2 #ifndef _LINUX_TRACE_EVENT_H
3 #define _LINUX_TRACE_EVENT_H
5 #include <linux/ring_buffer.h>
6 #include <linux/trace_seq.h>
7 #include <linux/percpu.h>
8 #include <linux/hardirq.h>
9 #include <linux/perf_event.h>
10 #include <linux/tracepoint.h>
18 struct trace_print_flags {
23 struct trace_print_flags_u64 {
24 unsigned long long mask;
28 const char *trace_print_flags_seq(struct trace_seq *p, const char *delim,
30 const struct trace_print_flags *flag_array);
32 const char *trace_print_symbols_seq(struct trace_seq *p, unsigned long val,
33 const struct trace_print_flags *symbol_array);
35 #if BITS_PER_LONG == 32
36 const char *trace_print_symbols_seq_u64(struct trace_seq *p,
37 unsigned long long val,
38 const struct trace_print_flags_u64
42 const char *trace_print_bitmask_seq(struct trace_seq *p, void *bitmask_ptr,
43 unsigned int bitmask_size);
45 const char *trace_print_hex_seq(struct trace_seq *p,
46 const unsigned char *buf, int len);
48 const char *trace_print_array_seq(struct trace_seq *p,
49 const void *buf, int count,
52 struct trace_iterator;
55 int trace_raw_output_prep(struct trace_iterator *iter,
56 struct trace_event *event);
59 * The trace entry - the most basic unit of tracing. This is what
60 * is printed in the end as a single line in the trace output, such as:
62 * bash-15816 [01] 235.197585: idle_cpu <- irq_enter
67 unsigned char preempt_count;
71 #define TRACE_EVENT_TYPE_MAX \
72 ((1 << (sizeof(((struct trace_entry *)0)->type) * 8)) - 1)
75 * Trace iterator - used by printout routines who present trace
76 * results to users and which routines might sleep, etc:
78 struct trace_iterator {
79 struct trace_array *tr;
81 struct trace_buffer *trace_buffer;
85 struct ring_buffer_iter **buffer_iter;
86 unsigned long iter_flags;
88 /* trace_seq for __print_flags() and __print_symbolic() etc. */
89 struct trace_seq tmp_seq;
91 cpumask_var_t started;
93 /* it's true when current open file is snapshot */
96 /* The below is zeroed out in pipe_read */
98 struct trace_entry *ent;
99 unsigned long lost_events;
108 /* All new field here will be zeroed out in pipe_read */
111 enum trace_iter_flags {
112 TRACE_FILE_LAT_FMT = 1,
113 TRACE_FILE_ANNOTATE = 2,
114 TRACE_FILE_TIME_IN_NS = 4,
118 typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
119 int flags, struct trace_event *event);
121 struct trace_event_functions {
122 trace_print_func trace;
123 trace_print_func raw;
124 trace_print_func hex;
125 trace_print_func binary;
129 struct hlist_node node;
130 struct list_head list;
132 struct trace_event_functions *funcs;
135 extern int register_trace_event(struct trace_event *event);
136 extern int unregister_trace_event(struct trace_event *event);
138 /* Return values for print_line callback */
140 TRACE_TYPE_PARTIAL_LINE = 0, /* Retry after flushing the seq */
141 TRACE_TYPE_HANDLED = 1,
142 TRACE_TYPE_UNHANDLED = 2, /* Relay to other output functions */
143 TRACE_TYPE_NO_CONSUME = 3 /* Handled but ask to not consume */
147 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
148 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
149 * simplifies those functions and keeps them in sync.
151 static inline enum print_line_t trace_handle_return(struct trace_seq *s)
153 return trace_seq_has_overflowed(s) ?
