1 /* 2 * auxtrace.h: AUX area 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 #ifndef __PERF_AUXTRACE_H 17 #define __PERF_AUXTRACE_H 18 19 #include <sys/types.h> 20 #include <stdbool.h> 21 #include <stddef.h> 22 #include <linux/list.h> 23 #include <linux/perf_event.h> 24 #include <linux/types.h> 25 26 #include "../perf.h" 27 #include "event.h" 28 #include "session.h" 29 #include "debug.h" 30 31 union perf_event; 32 struct perf_session; 33 struct perf_evlist; 34 struct perf_tool; 35 struct option; 36 struct record_opts; 37 struct auxtrace_info_event; 38 struct events_stats; 39 40 enum auxtrace_type { 41 PERF_AUXTRACE_UNKNOWN, 42 PERF_AUXTRACE_INTEL_PT, 43 PERF_AUXTRACE_INTEL_BTS, 44 PERF_AUXTRACE_CS_ETM, 45 }; 46 47 enum itrace_period_type { 48 PERF_ITRACE_PERIOD_INSTRUCTIONS, 49 PERF_ITRACE_PERIOD_TICKS, 50 PERF_ITRACE_PERIOD_NANOSECS, 51 }; 52 53 /** 54 * struct itrace_synth_opts - AUX area tracing synthesis options. 55 * @set: indicates whether or not options have been set 56 * @inject: indicates the event (not just the sample) must be fully synthesized 57 * because 'perf inject' will write it out 58 * @instructions: whether to synthesize 'instructions' events 59 * @branches: whether to synthesize 'branches' events 60 * @transactions: whether to synthesize events for transactions 61 * @errors: whether to synthesize decoder error events 62 * @dont_decode: whether to skip decoding entirely 63 * @log: write a decoding log 64 * @calls: limit branch samples to calls (can be combined with @returns) 65 * @returns: limit branch samples to returns (can be combined with @calls) 66 * @callchain: add callchain to 'instructions' events 67 * @thread_stack: feed branches to the thread_stack 68 * @last_branch: add branch context to 'instruction' events 69 * @callchain_sz: maximum callchain size 70 * @last_branch_sz: branch context size 71 * @period: 'instructions' events period 72 * @period_type: 'instructions' events period type 73 * @initial_skip: skip N events at the beginning. 74 */ 75 struct itrace_synth_opts { 76 bool set; 77 bool inject; 78 bool instructions; 79 bool branches; 80 bool transactions; 81 bool errors; 82 bool dont_decode; 83 bool log; 84 bool calls; 85 bool returns; 86 bool callchain; 87 bool thread_stack; 88 bool last_branch; 89 unsigned int callchain_sz; 90 unsigned int last_branch_sz; 91 unsigned long long period; 92 enum itrace_period_type period_type; 93 unsigned long initial_skip; 94 }; 95 96 /** 97 * struct auxtrace_index_entry - indexes a AUX area tracing event within a 98 * perf.data file. 99 * @file_offset: offset within the perf.data file 100 * @sz: size of the event 101 */ 102 struct auxtrace_index_entry { 103 u64 file_offset; 104 u64 sz; 105 }; 106 107 #define PERF_AUXTRACE_INDEX_ENTRY_COUNT 256 108 109 /** 110 * struct auxtrace_index - index of AUX area tracing events within a perf.data 111 * file. 112 * @list: linking a number of arrays of entries 113 * @nr: number of entries 114 * @entries: array of entries 115 */ 116 struct auxtrace_index { 117 struct list_head list; 118 size_t nr; 119 struct auxtrace_index_entry entries[PERF_AUXTRACE_INDEX_ENTRY_COUNT]; 120 }; 121 122 /** 123 * struct auxtrace - session callbacks to allow AUX area data decoding. 