2 * auxtrace.h: AUX area trace support
3 * Copyright (c) 2013-2015, Intel Corporation.
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.
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
16 #ifndef __PERF_AUXTRACE_H
17 #define __PERF_AUXTRACE_H
19 #include <sys/types.h>
22 #include <linux/list.h>
23 #include <linux/perf_event.h>
24 #include <linux/types.h>
37 struct auxtrace_info_event;
41 PERF_AUXTRACE_UNKNOWN,
42 PERF_AUXTRACE_INTEL_PT,
43 PERF_AUXTRACE_INTEL_BTS,
46 enum itrace_period_type {
47 PERF_ITRACE_PERIOD_INSTRUCTIONS,
48 PERF_ITRACE_PERIOD_TICKS,
49 PERF_ITRACE_PERIOD_NANOSECS,
53 * struct itrace_synth_opts - AUX area tracing synthesis options.
54 * @set: indicates whether or not options have been set
55 * @inject: indicates the event (not just the sample) must be fully synthesized
56 * because 'perf inject' will write it out
57 * @instructions: whether to synthesize 'instructions' events
58 * @branches: whether to synthesize 'branches' events
59 * @transactions: whether to synthesize events for transactions
60 * @errors: whether to synthesize decoder error events
61 * @dont_decode: whether to skip decoding entirely
62 * @log: write a decoding log
63 * @calls: limit branch samples to calls (can be combined with @returns)
64 * @returns: limit branch samples to returns (can be combined with @calls)
65 * @callchain: add callchain to 'instructions' events
66 * @thread_stack: feed branches to the thread_stack
67 * @last_branch: add branch context to 'instruction' events
68 * @callchain_sz: maximum callchain size
69 * @last_branch_sz: branch context size
70 * @period: 'instructions' events period
71 * @period_type: 'instructions' events period type
72 * @initial_skip: skip N events at the beginning.
74 struct itrace_synth_opts {
88 unsigned int callchain_sz;
89 unsigned int last_branch_sz;
90 unsigned long long period;
91 enum itrace_period_type period_type;
92 unsigned long initial_skip;
96 * struct auxtrace_index_entry - indexes a AUX area tracing event within a
98 * @file_offset: offset within the perf.data file
99 * @sz: size of the event
101 struct auxtrace_index_entry {
106 #define PERF_AUXTRACE_INDEX_ENTRY_COUNT 256
109 * struct auxtrace_index - index of AUX area tracing events within a perf.data
111 * @list: linking a number of arrays of entries
112 * @nr: number of entries
113 * @entries: array of entries
115 struct auxtrace_index {
116 struct list_head list;
118 struct auxtrace_index_entry entries[PERF_AUXTRACE_INDEX_ENTRY_COUNT];
122 * struct auxtrace - session callbacks to allow AUX area data decoding.
123 * @process_event: lets the decoder see all session events
124 * @flush_events: process any remaining data
125 * @free_events: free resources associated with event processing
126 * @free: free resources associated with the session
129 int (*process_event)(struct perf_session *session,
130 union perf_event *event,
131 struct perf_sample *sample,
132 struct perf_tool *tool);
133 int (*process_auxtrace_event)(struct perf_session *session,
134 union perf_event *event,
135 struct perf_tool *tool);
136 int (*flush_events)(struct perf_session *session,
137 struct perf_tool *tool);
138 void (*free_events)(struct perf_session *session);
139 void (*free)(struct perf_session *session);
143 * struct auxtrace_buffer - a buffer containing AUX area tracing data.
