1 /* SPDX-License-Identifier: GPL-2.0-only */
3 * auxtrace.h: AUX area trace support
4 * Copyright (c) 2013-2015, Intel Corporation.
7 #ifndef __PERF_AUXTRACE_H
8 #define __PERF_AUXTRACE_H
10 #include <sys/types.h>
14 #include <linux/list.h>
15 #include <linux/perf_event.h>
16 #include <linux/types.h>
17 #include <asm/bitsperlong.h>
31 struct auxtrace_info_event;
34 /* Auxtrace records must have the same alignment as perf event records */
35 #define PERF_AUXTRACE_RECORD_ALIGNMENT 8
38 PERF_AUXTRACE_UNKNOWN,
39 PERF_AUXTRACE_INTEL_PT,
40 PERF_AUXTRACE_INTEL_BTS,
42 PERF_AUXTRACE_ARM_SPE,
43 PERF_AUXTRACE_S390_CPUMSF,
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 * @default_no_sample: Default to no sampling.
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 * @ptwrites: whether to synthesize events for ptwrites
62 * @pwr_events: whether to synthesize power events
63 * @errors: whether to synthesize decoder error events
64 * @dont_decode: whether to skip decoding entirely
65 * @log: write a decoding log
66 * @calls: limit branch samples to calls (can be combined with @returns)
67 * @returns: limit branch samples to returns (can be combined with @calls)
68 * @callchain: add callchain to 'instructions' events
69 * @thread_stack: feed branches to the thread_stack
70 * @last_branch: add branch context to 'instruction' events
71 * @callchain_sz: maximum callchain size
72 * @last_branch_sz: branch context size
73 * @period: 'instructions' events period
74 * @period_type: 'instructions' events period type
75 * @initial_skip: skip N events at the beginning.
76 * @cpu_bitmap: CPUs for which to synthesize events, or NULL for all
77 * @ptime_range: time intervals to trace or NULL
78 * @range_num: number of time intervals to trace
80 struct itrace_synth_opts {
82 bool default_no_sample;
97 unsigned int callchain_sz;
98 unsigned int last_branch_sz;
99 unsigned long long period;
100 enum itrace_period_type period_type;
101 unsigned long initial_skip;
102 unsigned long *cpu_bitmap;
103 struct perf_time_interval *ptime_range;
108 * struct auxtrace_index_entry - indexes a AUX area tracing event within a
110 * @file_offset: offset within the perf.data file
111 * @sz: size of the event
113 struct auxtrace_index_entry {
118 #define PERF_AUXTRACE_INDEX_ENTRY_COUNT 256
121 * struct auxtrace_index - index of AUX area tracing events within a perf.data
123 * @list: linking a number of arrays of entries
124 * @nr: number of entries
125 * @entries: array of entries
127 struct auxtrace_index {
128 struct list_head list;
130 struct auxtrace_index_entry entries[PERF_AUXTRACE_INDEX_ENTRY_COUNT];
134 * struct auxtrace - session callbacks to allow AUX area data decoding.
135 * @process_event: lets the decoder see all session events
136 * @process_auxtrace_event: process a PERF_RECORD_AUXTRACE event
137 * @flush_events: process any remaining data
138 * @free_events: free resources associated with event processing
139 * @free: free resources associated with the session
142 int (*process_event)(struct perf_session *session,
143 union perf_event *event,
144 struct perf_sample *sample,
145 struct perf_tool *tool);
146 int (*process_auxtrace_event)(struct perf_session *session,
147 union perf_event *event,
148 struct perf_tool *tool);
149 int (*flush_events)(struct perf_session *session,
150 struct perf_tool *tool);
151 void (*free_events)(struct perf_session *session);
152 void (*free)(struct perf_session *session);
156 * struct auxtrace_buffer - a buffer containing AUX area tracing data.