124 * @process_event: lets the decoder see all session events 125 * @flush_events: process any remaining data 126 * @free_events: free resources associated with event processing 127 * @free: free resources associated with the session 128 */ 129 struct auxtrace { 130 int (*process_event)(struct perf_session *session, 131 union perf_event *event, 132 struct perf_sample *sample, 133 struct perf_tool *tool); 134 int (*process_auxtrace_event)(struct perf_session *session, 135 union perf_event *event, 136 struct perf_tool *tool); 137 int (*flush_events)(struct perf_session *session, 138 struct perf_tool *tool); 139 void (*free_events)(struct perf_session *session); 140 void (*free)(struct perf_session *session); 141 }; 142 143 /** 144 * struct auxtrace_buffer - a buffer containing AUX area tracing data. 145 * @list: buffers are queued in a list held by struct auxtrace_queue 146 * @size: size of the buffer in bytes 147 * @pid: in per-thread mode, the pid this buffer is associated with 148 * @tid: in per-thread mode, the tid this buffer is associated with 149 * @cpu: in per-cpu mode, the cpu this buffer is associated with 150 * @data: actual buffer data (can be null if the data has not been loaded) 151 * @data_offset: file offset at which the buffer can be read 152 * @mmap_addr: mmap address at which the buffer can be read 153 * @mmap_size: size of the mmap at @mmap_addr 154 * @data_needs_freeing: @data was malloc'd so free it when it is no longer 155 * needed 156 * @consecutive: the original data was split up and this buffer is consecutive 157 * to the previous buffer 158 * @offset: offset as determined by aux_head / aux_tail members of struct 159 * perf_event_mmap_page 160 * @reference: an implementation-specific reference determined when the data is 161 * recorded 162 * @buffer_nr: used to number each buffer 163 * @use_size: implementation actually only uses this number of bytes 164 * @use_data: implementation actually only uses data starting at this address 165 */ 166 struct auxtrace_buffer { 167 struct list_head list; 168 size_t size; 169 pid_t pid; 170 pid_t tid; 171 int cpu; 172 void *data; 173 off_t data_offset; 174 void *mmap_addr; 175 size_t mmap_size; 176 bool data_needs_freeing; 177 bool consecutive; 178 u64 offset; 179 u64 reference; 180 u64 buffer_nr; 181 size_t use_size; 182 void *use_data; 183 }; 184 185 /** 186 * struct auxtrace_queue - a queue of AUX area tracing data buffers. 187 * @head: head of buffer list 188 * @tid: in per-thread mode, the tid this queue is associated with 189 * @cpu: in per-cpu mode, the cpu this queue is associated with 190 * @set: %true once this queue has been dedicated to a specific thread or cpu 191 * @priv: implementation-specific data 192 */ 193 struct auxtrace_queue { 194 struct list_head head; 195 pid_t tid; 196 int cpu; 197 bool set; 198 void *priv; 199 }; 200 201 /** 202 * struct auxtrace_queues - an array of AUX area tracing queues. 203 * @queue_array: array of queues 204 * @nr_queues: number of queues 205 * @new_data: set whenever new data is queued 206 * @populated: queues have been fully populated using the auxtrace_index 207 * @next_buffer_nr: used to number each buffer 208 */ 209 struct auxtrace_queues { 210 struct auxtrace_queue *queue_array; 211 unsigned int nr_queues; 212 bool new_data; 213 bool populated; 214 u64 next_buffer_nr; 215 }; 216 217 /** 218 * struct auxtrace_heap_item - element of struct auxtrace_heap. 219 * @queue_nr: queue number 220 * @ordinal: value used for sorting (lowest ordinal is top of the heap) expected 221 * to be a timestamp 222 */ 223 struct auxtrace_heap_item { 224 unsigned int queue_nr; 225 u64 ordinal; 226 }; 227 228 /** 229 * struct auxtrace_heap - a heap suitable for sorting AUX area tracing queues. 