144 * @list: buffers are queued in a list held by struct auxtrace_queue
145 * @size: size of the buffer in bytes
146 * @pid: in per-thread mode, the pid this buffer is associated with
147 * @tid: in per-thread mode, the tid this buffer is associated with
148 * @cpu: in per-cpu mode, the cpu this buffer is associated with
149 * @data: actual buffer data (can be null if the data has not been loaded)
150 * @data_offset: file offset at which the buffer can be read
151 * @mmap_addr: mmap address at which the buffer can be read
152 * @mmap_size: size of the mmap at @mmap_addr
153 * @data_needs_freeing: @data was malloc'd so free it when it is no longer
155 * @consecutive: the original data was split up and this buffer is consecutive
156 * to the previous buffer
157 * @offset: offset as determined by aux_head / aux_tail members of struct
158 * perf_event_mmap_page
159 * @reference: an implementation-specific reference determined when the data is
161 * @buffer_nr: used to number each buffer
162 * @use_size: implementation actually only uses this number of bytes
163 * @use_data: implementation actually only uses data starting at this address
165 struct auxtrace_buffer {
166 struct list_head list;
175 bool data_needs_freeing;
185 * struct auxtrace_queue - a queue of AUX area tracing data buffers.
186 * @head: head of buffer list
187 * @tid: in per-thread mode, the tid this queue is associated with
188 * @cpu: in per-cpu mode, the cpu this queue is associated with
189 * @set: %true once this queue has been dedicated to a specific thread or cpu
190 * @priv: implementation-specific data
192 struct auxtrace_queue {
193 struct list_head head;
201 * struct auxtrace_queues - an array of AUX area tracing queues.
202 * @queue_array: array of queues
203 * @nr_queues: number of queues
204 * @new_data: set whenever new data is queued
205 * @populated: queues have been fully populated using the auxtrace_index
206 * @next_buffer_nr: used to number each buffer
208 struct auxtrace_queues {
209 struct auxtrace_queue *queue_array;
210 unsigned int nr_queues;
217 * struct auxtrace_heap_item - element of struct auxtrace_heap.
218 * @queue_nr: queue number
219 * @ordinal: value used for sorting (lowest ordinal is top of the heap) expected
222 struct auxtrace_heap_item {
223 unsigned int queue_nr;
228 * struct auxtrace_heap - a heap suitable for sorting AUX area tracing queues.
229 * @heap_array: the heap
230 * @heap_cnt: the number of elements in the heap
231 * @heap_sz: maximum number of elements (grows as needed)
233 struct auxtrace_heap {
234 struct auxtrace_heap_item *heap_array;
235 unsigned int heap_cnt;
236 unsigned int heap_sz;
240 * struct auxtrace_mmap - records an mmap of the auxtrace buffer.
241 * @base: address of mapped area
242 * @userpg: pointer to buffer's perf_event_mmap_page
243 * @mask: %0 if @len is not a power of two, otherwise (@len - %1)
244 * @len: size of mapped area
245 * @prev: previous aux_head
246 * @idx: index of this mmap
247 * @tid: tid for a per-thread mmap (also set if there is only 1 tid on a per-cpu
249 * @cpu: cpu number for a per-cpu mmap otherwise %-1
251 struct auxtrace_mmap {
263 * struct auxtrace_mmap_params - parameters to set up struct auxtrace_mmap.
264 * @mask: %0 if @len is not a power of two, otherwise (@len - %1)
265 * @offset: file offset of mapped area
266 * @len: size of mapped area
267 * @prot: mmap memory protection
268 * @idx: index of this mmap
269 * @tid: tid for a per-thread mmap (also set if there is only 1 tid on a per-cpu
271 * @cpu: cpu number for a per-cpu mmap otherwise %-1
273 struct auxtrace_mmap_params {
284 * struct auxtrace_record - callbacks for recording AUX area data.