154 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
157 void tracing_generic_entry_update(struct trace_entry *entry,
160 struct trace_event_file;
162 struct ring_buffer_event *
163 trace_event_buffer_lock_reserve(struct ring_buffer **current_buffer,
164 struct trace_event_file *trace_file,
165 int type, unsigned long len,
166 unsigned long flags, int pc);
167 struct ring_buffer_event *
168 trace_current_buffer_lock_reserve(struct ring_buffer **current_buffer,
169 int type, unsigned long len,
170 unsigned long flags, int pc);
171 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
172 struct ring_buffer_event *event,
173 unsigned long flags, int pc);
174 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
175 struct ring_buffer_event *event,
176 unsigned long flags, int pc);
177 void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer,
178 struct ring_buffer_event *event,
179 unsigned long flags, int pc,
180 struct pt_regs *regs);
181 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
182 struct ring_buffer_event *event);
184 void tracing_record_cmdline(struct task_struct *tsk);
186 int trace_output_call(struct trace_iterator *iter, char *name, char *fmt, ...);
192 TRACE_REG_UNREGISTER,
193 #ifdef CONFIG_PERF_EVENTS
194 TRACE_REG_PERF_REGISTER,
195 TRACE_REG_PERF_UNREGISTER,
197 TRACE_REG_PERF_CLOSE,
203 struct trace_event_call;
205 struct trace_event_class {
208 #ifdef CONFIG_PERF_EVENTS
211 int (*reg)(struct trace_event_call *event,
212 enum trace_reg type, void *data);
213 int (*define_fields)(struct trace_event_call *);
214 struct list_head *(*get_fields)(struct trace_event_call *);
215 struct list_head fields;
216 int (*raw_init)(struct trace_event_call *);
219 extern int trace_event_reg(struct trace_event_call *event,
220 enum trace_reg type, void *data);
222 struct trace_event_buffer {
223 struct ring_buffer *buffer;
224 struct ring_buffer_event *event;
225 struct trace_event_file *trace_file;
231 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
232 struct trace_event_file *trace_file,
235 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer);
238 TRACE_EVENT_FL_FILTERED_BIT,
239 TRACE_EVENT_FL_CAP_ANY_BIT,
240 TRACE_EVENT_FL_NO_SET_FILTER_BIT,
241 TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
242 TRACE_EVENT_FL_WAS_ENABLED_BIT,
243 TRACE_EVENT_FL_USE_CALL_FILTER_BIT,
244 TRACE_EVENT_FL_TRACEPOINT_BIT,
245 TRACE_EVENT_FL_KPROBE_BIT,
250 * FILTERED - The event has a filter attached
251 * CAP_ANY - Any user can enable for perf
252 * NO_SET_FILTER - Set when filter has error and is to be ignored
253 * IGNORE_ENABLE - For trace internal events, do not enable with debugfs file
254 * WAS_ENABLED - Set and stays set when an event was ever enabled
255 * (used for module unloading, if a module event is enabled,
256 * it is best to clear the buffers that used it).
257 * USE_CALL_FILTER - For trace internal events, don't use file filter
258 * TRACEPOINT - Event is a tracepoint
259 * KPROBE - Event is a kprobe
262 TRACE_EVENT_FL_FILTERED = (1 << TRACE_EVENT_FL_FILTERED_BIT),
263 TRACE_EVENT_FL_CAP_ANY = (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
264 TRACE_EVENT_FL_NO_SET_FILTER = (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
265 TRACE_EVENT_FL_IGNORE_ENABLE = (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
266 TRACE_EVENT_FL_WAS_ENABLED = (1 << TRACE_EVENT_FL_WAS_ENABLED_BIT),
267 TRACE_EVENT_FL_USE_CALL_FILTER = (1 << TRACE_EVENT_FL_USE_CALL_FILTER_BIT),
268 TRACE_EVENT_FL_TRACEPOINT = (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
269 TRACE_EVENT_FL_KPROBE = (1 << TRACE_EVENT_FL_KPROBE_BIT),
272 struct trace_event_call {
273 struct list_head list;
274 struct trace_event_class *class;
277 /* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
278 struct tracepoint *tp;
280 struct trace_event event;
282 struct event_filter *filter;
286 * bit 0: filter_active
287 * bit 1: allow trace by non root (cap any)
288 * bit 2: failed to apply filter
289 * bit 3: trace internal event (do not enable)
290 * bit 4: Event was enabled by module
291 * bit 5: use call filter rather than file filter
292 * bit 6: Event is a tracepoint
294 int flags; /* static flags of different events */
296 #ifdef CONFIG_PERF_EVENTS
298 struct hlist_head __percpu *perf_events;
299 struct bpf_prog *prog;
301 int (*perf_perm)(struct trace_event_call *,
302 struct perf_event *);
306 static inline const char *
307 trace_event_name(struct trace_event_call *call)
309 if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
310 return call->tp ? call->tp->name : NULL;
316 struct trace_subsystem_dir;
319 EVENT_FILE_FL_ENABLED_BIT,
320 EVENT_FILE_FL_RECORDED_CMD_BIT,
321 EVENT_FILE_FL_FILTERED_BIT,
322 EVENT_FILE_FL_NO_SET_FILTER_BIT,
323 EVENT_FILE_FL_SOFT_MODE_BIT,