157 * @list: buffers are queued in a list held by struct auxtrace_queue
158 * @size: size of the buffer in bytes
159 * @pid: in per-thread mode, the pid this buffer is associated with
160 * @tid: in per-thread mode, the tid this buffer is associated with
161 * @cpu: in per-cpu mode, the cpu this buffer is associated with
162 * @data: actual buffer data (can be null if the data has not been loaded)
163 * @data_offset: file offset at which the buffer can be read
164 * @mmap_addr: mmap address at which the buffer can be read
165 * @mmap_size: size of the mmap at @mmap_addr
166 * @data_needs_freeing: @data was malloc'd so free it when it is no longer
168 * @consecutive: the original data was split up and this buffer is consecutive
169 * to the previous buffer
170 * @offset: offset as determined by aux_head / aux_tail members of struct
171 * perf_event_mmap_page
172 * @reference: an implementation-specific reference determined when the data is
174 * @buffer_nr: used to number each buffer
175 * @use_size: implementation actually only uses this number of bytes
176 * @use_data: implementation actually only uses data starting at this address
178 struct auxtrace_buffer {
179 struct list_head list;
188 bool data_needs_freeing;
198 * struct auxtrace_queue - a queue of AUX area tracing data buffers.
199 * @head: head of buffer list
200 * @tid: in per-thread mode, the tid this queue is associated with
201 * @cpu: in per-cpu mode, the cpu this queue is associated with
202 * @set: %true once this queue has been dedicated to a specific thread or cpu
203 * @priv: implementation-specific data
205 struct auxtrace_queue {
206 struct list_head head;
214 * struct auxtrace_queues - an array of AUX area tracing queues.
215 * @queue_array: array of queues
216 * @nr_queues: number of queues
217 * @new_data: set whenever new data is queued
218 * @populated: queues have been fully populated using the auxtrace_index
219 * @next_buffer_nr: used to number each buffer
221 struct auxtrace_queues {
222 struct auxtrace_queue *queue_array;
223 unsigned int nr_queues;
230 * struct auxtrace_heap_item - element of struct auxtrace_heap.
231 * @queue_nr: queue number
232 * @ordinal: value used for sorting (lowest ordinal is top of the heap) expected
235 struct auxtrace_heap_item {
236 unsigned int queue_nr;
241 * struct auxtrace_heap - a heap suitable for sorting AUX area tracing queues.
242 * @heap_array: the heap
243 * @heap_cnt: the number of elements in the heap
244 * @heap_sz: maximum number of elements (grows as needed)
246 struct auxtrace_heap {
247 struct auxtrace_heap_item *heap_array;
248 unsigned int heap_cnt;
249 unsigned int heap_sz;
253 * struct auxtrace_mmap - records an mmap of the auxtrace buffer.
254 * @base: address of mapped area
255 * @userpg: pointer to buffer's perf_event_mmap_page
256 * @mask: %0 if @len is not a power of two, otherwise (@len - %1)
257 * @len: size of mapped area
258 * @prev: previous aux_head
259 * @idx: index of this mmap
260 * @tid: tid for a per-thread mmap (also set if there is only 1 tid on a per-cpu
262 * @cpu: cpu number for a per-cpu mmap otherwise %-1
264 struct auxtrace_mmap {
276 * struct auxtrace_mmap_params - parameters to set up struct auxtrace_mmap.
277 * @mask: %0 if @len is not a power of two, otherwise (@len - %1)
278 * @offset: file offset of mapped area
279 * @len: size of mapped area
280 * @prot: mmap memory protection
281 * @idx: index of this mmap
282 * @tid: tid for a per-thread mmap (also set if there is only 1 tid on a per-cpu
284 * @cpu: cpu number for a per-cpu mmap otherwise %-1
286 struct auxtrace_mmap_params {
297 * struct auxtrace_record - callbacks for recording AUX area data.
