratio_n = auxtrace_info->priv[INTEL_PT_TSC_CTC_N]; | |
2467 | pt->tsc_ctc_ratio_d = auxtrace_info->priv[INTEL_PT_TSC_CTC_D]; | |
2468 | pt->cyc_bit = auxtrace_info->priv[INTEL_PT_CYC_BIT]; | |
2469 | intel_pt_print_info(&auxtrace_info->priv[0], INTEL_PT_MTC_BIT, | |
2470 | INTEL_PT_CYC_BIT); | |
2471 | } | |
2472 | ||
40b746a0 | 2473 | if (intel_pt_has(auxtrace_info, INTEL_PT_MAX_NONTURBO_RATIO)) { |
fa8025c3 AH |
2474 | pt->max_non_turbo_ratio = |
2475 | auxtrace_info->priv[INTEL_PT_MAX_NONTURBO_RATIO]; | |
2476 | intel_pt_print_info(&auxtrace_info->priv[0], | |
2477 | INTEL_PT_MAX_NONTURBO_RATIO, | |
2478 | INTEL_PT_MAX_NONTURBO_RATIO); | |
2479 | } | |
2480 | ||
2b9e32c4 AH |
2481 | info = &auxtrace_info->priv[INTEL_PT_FILTER_STR_LEN] + 1; |
2482 | info_end = (void *)info + auxtrace_info->header.size; | |
2483 | ||
2484 | if (intel_pt_has(auxtrace_info, INTEL_PT_FILTER_STR_LEN)) { | |
2485 | size_t len; | |
2486 | ||
2487 | len = auxtrace_info->priv[INTEL_PT_FILTER_STR_LEN]; | |
2488 | intel_pt_print_info(&auxtrace_info->priv[0], | |
2489 | INTEL_PT_FILTER_STR_LEN, | |
2490 | INTEL_PT_FILTER_STR_LEN); | |
2491 | if (len) { | |
2492 | const char *filter = (const char *)info; | |
2493 | ||
2494 | len = roundup(len + 1, 8); | |
2495 | info += len >> 3; | |
2496 | if ((void *)info > info_end) { | |
2497 | pr_err("%s: bad filter string length\n", __func__); | |
2498 | err = -EINVAL; | |
2499 | goto err_free_queues; | |
2500 | } | |
2501 | pt->filter = memdup(filter, len); | |
2502 | if (!pt->filter) { | |
2503 | err = -ENOMEM; | |
2504 | goto err_free_queues; | |
2505 | } | |
2506 | if (session->header.needs_swap) | |
2507 | mem_bswap_64(pt->filter, len); | |
2508 | if (pt->filter[len - 1]) { | |
2509 | pr_err("%s: filter string not null terminated\n", __func__); | |
2510 | err = -EINVAL; | |
2511 | goto err_free_queues; | |
2512 | } | |
2acee108 AH |
2513 | err = addr_filters__parse_bare_filter(&pt->filts, |
2514 | filter); | |
2515 | if (err) | |
2516 | goto err_free_queues; | |
2b9e32c4 AH |
2517 | } |
2518 | intel_pt_print_info_str("Filter string", pt->filter); | |
2519 | } | |
2520 | ||
90e457f7 AH |
2521 | pt->timeless_decoding = intel_pt_timeless_decoding(pt); |
2522 | pt->have_tsc = intel_pt_have_tsc(pt); | |
2523 | pt->sampling_mode = false; | |
2524 | pt->est_tsc = !pt->timeless_decoding; | |
2525 | ||
2526 | pt->unknown_thread = thread__new(999999999, 999999999); | |
2527 | if (!pt->unknown_thread) { | |
2528 | err = -ENOMEM; | |
2529 | goto err_free_queues; | |
2530 | } | |
3a4acda1 AH |
2531 | |
2532 | /* | |
2533 | * Since this thread will not be kept in any rbtree not in a | |
2534 | * list, initialize its list node so that at thread__put() the | |
2535 | * current thread lifetime assuption is kept and we don't segfault | |
2536 | * at list_del_init(). | |
2537 | */ | |
2538 | INIT_LIST_HEAD(&pt->unknown_thread->node); | |
2539 | ||
90e457f7 AH |
2540 | err = thread__set_comm(pt->unknown_thread, "unknown", 0); |
2541 | if (err) | |
2542 | goto err_delete_thread; | |
2543 | if (thread__init_map_groups(pt->unknown_thread, pt->machine)) { | |
2544 | err = -ENOMEM; | |
2545 | goto err_delete_thread; | |
2546 | } | |
2547 | ||
2548 | pt->auxtrace.process_event = intel_pt_process_event; | |
2549 | pt->auxtrace.process_auxtrace_event = intel_pt_process_auxtrace_event; | |
2550 | pt->auxtrace.flush_events = intel_pt_flush; | |
2551 | pt->auxtrace.free_events = intel_pt_free_events; | |