230 * @heap_array: the heap 231 * @heap_cnt: the number of elements in the heap 232 * @heap_sz: maximum number of elements (grows as needed) 233 */ 234 struct auxtrace_heap { 235 struct auxtrace_heap_item *heap_array; 236 unsigned int heap_cnt; 237 unsigned int heap_sz; 238 }; 239 240 /** 241 * struct auxtrace_mmap - records an mmap of the auxtrace buffer. 242 * @base: address of mapped area 243 * @userpg: pointer to buffer's perf_event_mmap_page 244 * @mask: %0 if @len is not a power of two, otherwise (@len - %1) 245 * @len: size of mapped area 246 * @prev: previous aux_head 247 * @idx: index of this mmap 248 * @tid: tid for a per-thread mmap (also set if there is only 1 tid on a per-cpu 249 * mmap) otherwise %0 250 * @cpu: cpu number for a per-cpu mmap otherwise %-1 251 */ 252 struct auxtrace_mmap { 253 void *base; 254 void *userpg; 255 size_t mask; 256 size_t len; 257 u64 prev; 258 int idx; 259 pid_t tid; 260 int cpu; 261 }; 262 263 /** 264 * struct auxtrace_mmap_params - parameters to set up struct auxtrace_mmap. 265 * @mask: %0 if @len is not a power of two, otherwise (@len - %1) 266 * @offset: file offset of mapped area 267 * @len: size of mapped area 268 * @prot: mmap memory protection 269 * @idx: index of this mmap 270 * @tid: tid for a per-thread mmap (also set if there is only 1 tid on a per-cpu 271 * mmap) otherwise %0 272 * @cpu: cpu number for a per-cpu mmap otherwise %-1 273 */ 274 struct auxtrace_mmap_params { 275 size_t mask; 276 off_t offset; 277 size_t len; 278 int prot; 279 int idx; 280 pid_t tid; 281 int cpu; 282 }; 283 284 /** 285 * struct auxtrace_record - callbacks for recording AUX area data. 286 * @recording_options: validate and process recording options 287 * @info_priv_size: return the size of the private data in auxtrace_info_event 288 * @info_fill: fill-in the private data in auxtrace_info_event 289 * @free: free this auxtrace record structure 290 * @snapshot_start: starting a snapshot 291 * @snapshot_finish: finishing a snapshot 292 * @find_snapshot: find data to snapshot within auxtrace mmap 293 * @parse_snapshot_options: parse snapshot options 294 * @reference: provide a 64-bit reference number for auxtrace_event 295 * @read_finish: called after reading from an auxtrace mmap 296 */ 297 struct auxtrace_record { 298 int (*recording_options)(struct auxtrace_record *itr, 299 struct perf_evlist *evlist, 300 struct record_opts *opts); 301 size_t (*info_priv_size)(struct auxtrace_record *itr, 302 struct perf_evlist *evlist); 303 int (*info_fill)(struct auxtrace_record *itr, 304 struct perf_session *session, 305 struct auxtrace_info_event *auxtrace_info, 306 size_t priv_size); 307 void (*free)(struct auxtrace_record *itr); 308 int (*snapshot_start)(struct auxtrace_record *itr); 309 int (*snapshot_finish)(struct auxtrace_record *itr); 310 int (*find_snapshot)(struct auxtrace_record *itr, int idx, 311 struct auxtrace_mmap *mm, unsigned char *data, 312 u64 *head, u64 *old); 313 int (*parse_snapshot_options)(struct auxtrace_record *itr, 314 struct record_opts *opts, 315 const char *str); 316 u64 (*reference)(struct auxtrace_record *itr); 317 int (*read_finish)(struct auxtrace_record *itr, int idx); 318 unsigned int alignment; 319 }; 320 321 #ifdef HAVE_AUXTRACE_SUPPORT 322 323 /* 324 * In snapshot mode the mmapped page is read-only which makes using 325 * __sync_val_compare_and_swap() problematic. However, snapshot mode expects 326 * the buffer is not updated while the snapshot is made (e.g. Intel PT disables 327 * the event) so there is not a race anyway. 