298 * @recording_options: validate and process recording options
299 * @info_priv_size: return the size of the private data in auxtrace_info_event
300 * @info_fill: fill-in the private data in auxtrace_info_event
301 * @free: free this auxtrace record structure
302 * @snapshot_start: starting a snapshot
303 * @snapshot_finish: finishing a snapshot
304 * @find_snapshot: find data to snapshot within auxtrace mmap
305 * @parse_snapshot_options: parse snapshot options
306 * @reference: provide a 64-bit reference number for auxtrace_event
307 * @read_finish: called after reading from an auxtrace mmap
308 * @alignment: alignment (if any) for AUX area data
310 struct auxtrace_record {
311 int (*recording_options)(struct auxtrace_record *itr,
312 struct perf_evlist *evlist,
313 struct record_opts *opts);
314 size_t (*info_priv_size)(struct auxtrace_record *itr,
315 struct perf_evlist *evlist);
316 int (*info_fill)(struct auxtrace_record *itr,
317 struct perf_session *session,
318 struct auxtrace_info_event *auxtrace_info,
320 void (*free)(struct auxtrace_record *itr);
321 int (*snapshot_start)(struct auxtrace_record *itr);
322 int (*snapshot_finish)(struct auxtrace_record *itr);
323 int (*find_snapshot)(struct auxtrace_record *itr, int idx,
324 struct auxtrace_mmap *mm, unsigned char *data,
325 u64 *head, u64 *old);
326 int (*parse_snapshot_options)(struct auxtrace_record *itr,
327 struct record_opts *opts,
329 u64 (*reference)(struct auxtrace_record *itr);
330 int (*read_finish)(struct auxtrace_record *itr, int idx);
331 unsigned int alignment;
335 * struct addr_filter - address filter.
337 * @range: true if it is a range filter
338 * @start: true if action is 'filter' or 'start'
339 * @action: 'filter', 'start' or 'stop' ('tracestop' is accepted but converted
341 * @sym_from: symbol name for the filter address
342 * @sym_to: symbol name that determines the filter size
343 * @sym_from_idx: selects n'th from symbols with the same name (0 means global
344 * and less than 0 means symbol must be unique)
345 * @sym_to_idx: same as @sym_from_idx but for @sym_to
346 * @addr: filter address
347 * @size: filter region size (for range filters)
348 * @filename: DSO file name or NULL for the kernel
349 * @str: allocated string that contains the other string members
352 struct list_head list;
356 const char *sym_from;
362 const char *filename;
367 * struct addr_filters - list of address filters.
368 * @head: list of address filters
369 * @cnt: number of address filters
371 struct addr_filters {
372 struct list_head head;
376 #ifdef HAVE_AUXTRACE_SUPPORT
379 * In snapshot mode the mmapped page is read-only which makes using
380 * __sync_val_compare_and_swap() problematic. However, snapshot mode expects
381 * the buffer is not updated while the snapshot is made (e.g. Intel PT disables
382 * the event) so there is not a race anyway.