324 EVENT_FILE_FL_SOFT_DISABLED_BIT,
325 EVENT_FILE_FL_TRIGGER_MODE_BIT,
326 EVENT_FILE_FL_TRIGGER_COND_BIT,
331 * ENABLED - The event is enabled
332 * RECORDED_CMD - The comms should be recorded at sched_switch
333 * FILTERED - The event has a filter attached
334 * NO_SET_FILTER - Set when filter has error and is to be ignored
335 * SOFT_MODE - The event is enabled/disabled by SOFT_DISABLED
336 * SOFT_DISABLED - When set, do not trace the event (even though its
337 * tracepoint may be enabled)
338 * TRIGGER_MODE - When set, invoke the triggers associated with the event
339 * TRIGGER_COND - When set, one or more triggers has an associated filter
342 EVENT_FILE_FL_ENABLED = (1 << EVENT_FILE_FL_ENABLED_BIT),
343 EVENT_FILE_FL_RECORDED_CMD = (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
344 EVENT_FILE_FL_FILTERED = (1 << EVENT_FILE_FL_FILTERED_BIT),
345 EVENT_FILE_FL_NO_SET_FILTER = (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
346 EVENT_FILE_FL_SOFT_MODE = (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
347 EVENT_FILE_FL_SOFT_DISABLED = (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
348 EVENT_FILE_FL_TRIGGER_MODE = (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
349 EVENT_FILE_FL_TRIGGER_COND = (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
352 struct trace_event_file {
353 struct list_head list;
354 struct trace_event_call *event_call;
355 struct event_filter *filter;
357 struct trace_array *tr;
358 struct trace_subsystem_dir *system;
359 struct list_head triggers;
364 * bit 1: enabled cmd record
365 * bit 2: enable/disable with the soft disable bit
366 * bit 3: soft disabled
367 * bit 4: trigger enabled
369 * Note: The bits must be set atomically to prevent races
370 * from other writers. Reads of flags do not need to be in
371 * sync as they occur in critical sections. But the way flags
372 * is currently used, these changes do not affect the code
373 * except that when a change is made, it may have a slight
374 * delay in propagating the changes to other CPUs due to
375 * caching and such. Which is mostly OK ;-)
378 atomic_t sm_ref; /* soft-mode reference counter */
379 atomic_t tm_ref; /* trigger-mode reference counter */
382 #define __TRACE_EVENT_FLAGS(name, value) \
383 static int __init trace_init_flags_##name(void) \
385 event_##name.flags |= value; \
388 early_initcall(trace_init_flags_##name);
390 #define __TRACE_EVENT_PERF_PERM(name, expr...) \
391 static int perf_perm_##name(struct trace_event_call *tp_event, \
392 struct perf_event *p_event) \
394 return ({ expr; }); \
396 static int __init trace_init_perf_perm_##name(void) \
398 event_##name.perf_perm = &perf_perm_##name; \
401 early_initcall(trace_init_perf_perm_##name);
403 #define PERF_MAX_TRACE_SIZE 2048
405 #define MAX_FILTER_STR_VAL 256 /* Should handle KSYM_SYMBOL_LEN */
407 enum event_trigger_type {
409 ETT_TRACE_ONOFF = (1 << 0),
410 ETT_SNAPSHOT = (1 << 1),
411 ETT_STACKTRACE = (1 << 2),
412 ETT_EVENT_ENABLE = (1 << 3),
415 extern int filter_match_preds(struct event_filter *filter, void *rec);
417 extern int filter_check_discard(struct trace_event_file *file, void *rec,
418 struct ring_buffer *buffer,
419 struct ring_buffer_event *event);
420 extern int call_filter_check_discard(struct trace_event_call *call, void *rec,
421 struct ring_buffer *buffer,
422 struct ring_buffer_event *event);
423 extern enum event_trigger_type event_triggers_call(struct trace_event_file *file,
425 extern void event_triggers_post_call(struct trace_event_file *file,
426 enum event_trigger_type tt);
429 * trace_trigger_soft_disabled - do triggers and test if soft disabled
430 * @file: The file pointer of the event to test
432 * If any triggers without filters are attached to this event, they
433 * will be called here. If the event is soft disabled and has no
434 * triggers that require testing the fields, it will return true,
438 trace_trigger_soft_disabled(struct trace_event_file *file)
440 unsigned long eflags = file->flags;
442 if (!(eflags & EVENT_FILE_FL_TRIGGER_COND)) {
443 if (eflags & EVENT_FILE_FL_TRIGGER_MODE)
444 event_triggers_call(file, NULL);
445 if (eflags & EVENT_FILE_FL_SOFT_DISABLED)
452 * Helper function for event_trigger_unlock_commit{_regs}().
453 * If there are event triggers attached to this event that requires
454 * filtering against its fields, then they wil be called as the
455 * entry already holds the field information of the current event.
457 * It also checks if the event should be discarded or not.
458 * It is to be discarded if the event is soft disabled and the
459 * event was only recorded to process triggers, or if the event
460 * filter is active and this event did not match the filters.
462 * Returns true if the event is discarded, false otherwise.