328 */ 329 static inline u64 auxtrace_mmap__read_snapshot_head(struct auxtrace_mmap *mm) 330 { 331 struct perf_event_mmap_page *pc = mm->userpg; 332 u64 head = ACCESS_ONCE(pc->aux_head); 333 334 /* Ensure all reads are done after we read the head */ 335 rmb(); 336 return head; 337 } 338 339 static inline u64 auxtrace_mmap__read_head(struct auxtrace_mmap *mm) 340 { 341 struct perf_event_mmap_page *pc = mm->userpg; 342 #if BITS_PER_LONG == 64 || !defined(HAVE_SYNC_COMPARE_AND_SWAP_SUPPORT) 343 u64 head = ACCESS_ONCE(pc->aux_head); 344 #else 345 u64 head = __sync_val_compare_and_swap(&pc->aux_head, 0, 0); 346 #endif 347 348 /* Ensure all reads are done after we read the head */ 349 rmb(); 350 return head; 351 } 352 353 static inline void auxtrace_mmap__write_tail(struct auxtrace_mmap *mm, u64 tail) 354 { 355 struct perf_event_mmap_page *pc = mm->userpg; 356 #if BITS_PER_LONG != 64 && defined(HAVE_SYNC_COMPARE_AND_SWAP_SUPPORT) 357 u64 old_tail; 358 #endif 359 360 /* Ensure all reads are done before we write the tail out */ 361 mb(); 362 #if BITS_PER_LONG == 64 || !defined(HAVE_SYNC_COMPARE_AND_SWAP_SUPPORT) 363 pc->aux_tail = tail; 364 #else 365 do { 366 old_tail = __sync_val_compare_and_swap(&pc->aux_tail, 0, 0); 367 } while (!__sync_bool_compare_and_swap(&pc->aux_tail, old_tail, tail)); 368 #endif 369 } 370 371 int auxtrace_mmap__mmap(struct auxtrace_mmap *mm, 372 struct auxtrace_mmap_params *mp, 373 void *userpg, int fd); 374 void auxtrace_mmap__munmap(struct auxtrace_mmap *mm); 375 void auxtrace_mmap_params__init(struct auxtrace_mmap_params *mp, 376 off_t auxtrace_offset, 377 unsigned int auxtrace_pages, 378 bool auxtrace_overwrite); 379 void auxtrace_mmap_params__set_idx(struct auxtrace_mmap_params *mp, 380 struct perf_evlist *evlist, int idx, 381 bool per_cpu); 382 383 typedef int (*process_auxtrace_t)(struct perf_tool *tool, 384 union perf_event *event, void *data1, 385 size_t len1, void *data2, size_t len2); 386 387 int auxtrace_mmap__read(struct auxtrace_mmap *mm, struct auxtrace_record *itr, 388 struct perf_tool *tool, process_auxtrace_t fn); 389 390 int auxtrace_mmap__read_snapshot(struct auxtrace_mmap *mm, 391 struct auxtrace_record *itr, 392 struct perf_tool *tool, process_auxtrace_t fn, 393 size_t snapshot_size); 394 395 int auxtrace_queues__init(struct auxtrace_queues *queues); 396 int auxtrace_queues__add_event(struct auxtrace_queues *queues, 397 struct perf_session *session, 398 union perf_event *event, off_t data_offset, 399 struct auxtrace_buffer **buffer_ptr); 400 void auxtrace_queues__free(struct auxtrace_queues *queues); 401 int auxtrace_queues__process_index(struct auxtrace_queues *queues, 402 struct perf_session *session); 403 struct auxtrace_buffer *auxtrace_buffer__next(struct auxtrace_queue *queue, 404 struct auxtrace_buffer *buffer); 405 void *auxtrace_buffer__get_data(struct auxtrace_buffer *buffer, int fd); 406 void auxtrace_buffer__put_data(struct auxtrace_buffer *buffer); 407 void auxtrace_buffer__drop_data(struct auxtrace_buffer *buffer); 408 void auxtrace_buffer__free(struct auxtrace_buffer *buffer); 409 410 int auxtrace_heap__add(struct auxtrace_heap *heap, unsigned int queue_nr, 411 u64 ordinal); 412 void auxtrace_heap__pop(struct auxtrace_heap *heap); 413 void auxtrace_heap__free(struct auxtrace_heap *heap); 414 415 struct auxtrace_cache_entry { 416 struct hlist_node hash; 417 u32 key; 418 }; 419 420 struct auxtrace_cache *auxtrace_cache__new(unsigned int bits, size_t entry_size, 421 unsigned int limit_percent); 422 void auxtrace_cache__free(struct auxtrace_cache *auxtrace_cache); 423 void *auxtrace_cache__alloc_entry(struct auxtrace_cache *c); 424 void auxtrace_cache__free_entry(struct auxtrace_cache *c, void *entry); 425 int auxtrace_cache__add(struct auxtrace_cache *c, u32 key, 426 struct auxtrace_cache_entry *entry); 427 void *auxtrace_cache__lookup(struct auxtrace_cache *c, u32 key); 428 429 struct auxtrace_record *auxtrace_record__init(struct perf_evlist *evlist, 430 int *err); 431 432 int auxtrace_parse_snapshot_options(struct auxtrace_record *itr, 433 struct record_opts *opts, 434 const char *str); 435 int auxtrace_record__options(struct auxtrace_record *itr, 436 struct perf_evlist *evlist, 437 struct record_opts *opts); 438 size_t auxtrace_record__info_priv_size(struct auxtrace_record *itr, 439 struct perf_evlist *evlist); 440 int auxtrace_record__info_fill(struct auxtrace_record *itr, 441 struct perf_session *session, 442 struct auxtrace_info_event *auxtrace_info, 443 size_t priv_size); 444 void auxtrace_record__free(struct auxtrace_record *itr); 445 int auxtrace_record__snapshot_start(struct auxtrace_record *itr); 446 int auxtrace_record__snapshot_finish(struct auxtrace_record *itr); 447 int auxtrace_record__find_snapshot(struct auxtrace_record *itr, int idx, 448 struct auxtrace_mmap *mm, 449 unsigned char *data, u64 *head, u64 *old); 450 u64 auxtrace_record__reference(struct auxtrace_record *itr); 451 452 int auxtrace_index__auxtrace_event(struct list_head *head, union perf_event *event, 453 off_t file_offset); 454 int auxtrace_index__write(int fd, struct list_head *head); 455 int auxtrace_index__process(int fd, u64 size, struct perf_session *session, 456 bool needs_swap); 457 void auxtrace_index__free(struct list_head *head); 458 459 void auxtrace_synth_error(struct auxtrace_error_event *auxtrace_error, int type, 460 int code, int cpu, pid_t pid, pid_t tid, u64 ip, 461 const char *msg); 462 463 int perf_event__synthesize_auxtrace_info(struct auxtrace_record *itr, 464 struct perf_tool *tool, 465 struct perf_session *session, 466 perf_event__handler_t process); 467 int perf_event__process_auxtrace_info(struct perf_tool *tool, 468 union perf_event *event, 469 struct perf_session *session); 470 s64 perf_event__process_auxtrace(struct perf_tool *tool, 471 union perf_event *event, 472 struct perf_session *session); 473 int perf_event__process_auxtrace_error(struct perf_tool *tool, 474 union perf_event *event, 475 struct perf_session *session); 476 int itrace_parse_synth_opts(const struct option *opt, const char *str, 477 int unset); 478 void itrace_synth_opts__set_default(struct itrace_synth_opts *synth_opts); 479 480 size_t perf_event__fprintf_auxtrace_error(union perf_event *event, FILE *fp); 481 void perf_session__auxtrace_error_inc(struct perf_session *session, 482 union perf_event *event); 483 void events_stats__auxtrace_error_warn(const struct events_stats *stats); 484 485 static inline int auxtrace__process_event(struct perf_session *session, 486 union perf_event *event, 487 struct perf_sample *sample, 488 struct perf_tool *tool) 489 { 490 if (!session->auxtrace) 491 return 0; 492 493 return session->auxtrace->process_event(session, event, sample, tool); 494 } 495 496 static inline int auxtrace__flush_events(struct perf_session *session, 497 struct perf_tool *tool) 498 { 499 if (!session->auxtrace) 500 return 0; 501 502 return session->auxtrace->flush_events(session, tool); 503 } 504 505 static inline void auxtrace__free_events(struct perf_session *session) 506 { 507 if (!session->auxtrace) 508 return; 509 510 return session->auxtrace->free_events(session); 511 } 512 513 static inline void auxtrace__free(struct perf_session *session) 514 { 515 if (!session->auxtrace) 516 return; 