384 static inline u64 auxtrace_mmap__read_snapshot_head(struct auxtrace_mmap *mm)
386 struct perf_event_mmap_page *pc = mm->userpg;
387 u64 head = READ_ONCE(pc->aux_head);
389 /* Ensure all reads are done after we read the head */
394 static inline u64 auxtrace_mmap__read_head(struct auxtrace_mmap *mm)
396 struct perf_event_mmap_page *pc = mm->userpg;
397 #if BITS_PER_LONG == 64 || !defined(HAVE_SYNC_COMPARE_AND_SWAP_SUPPORT)
398 u64 head = READ_ONCE(pc->aux_head);
400 u64 head = __sync_val_compare_and_swap(&pc->aux_head, 0, 0);
403 /* Ensure all reads are done after we read the head */
408 static inline void auxtrace_mmap__write_tail(struct auxtrace_mmap *mm, u64 tail)
410 struct perf_event_mmap_page *pc = mm->userpg;
411 #if BITS_PER_LONG != 64 && defined(HAVE_SYNC_COMPARE_AND_SWAP_SUPPORT)
415 /* Ensure all reads are done before we write the tail out */
417 #if BITS_PER_LONG == 64 || !defined(HAVE_SYNC_COMPARE_AND_SWAP_SUPPORT)
421 old_tail = __sync_val_compare_and_swap(&pc->aux_tail, 0, 0);
422 } while (!__sync_bool_compare_and_swap(&pc->aux_tail, old_tail, tail));
426 int auxtrace_mmap__mmap(struct auxtrace_mmap *mm,
427 struct auxtrace_mmap_params *mp,
428 void *userpg, int fd);
429 void auxtrace_mmap__munmap(struct auxtrace_mmap *mm);
430 void auxtrace_mmap_params__init(struct auxtrace_mmap_params *mp,
431 off_t auxtrace_offset,
432 unsigned int auxtrace_pages,
433 bool auxtrace_overwrite);
434 void auxtrace_mmap_params__set_idx(struct auxtrace_mmap_params *mp,
435 struct perf_evlist *evlist, int idx,
438 typedef int (*process_auxtrace_t)(struct perf_tool *tool,
439 struct perf_mmap *map,
440 union perf_event *event, void *data1,
441 size_t len1, void *data2, size_t len2);
443 int auxtrace_mmap__read(struct perf_mmap *map, struct auxtrace_record *itr,
444 struct perf_tool *tool, process_auxtrace_t fn);
446 int auxtrace_mmap__read_snapshot(struct perf_mmap *map,
447 struct auxtrace_record *itr,
448 struct perf_tool *tool, process_auxtrace_t fn,
449 size_t snapshot_size);
451 int auxtrace_queues__init(struct auxtrace_queues *queues);
452 int auxtrace_queues__add_event(struct auxtrace_queues *queues,
453 struct perf_session *session,
454 union perf_event *event, off_t data_offset,
455 struct auxtrace_buffer **buffer_ptr);
456 void auxtrace_queues__free(struct auxtrace_queues *queues);
457 int auxtrace_queues__process_index(struct auxtrace_queues *queues,
458 struct perf_session *session);
459 struct auxtrace_buffer *auxtrace_buffer__next(struct auxtrace_queue *queue,
460 struct auxtrace_buffer *buffer);
461 void *auxtrace_buffer__get_data(struct auxtrace_buffer *buffer, int fd);
462 void auxtrace_buffer__put_data(struct auxtrace_buffer *buffer);
463 void auxtrace_buffer__drop_data(struct auxtrace_buffer *buffer);
464 void auxtrace_buffer__free(struct auxtrace_buffer *buffer);
466 int auxtrace_heap__add(struct auxtrace_heap *heap, unsigned int queue_nr,
468 void auxtrace_heap__pop(struct auxtrace_heap *heap);
469 void auxtrace_heap__free(struct auxtrace_heap *heap);
471 struct auxtrace_cache_entry {
472 struct hlist_node hash;
476 struct auxtrace_cache *auxtrace_cache__new(unsigned int bits, size_t entry_size,
477 unsigned int limit_percent);
478 void auxtrace_cache__free(struct auxtrace_cache *auxtrace_cache);
479 void *auxtrace_cache__alloc_entry(struct auxtrace_cache *c);
480 void auxtrace_cache__free_entry(struct auxtrace_cache *c, void *entry);
481 int auxtrace_cache__add(struct auxtrace_cache *c, u32 key,
482 struct auxtrace_cache_entry *entry);
483 void *auxtrace_cache__lookup(struct auxtrace_cache *c, u32 key);