465 __event_trigger_test_discard(struct trace_event_file *file,
466 struct ring_buffer *buffer,
467 struct ring_buffer_event *event,
469 enum event_trigger_type *tt)
471 unsigned long eflags = file->flags;
473 if (eflags & EVENT_FILE_FL_TRIGGER_COND)
474 *tt = event_triggers_call(file, entry);
476 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags))
477 ring_buffer_discard_commit(buffer, event);
478 else if (!filter_check_discard(file, entry, buffer, event))
485 * event_trigger_unlock_commit - handle triggers and finish event commit
486 * @file: The file pointer assoctiated to the event
487 * @buffer: The ring buffer that the event is being written to
488 * @event: The event meta data in the ring buffer
489 * @entry: The event itself
490 * @irq_flags: The state of the interrupts at the start of the event
491 * @pc: The state of the preempt count at the start of the event.
493 * This is a helper function to handle triggers that require data
494 * from the event itself. It also tests the event against filters and
495 * if the event is soft disabled and should be discarded.
498 event_trigger_unlock_commit(struct trace_event_file *file,
499 struct ring_buffer *buffer,
500 struct ring_buffer_event *event,
501 void *entry, unsigned long irq_flags, int pc)
503 enum event_trigger_type tt = ETT_NONE;
505 if (!__event_trigger_test_discard(file, buffer, event, entry, &tt))
506 trace_buffer_unlock_commit(buffer, event, irq_flags, pc);
509 event_triggers_post_call(file, tt);
513 * event_trigger_unlock_commit_regs - handle triggers and finish event commit
514 * @file: The file pointer assoctiated to the event
515 * @buffer: The ring buffer that the event is being written to
516 * @event: The event meta data in the ring buffer
517 * @entry: The event itself
518 * @irq_flags: The state of the interrupts at the start of the event
519 * @pc: The state of the preempt count at the start of the event.
521 * This is a helper function to handle triggers that require data
522 * from the event itself. It also tests the event against filters and
523 * if the event is soft disabled and should be discarded.
525 * Same as event_trigger_unlock_commit() but calls
526 * trace_buffer_unlock_commit_regs() instead of trace_buffer_unlock_commit().
529 event_trigger_unlock_commit_regs(struct trace_event_file *file,
530 struct ring_buffer *buffer,
531 struct ring_buffer_event *event,
532 void *entry, unsigned long irq_flags, int pc,
533 struct pt_regs *regs)
535 enum event_trigger_type tt = ETT_NONE;
537 if (!__event_trigger_test_discard(file, buffer, event, entry, &tt))
538 trace_buffer_unlock_commit_regs(buffer, event,
539 irq_flags, pc, regs);
542 event_triggers_post_call(file, tt);
545 #ifdef CONFIG_BPF_SYSCALL
546 unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx);
548 static inline unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx)
556 FILTER_STATIC_STRING,
562 extern int trace_event_raw_init(struct trace_event_call *call);
563 extern int trace_define_field(struct trace_event_call *call, const char *type,
564 const char *name, int offset, int size,
565 int is_signed, int filter_type);
566 extern int trace_add_event_call(struct trace_event_call *call);
567 extern int trace_remove_event_call(struct trace_event_call *call);
569 #define is_signed_type(type) (((type)(-1)) < (type)1)
571 int trace_set_clr_event(const char *system, const char *event, int set);
574 * The double __builtin_constant_p is because gcc will give us an error
575 * if we try to allocate the static variable to fmt if it is not a
576 * constant. Even with the outer if statement optimizing out.
578 #define event_trace_printk(ip, fmt, args...) \
580 __trace_printk_check_format(fmt, ##args); \
581 tracing_record_cmdline(current); \
582 if (__builtin_constant_p(fmt)) { \
583 static const char *trace_printk_fmt \
584 __attribute__((section("__trace_printk_fmt"))) = \
585 __builtin_constant_p(fmt) ? fmt : NULL; \
587 __trace_bprintk(ip, trace_printk_fmt, ##args); \
589 __trace_printk(ip, fmt, ##args); \
592 #ifdef CONFIG_PERF_EVENTS
595 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
597 extern int perf_trace_init(struct perf_event *event);
598 extern void perf_trace_destroy(struct perf_event *event);
599 extern int perf_trace_add(struct perf_event *event, int flags);
600 extern void perf_trace_del(struct perf_event *event, int flags);
601 extern int ftrace_profile_set_filter(struct perf_event *event, int event_id,
603 extern void ftrace_profile_free_filter(struct perf_event *event);
604 extern void *perf_trace_buf_prepare(int size, unsigned short type,
605 struct pt_regs **regs, int *rctxp);
608 perf_trace_buf_submit(void *raw_data, int size, int rctx, u64 addr,
609 u64 count, struct pt_regs *regs, void *head,
610 struct task_struct *task)
612 perf_tp_event(addr, count, raw_data, size, regs, head, rctx, task);
616 #endif /* _LINUX_TRACE_EVENT_H */