517 518 return session->auxtrace->free(session); 519 } 520 521 #else 522 523 static inline struct auxtrace_record * 524 auxtrace_record__init(struct perf_evlist *evlist __maybe_unused, 525 int *err) 526 { 527 *err = 0; 528 return NULL; 529 } 530 531 static inline 532 void auxtrace_record__free(struct auxtrace_record *itr __maybe_unused) 533 { 534 } 535 536 static inline int 537 perf_event__synthesize_auxtrace_info(struct auxtrace_record *itr __maybe_unused, 538 struct perf_tool *tool __maybe_unused, 539 struct perf_session *session __maybe_unused, 540 perf_event__handler_t process __maybe_unused) 541 { 542 return -EINVAL; 543 } 544 545 static inline 546 int auxtrace_record__options(struct auxtrace_record *itr __maybe_unused, 547 struct perf_evlist *evlist __maybe_unused, 548 struct record_opts *opts __maybe_unused) 549 { 550 return 0; 551 } 552 553 #define perf_event__process_auxtrace_info 0 554 #define perf_event__process_auxtrace 0 555 #define perf_event__process_auxtrace_error 0 556 557 static inline 558 void perf_session__auxtrace_error_inc(struct perf_session *session 559 __maybe_unused, 560 union perf_event *event 561 __maybe_unused) 562 { 563 } 564 565 static inline 566 void events_stats__auxtrace_error_warn(const struct events_stats *stats 567 __maybe_unused) 568 { 569 } 570 571 static inline 572 int itrace_parse_synth_opts(const struct option *opt __maybe_unused, 573 const char *str __maybe_unused, 574 int unset __maybe_unused) 575 { 576 pr_err("AUX area tracing not supported\n"); 577 return -EINVAL; 578 } 579 580 static inline 581 int auxtrace_parse_snapshot_options(struct auxtrace_record *itr __maybe_unused, 582 struct record_opts *opts __maybe_unused, 583 const char *str) 584 { 585 if (!str) 586 return 0; 587 pr_err("AUX area tracing not supported\n"); 588 return -EINVAL; 589 } 590 591 static inline 592 int auxtrace__process_event(struct perf_session *session __maybe_unused, 593 union perf_event *event __maybe_unused, 594 struct perf_sample *sample __maybe_unused, 595 struct perf_tool *tool __maybe_unused) 596 { 597 return 0; 598 } 599 600 static inline 601 int auxtrace__flush_events(struct perf_session *session __maybe_unused, 602 struct perf_tool *tool __maybe_unused) 603 { 604 return 0; 605 } 606 607 static inline 608 void auxtrace__free_events(struct perf_session *session __maybe_unused) 609 { 610 } 611 612 static inline 613 void auxtrace_cache__free(struct auxtrace_cache *auxtrace_cache __maybe_unused) 614 { 615 } 616 617 static inline 618 void auxtrace__free(struct perf_session *session __maybe_unused) 619 { 620 } 621 622 static inline 623 int auxtrace_index__write(int fd __maybe_unused, 624 struct list_head *head __maybe_unused) 625 { 626 return -EINVAL; 627 } 628 629 static inline 630 int auxtrace_index__process(int fd __maybe_unused, 631 u64 size __maybe_unused, 632 struct perf_session *session __maybe_unused, 633 bool needs_swap __maybe_unused) 634 { 635 return -EINVAL; 636 } 637 638 static inline 639 void auxtrace_index__free(struct list_head *head __maybe_unused) 640 { 641 } 642 643 int auxtrace_mmap__mmap(struct auxtrace_mmap *mm, 644 struct auxtrace_mmap_params *mp, 645 void *userpg, int fd); 646 void auxtrace_mmap__munmap(struct auxtrace_mmap *mm); 647 void auxtrace_mmap_params__init(struct auxtrace_mmap_params *mp, 648 off_t auxtrace_offset, 649 unsigned int auxtrace_pages, 650 bool auxtrace_overwrite); 651 void auxtrace_mmap_params__set_idx(struct auxtrace_mmap_params *mp, 652 struct perf_evlist *evlist, int idx, 653 bool per_cpu); 654 655 #endif 656 657 #endif 658