285 * @recording_options: validate and process recording options
286 * @info_priv_size: return the size of the private data in auxtrace_info_event
287 * @info_fill: fill-in the private data in auxtrace_info_event
288 * @free: free this auxtrace record structure
289 * @snapshot_start: starting a snapshot
290 * @snapshot_finish: finishing a snapshot
291 * @find_snapshot: find data to snapshot within auxtrace mmap
292 * @parse_snapshot_options: parse snapshot options
293 * @reference: provide a 64-bit reference number for auxtrace_event
294 * @read_finish: called after reading from an auxtrace mmap
296 struct auxtrace_record {
297 int (*recording_options)(struct auxtrace_record *itr,
298 struct perf_evlist *evlist,
299 struct record_opts *opts);
300 size_t (*info_priv_size)(struct auxtrace_record *itr,
301 struct perf_evlist *evlist);
302 int (*info_fill)(struct auxtrace_record *itr,
303 struct perf_session *session,
304 struct auxtrace_info_event *auxtrace_info,
306 void (*free)(struct auxtrace_record *itr);
307 int (*snapshot_start)(struct auxtrace_record *itr);
308 int (*snapshot_finish)(struct auxtrace_record *itr);
309 int (*find_snapshot)(struct auxtrace_record *itr, int idx,
310 struct auxtrace_mmap *mm, unsigned char *data,
311 u64 *head, u64 *old);
312 int (*parse_snapshot_options)(struct auxtrace_record *itr,
313 struct record_opts *opts,
315 u64 (*reference)(struct auxtrace_record *itr);
316 int (*read_finish)(struct auxtrace_record *itr, int idx);
317 unsigned int alignment;
320 #ifdef HAVE_AUXTRACE_SUPPORT
323 * In snapshot mode the mmapped page is read-only which makes using
324 * __sync_val_compare_and_swap() problematic. However, snapshot mode expects
325 * the buffer is not updated while the snapshot is made (e.g. Intel PT disables
326 * the event) so there is not a race anyway.
328 static inline u64 auxtrace_mmap__read_snapshot_head(struct auxtrace_mmap *mm)
330 struct perf_event_mmap_page *pc = mm->userpg;
331 u64 head = ACCESS_ONCE(pc->aux_head);
333 /* Ensure all reads are done after we read the head */
338 static inline u64 auxtrace_mmap__read_head(struct auxtrace_mmap *mm)
340 struct perf_event_mmap_page *pc = mm->userpg;
341 #if BITS_PER_LONG == 64 || !defined(HAVE_SYNC_COMPARE_AND_SWAP_SUPPORT)
342 u64 head = ACCESS_ONCE(pc->aux_head);
344 u64 head = __sync_val_compare_and_swap(&pc->aux_head, 0, 0);
347 /* Ensure all reads are done after we read the head */
352 static inline void auxtrace_mmap__write_tail(struct auxtrace_mmap *mm, u64 tail)
354 struct perf_event_mmap_page *pc = mm->userpg;
355 #if BITS_PER_LONG != 64 && defined(HAVE_SYNC_COMPARE_AND_SWAP_SUPPORT)
359 /* Ensure all reads are done before we write the tail out */
361 #if BITS_PER_LONG == 64 || !defined(HAVE_SYNC_COMPARE_AND_SWAP_SUPPORT)
365 old_tail = __sync_val_compare_and_swap(&pc->aux_tail, 0, 0);
366 } while (!__sync_bool_compare_and_swap(&pc->aux_tail, old_tail, tail));
370 int auxtrace_mmap__mmap(struct auxtrace_mmap *mm,
371 struct auxtrace_mmap_params *mp,
372 void *userpg, int fd);
373 void auxtrace_mmap__munmap(struct auxtrace_mmap *mm);
374 void auxtrace_mmap_params__init(struct auxtrace_mmap_params *mp,
375 off_t auxtrace_offset,
376 unsigned int auxtrace_pages,
377 bool auxtrace_overwrite);
378 void auxtrace_mmap_params__set_idx(struct auxtrace_mmap_params *mp,
379 struct perf_evlist *evlist, int idx,
382 typedef int (*process_auxtrace_t)(struct perf_tool *tool,
383 union perf_event *event, void *data1,
384 size_t len1, void *data2, size_t len2);
386 int auxtrace_mmap__read(struct auxtrace_mmap *mm, struct auxtrace_record *itr,
387 struct perf_tool *tool, process_auxtrace_t fn);
389 int auxtrace_mmap__read_snapshot(struct auxtrace_mmap *mm,