485 struct auxtrace_record *auxtrace_record__init(struct perf_evlist *evlist,
488 int auxtrace_parse_snapshot_options(struct auxtrace_record *itr,
489 struct record_opts *opts,
491 int auxtrace_record__options(struct auxtrace_record *itr,
492 struct perf_evlist *evlist,
493 struct record_opts *opts);
494 size_t auxtrace_record__info_priv_size(struct auxtrace_record *itr,
495 struct perf_evlist *evlist);
496 int auxtrace_record__info_fill(struct auxtrace_record *itr,
497 struct perf_session *session,
498 struct auxtrace_info_event *auxtrace_info,
500 void auxtrace_record__free(struct auxtrace_record *itr);
501 int auxtrace_record__snapshot_start(struct auxtrace_record *itr);
502 int auxtrace_record__snapshot_finish(struct auxtrace_record *itr);
503 int auxtrace_record__find_snapshot(struct auxtrace_record *itr, int idx,
504 struct auxtrace_mmap *mm,
505 unsigned char *data, u64 *head, u64 *old);
506 u64 auxtrace_record__reference(struct auxtrace_record *itr);
508 int auxtrace_index__auxtrace_event(struct list_head *head, union perf_event *event,
510 int auxtrace_index__write(int fd, struct list_head *head);
511 int auxtrace_index__process(int fd, u64 size, struct perf_session *session,
513 void auxtrace_index__free(struct list_head *head);
515 void auxtrace_synth_error(struct auxtrace_error_event *auxtrace_error, int type,
516 int code, int cpu, pid_t pid, pid_t tid, u64 ip,
517 const char *msg, u64 timestamp);
519 int perf_event__synthesize_auxtrace_info(struct auxtrace_record *itr,
520 struct perf_tool *tool,
521 struct perf_session *session,
522 perf_event__handler_t process);
523 int perf_event__process_auxtrace_info(struct perf_session *session,
524 union perf_event *event);
525 s64 perf_event__process_auxtrace(struct perf_session *session,
526 union perf_event *event);
527 int perf_event__process_auxtrace_error(struct perf_session *session,
528 union perf_event *event);
529 int itrace_parse_synth_opts(const struct option *opt, const char *str,
531 void itrace_synth_opts__set_default(struct itrace_synth_opts *synth_opts,
534 size_t perf_event__fprintf_auxtrace_error(union perf_event *event, FILE *fp);
535 void perf_session__auxtrace_error_inc(struct perf_session *session,
536 union perf_event *event);
537 void events_stats__auxtrace_error_warn(const struct events_stats *stats);
539 void addr_filters__init(struct addr_filters *filts);
540 void addr_filters__exit(struct addr_filters *filts);
541 int addr_filters__parse_bare_filter(struct addr_filters *filts,
543 int auxtrace_parse_filters(struct perf_evlist *evlist);
545 static inline int auxtrace__process_event(struct perf_session *session,
546 union perf_event *event,
547 struct perf_sample *sample,
548 struct perf_tool *tool)
550 if (!session->auxtrace)
553 return session->auxtrace->process_event(session, event, sample, tool);
556 static inline int auxtrace__flush_events(struct perf_session *session,
557 struct perf_tool *tool)
559 if (!session->auxtrace)
562 return session->auxtrace->flush_events(session, tool);
565 static inline void auxtrace__free_events(struct perf_session *session)
567 if (!session->auxtrace)
570 return session->auxtrace->free_events(session);
573 static inline void auxtrace__free(struct perf_session *session)
575 if (!session->auxtrace)
578 return session->auxtrace->free(session);
581 #define ITRACE_HELP \
582 " i: synthesize instructions events\n" \
583 " b: synthesize branches events\n" \
584 " c: synthesize branches events (calls only)\n" \
585 " r: synthesize branches events (returns only)\n" \
586 " x: synthesize transactions events\n" \
587 " w: synthesize ptwrite events\n" \
588 " p: synthesize power events\n" \