2552 | pt->auxtrace.free = intel_pt_free; | |
2553 | session->auxtrace = &pt->auxtrace; | |
2554 | ||
2555 | if (dump_trace) | |
2556 | return 0; | |
2557 | ||
2558 | if (pt->have_sched_switch == 1) { | |
2559 | pt->switch_evsel = intel_pt_find_sched_switch(session->evlist); | |
2560 | if (!pt->switch_evsel) { | |
2561 | pr_err("%s: missing sched_switch event\n", __func__); | |
4d34e10a | 2562 | err = -EINVAL; |
90e457f7 AH |
2563 | goto err_delete_thread; |
2564 | } | |
86c27869 AH |
2565 | } else if (pt->have_sched_switch == 2 && |
2566 | !intel_pt_find_switch(session->evlist)) { | |
2567 | pr_err("%s: missing context_switch attribute flag\n", __func__); | |
4d34e10a | 2568 | err = -EINVAL; |
86c27869 | 2569 | goto err_delete_thread; |
90e457f7 AH |
2570 | } |
2571 | ||
2572 | if (session->itrace_synth_opts && session->itrace_synth_opts->set) { | |
2573 | pt->synth_opts = *session->itrace_synth_opts; | |
2574 | } else { | |
2575 | itrace_synth_opts__set_default(&pt->synth_opts); | |
2576 | if (use_browser != -1) { | |
2577 | pt->synth_opts.branches = false; | |
2578 | pt->synth_opts.callchain = true; | |
2579 | } | |
50f73637 AH |
2580 | if (session->itrace_synth_opts) |
2581 | pt->synth_opts.thread_stack = | |
2582 | session->itrace_synth_opts->thread_stack; | |
90e457f7 AH |
2583 | } |
2584 | ||
2585 | if (pt->synth_opts.log) | |
2586 | intel_pt_log_enable(); | |
2587 | ||
2588 | /* Maximum non-turbo ratio is TSC freq / 100 MHz */ | |
2589 | if (pt->tc.time_mult) { | |
2590 | u64 tsc_freq = intel_pt_ns_to_ticks(pt, 1000000000); | |
2591 | ||
fa8025c3 AH |
2592 | if (!pt->max_non_turbo_ratio) |
2593 | pt->max_non_turbo_ratio = | |
2594 | (tsc_freq + 50000000) / 100000000; | |
90e457f7 AH |
2595 | intel_pt_log("TSC frequency %"PRIu64"\n", tsc_freq); |
2596 | intel_pt_log("Maximum non-turbo ratio %u\n", | |
2597 | pt->max_non_turbo_ratio); | |
37973075 | 2598 | pt->cbr2khz = tsc_freq / pt->max_non_turbo_ratio / 1000; |
90e457f7 AH |
2599 | } |
2600 | ||
2601 | if (pt->synth_opts.calls) | |
2602 | pt->branches_filter |= PERF_IP_FLAG_CALL | PERF_IP_FLAG_ASYNC | | |
2603 | PERF_IP_FLAG_TRACE_END; | |
2604 | if (pt->synth_opts.returns) | |
2605 | pt->branches_filter |= PERF_IP_FLAG_RETURN | | |
2606 | PERF_IP_FLAG_TRACE_BEGIN; | |
2607 | ||
2608 | if (pt->synth_opts.callchain && !symbol_conf.use_callchain) { | |
2609 | symbol_conf.use_callchain = true; | |
2610 | if (callchain_register_param(&callchain_param) < 0) { | |
2611 | symbol_conf.use_callchain = false; | |
2612 | pt->synth_opts.callchain = false; | |
2613 | } | |
2614 | } | |
2615 | ||
2616 | err = intel_pt_synth_events(pt, session); | |
2617 | if (err) | |
2618 | goto err_delete_thread; | |
2619 | ||
2620 | err = auxtrace_queues__process_index(&pt->queues, session); | |
2621 | if (err) | |
2622 | goto err_delete_thread; | |
2623 | ||
2624 | if (pt->queues.populated) | |
2625 | pt->data_queued = true; | |
2626 | ||
2627 | if (pt->timeless_decoding) | |
2628 | pr_debug2("Intel PT decoding without timestamps\n"); | |
2629 | ||
2630 | return 0; | |
2631 | ||
2632 | err_delete_thread: | |
abd82868 | 2633 | thread__zput(pt->unknown_thread); |
90e457f7 AH |
2634 | err_free_queues: |
2635 | intel_pt_log_disable(); | |
2636 | auxtrace_queues__free(&pt->queues); | |
2637 | session->auxtrace = NULL; | |
2638 | err_free: | |
2acee108 | 2639 | addr_filters__exit(&pt->filts); |
2b9e32c4 | 2640 | zfree(&pt->filter); |
90e457f7 AH |
2641 | free(pt); |
2642 | return err; | |
2643 | } |