390 struct auxtrace_record *itr,
391 struct perf_tool *tool, process_auxtrace_t fn,
392 size_t snapshot_size);
394 int auxtrace_queues__init(struct auxtrace_queues *queues);
395 int auxtrace_queues__add_event(struct auxtrace_queues *queues,
396 struct perf_session *session,
397 union perf_event *event, off_t data_offset,
398 struct auxtrace_buffer **buffer_ptr);
399 void auxtrace_queues__free(struct auxtrace_queues *queues);
400 int auxtrace_queues__process_index(struct auxtrace_queues *queues,
401 struct perf_session *session);
402 struct auxtrace_buffer *auxtrace_buffer__next(struct auxtrace_queue *queue,
403 struct auxtrace_buffer *buffer);
404 void *auxtrace_buffer__get_data(struct auxtrace_buffer *buffer, int fd);
405 void auxtrace_buffer__put_data(struct auxtrace_buffer *buffer);
406 void auxtrace_buffer__drop_data(struct auxtrace_buffer *buffer);
407 void auxtrace_buffer__free(struct auxtrace_buffer *buffer);
409 int auxtrace_heap__add(struct auxtrace_heap *heap, unsigned int queue_nr,
411 void auxtrace_heap__pop(struct auxtrace_heap *heap);
412 void auxtrace_heap__free(struct auxtrace_heap *heap);
414 struct auxtrace_cache_entry {
415 struct hlist_node hash;
419 struct auxtrace_cache *auxtrace_cache__new(unsigned int bits, size_t entry_size,
420 unsigned int limit_percent);
421 void auxtrace_cache__free(struct auxtrace_cache *auxtrace_cache);
422 void *auxtrace_cache__alloc_entry(struct auxtrace_cache *c);
423 void auxtrace_cache__free_entry(struct auxtrace_cache *c, void *entry);
424 int auxtrace_cache__add(struct auxtrace_cache *c, u32 key,
425 struct auxtrace_cache_entry *entry);
426 void *auxtrace_cache__lookup(struct auxtrace_cache *c, u32 key);
428 struct auxtrace_record *auxtrace_record__init(struct perf_evlist *evlist,
431 int auxtrace_parse_snapshot_options(struct auxtrace_record *itr,
432 struct record_opts *opts,
434 int auxtrace_record__options(struct auxtrace_record *itr,
435 struct perf_evlist *evlist,
436 struct record_opts *opts);
437 size_t auxtrace_record__info_priv_size(struct auxtrace_record *itr,
438 struct perf_evlist *evlist);
439 int auxtrace_record__info_fill(struct auxtrace_record *itr,
440 struct perf_session *session,
441 struct auxtrace_info_event *auxtrace_info,
443 void auxtrace_record__free(struct auxtrace_record *itr);
444 int auxtrace_record__snapshot_start(struct auxtrace_record *itr);
445 int auxtrace_record__snapshot_finish(struct auxtrace_record *itr);
446 int auxtrace_record__find_snapshot(struct auxtrace_record *itr, int idx,
447 struct auxtrace_mmap *mm,
448 unsigned char *data, u64 *head, u64 *old);
449 u64 auxtrace_record__reference(struct auxtrace_record *itr);
451 int auxtrace_index__auxtrace_event(struct list_head *head, union perf_event *event,
453 int auxtrace_index__write(int fd, struct list_head *head);
454 int auxtrace_index__process(int fd, u64 size, struct perf_session *session,
456 void auxtrace_index__free(struct list_head *head);
458 void auxtrace_synth_error(struct auxtrace_error_event *auxtrace_error, int type,
459 int code, int cpu, pid_t pid, pid_t tid, u64 ip,
462 int perf_event__synthesize_auxtrace_info(struct auxtrace_record *itr,
463 struct perf_tool *tool,
464 struct perf_session *session,
465 perf_event__handler_t process);
466 int perf_event__process_auxtrace_info(struct perf_tool *tool,
467 union perf_event *event,
468 struct perf_session *session);
469 s64 perf_event__process_auxtrace(struct perf_tool *tool,
470 union perf_event *event,
471 struct perf_session *session);
472 int perf_event__process_auxtrace_error(struct perf_tool *tool,
473 union perf_event *event,
474 struct perf_session *session);
475 int itrace_parse_synth_opts(const struct option *opt, const char *str,