589 " e: synthesize error events\n" \
590 " d: create a debug log\n" \
591 " g[len]: synthesize a call chain (use with i or x)\n" \
592 " l[len]: synthesize last branch entries (use with i or x)\n" \
593 " sNUMBER: skip initial number of events\n" \
594 " PERIOD[ns|us|ms|i|t]: specify period to sample stream\n" \
595 " concatenate multiple options. Default is ibxwpe or cewp\n"
598 void itrace_synth_opts__set_time_range(struct itrace_synth_opts *opts,
599 struct perf_time_interval *ptime_range,
602 opts->ptime_range = ptime_range;
603 opts->range_num = range_num;
607 void itrace_synth_opts__clear_time_range(struct itrace_synth_opts *opts)
609 opts->ptime_range = NULL;
615 static inline struct auxtrace_record *
616 auxtrace_record__init(struct perf_evlist *evlist __maybe_unused,
624 void auxtrace_record__free(struct auxtrace_record *itr __maybe_unused)
629 perf_event__synthesize_auxtrace_info(struct auxtrace_record *itr __maybe_unused,
630 struct perf_tool *tool __maybe_unused,
631 struct perf_session *session __maybe_unused,
632 perf_event__handler_t process __maybe_unused)
638 int auxtrace_record__options(struct auxtrace_record *itr __maybe_unused,
639 struct perf_evlist *evlist __maybe_unused,
640 struct record_opts *opts __maybe_unused)
645 #define perf_event__process_auxtrace_info 0
646 #define perf_event__process_auxtrace 0
647 #define perf_event__process_auxtrace_error 0
650 void perf_session__auxtrace_error_inc(struct perf_session *session
652 union perf_event *event
658 void events_stats__auxtrace_error_warn(const struct events_stats *stats
664 int itrace_parse_synth_opts(const struct option *opt __maybe_unused,
665 const char *str __maybe_unused,
666 int unset __maybe_unused)
668 pr_err("AUX area tracing not supported\n");
673 int auxtrace_parse_snapshot_options(struct auxtrace_record *itr __maybe_unused,
674 struct record_opts *opts __maybe_unused,
679 pr_err("AUX area tracing not supported\n");
684 int auxtrace__process_event(struct perf_session *session __maybe_unused,
685 union perf_event *event __maybe_unused,
686 struct perf_sample *sample __maybe_unused,
687 struct perf_tool *tool __maybe_unused)
693 int auxtrace__flush_events(struct perf_session *session __maybe_unused,
694 struct perf_tool *tool __maybe_unused)
700 void auxtrace__free_events(struct perf_session *session __maybe_unused)
705 void auxtrace_cache__free(struct auxtrace_cache *auxtrace_cache __maybe_unused)
710 void auxtrace__free(struct perf_session *session __maybe_unused)
715 int auxtrace_index__write(int fd __maybe_unused,
716 struct list_head *head __maybe_unused)
722 int auxtrace_index__process(int fd __maybe_unused,
723 u64 size __maybe_unused,
724 struct perf_session *session __maybe_unused,
725 bool needs_swap __maybe_unused)
731 void auxtrace_index__free(struct list_head *head __maybe_unused)
736 int auxtrace_parse_filters(struct perf_evlist *evlist __maybe_unused)
741 int auxtrace_mmap__mmap(struct auxtrace_mmap *mm,
742 struct auxtrace_mmap_params *mp,
743 void *userpg, int fd);
744 void auxtrace_mmap__munmap(struct auxtrace_mmap *mm);
745 void auxtrace_mmap_params__init(struct auxtrace_mmap_params *mp,
746 off_t auxtrace_offset,
747 unsigned int auxtrace_pages,
748 bool auxtrace_overwrite);
749 void auxtrace_mmap_params__set_idx(struct auxtrace_mmap_params *mp,
750 struct perf_evlist *evlist, int idx,
753 #define ITRACE_HELP ""
756 void itrace_synth_opts__set_time_range(struct itrace_synth_opts *opts
758 struct perf_time_interval *ptime_range
760 int range_num __maybe_unused)
765 void itrace_synth_opts__clear_time_range(struct itrace_synth_opts *opts