477 void itrace_synth_opts__set_default(struct itrace_synth_opts *synth_opts);
479 size_t perf_event__fprintf_auxtrace_error(union perf_event *event, FILE *fp);
480 void perf_session__auxtrace_error_inc(struct perf_session *session,
481 union perf_event *event);
482 void events_stats__auxtrace_error_warn(const struct events_stats *stats);
484 static inline int auxtrace__process_event(struct perf_session *session,
485 union perf_event *event,
486 struct perf_sample *sample,
487 struct perf_tool *tool)
489 if (!session->auxtrace)
492 return session->auxtrace->process_event(session, event, sample, tool);
495 static inline int auxtrace__flush_events(struct perf_session *session,
496 struct perf_tool *tool)
498 if (!session->auxtrace)
501 return session->auxtrace->flush_events(session, tool);
504 static inline void auxtrace__free_events(struct perf_session *session)
506 if (!session->auxtrace)
509 return session->auxtrace->free_events(session);
512 static inline void auxtrace__free(struct perf_session *session)
514 if (!session->auxtrace)
517 return session->auxtrace->free(session);
522 static inline struct auxtrace_record *
523 auxtrace_record__init(struct perf_evlist *evlist __maybe_unused,
531 void auxtrace_record__free(struct auxtrace_record *itr __maybe_unused)
536 perf_event__synthesize_auxtrace_info(struct auxtrace_record *itr __maybe_unused,
537 struct perf_tool *tool __maybe_unused,
538 struct perf_session *session __maybe_unused,
539 perf_event__handler_t process __maybe_unused)
545 int auxtrace_record__options(struct auxtrace_record *itr __maybe_unused,
546 struct perf_evlist *evlist __maybe_unused,
547 struct record_opts *opts __maybe_unused)
552 #define perf_event__process_auxtrace_info 0
553 #define perf_event__process_auxtrace 0
554 #define perf_event__process_auxtrace_error 0
557 void perf_session__auxtrace_error_inc(struct perf_session *session
559 union perf_event *event
565 void events_stats__auxtrace_error_warn(const struct events_stats *stats
571 int itrace_parse_synth_opts(const struct option *opt __maybe_unused,
572 const char *str __maybe_unused,
573 int unset __maybe_unused)
575 pr_err("AUX area tracing not supported\n");
580 int auxtrace_parse_snapshot_options(struct auxtrace_record *itr __maybe_unused,
581 struct record_opts *opts __maybe_unused,
586 pr_err("AUX area tracing not supported\n");
591 int auxtrace__process_event(struct perf_session *session __maybe_unused,
592 union perf_event *event __maybe_unused,
593 struct perf_sample *sample __maybe_unused,
594 struct perf_tool *tool __maybe_unused)
600 int auxtrace__flush_events(struct perf_session *session __maybe_unused,
601 struct perf_tool *tool __maybe_unused)
607 void auxtrace__free_events(struct perf_session *session __maybe_unused)
612 void auxtrace_cache__free(struct auxtrace_cache *auxtrace_cache __maybe_unused)
617 void auxtrace__free(struct perf_session *session __maybe_unused)
622 int auxtrace_index__write(int fd __maybe_unused,
623 struct list_head *head __maybe_unused)
629 int auxtrace_index__process(int fd __maybe_unused,
630 u64 size __maybe_unused,
631 struct perf_session *session __maybe_unused,
632 bool needs_swap __maybe_unused)
638 void auxtrace_index__free(struct list_head *head __maybe_unused)
642 int auxtrace_mmap__mmap(struct auxtrace_mmap *mm,
643 struct auxtrace_mmap_params *mp,
644 void *userpg, int fd);
645 void auxtrace_mmap__munmap(struct auxtrace_mmap *mm);
646 void auxtrace_mmap_params__init(struct auxtrace_mmap_params *mp,
647 off_t auxtrace_offset,
648 unsigned int auxtrace_pages,
649 bool auxtrace_overwrite);
650 void auxtrace_mmap_params__set_idx(struct auxtrace_mmap_params *mp,
651 struct perf_evlist *evlist, int idx,