2 #ifndef _LINUX_FTRACE_EVENT_H
3 #define _LINUX_FTRACE_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>
16 struct trace_print_flags
{
21 struct trace_print_flags_u64
{
22 unsigned long long mask
;
26 const char *ftrace_print_flags_seq(struct trace_seq
*p
, const char *delim
,
28 const struct trace_print_flags
*flag_array
);
30 const char *ftrace_print_symbols_seq(struct trace_seq
*p
, unsigned long val
,
31 const struct trace_print_flags
*symbol_array
);
33 #if BITS_PER_LONG == 32
34 const char *ftrace_print_symbols_seq_u64(struct trace_seq
*p
,
35 unsigned long long val
,
36 const struct trace_print_flags_u64
40 const char *ftrace_print_hex_seq(struct trace_seq
*p
,
41 const unsigned char *buf
, int len
);
43 struct trace_iterator
;
46 int ftrace_raw_output_prep(struct trace_iterator
*iter
,
47 struct trace_event
*event
);
50 * The trace entry - the most basic unit of tracing. This is what
51 * is printed in the end as a single line in the trace output, such as:
53 * bash-15816 [01] 235.197585: idle_cpu <- irq_enter
58 unsigned char preempt_count
;
62 #define FTRACE_MAX_EVENT \
63 ((1 << (sizeof(((struct trace_entry *)0)->type) * 8)) - 1)
66 * Trace iterator - used by printout routines who present trace
67 * results to users and which routines might sleep, etc:
69 struct trace_iterator
{
70 struct trace_array
*tr
;
72 struct trace_buffer
*trace_buffer
;
76 struct ring_buffer_iter
**buffer_iter
;
77 unsigned long iter_flags
;
79 /* trace_seq for __print_flags() and __print_symbolic() etc. */
80 struct trace_seq tmp_seq
;
82 cpumask_var_t started
;
84 /* it's true when current open file is snapshot */
87 /* The below is zeroed out in pipe_read */
89 struct trace_entry
*ent
;
90 unsigned long lost_events
;
99 /* All new field here will be zeroed out in pipe_read */
102 enum trace_iter_flags
{
103 TRACE_FILE_LAT_FMT
= 1,
104 TRACE_FILE_ANNOTATE
= 2,
105 TRACE_FILE_TIME_IN_NS
= 4,
109 typedef enum print_line_t (*trace_print_func
)(struct trace_iterator
*iter
,
110 int flags
, struct trace_event
*event
);
112 struct trace_event_functions
{
113 trace_print_func trace
;
114 trace_print_func raw
;
115 trace_print_func hex
;
116 trace_print_func binary
;
120 struct hlist_node node
;
121 struct list_head list
;
123 struct trace_event_functions
*funcs
;
126 extern int register_ftrace_event(struct trace_event
*event
);
127 extern int unregister_ftrace_event(struct trace_event
*event
);
129 /* Return values for print_line callback */
131 TRACE_TYPE_PARTIAL_LINE
= 0, /* Retry after flushing the seq */
132 TRACE_TYPE_HANDLED
= 1,
133 TRACE_TYPE_UNHANDLED
= 2, /* Relay to other output functions */
134 TRACE_TYPE_NO_CONSUME
= 3 /* Handled but ask to not consume */
137 void tracing_generic_entry_update(struct trace_entry
*entry
,
140 struct ftrace_event_file
;
142 struct ring_buffer_event
*
143 trace_event_buffer_lock_reserve(struct ring_buffer
**current_buffer
,
144 struct ftrace_event_file
*ftrace_file
,
145 int type
, unsigned long len
,
146 unsigned long flags
, int pc
);
147 struct ring_buffer_event
*
148 trace_current_buffer_lock_reserve(struct ring_buffer
**current_buffer
,
149 int type
, unsigned long len
,
150 unsigned long flags
, int pc
);
151 void trace_current_buffer_unlock_commit(struct ring_buffer
*buffer
,
152 struct ring_buffer_event
*event
,
153 unsigned long flags
, int pc
);
154 void trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
155 struct ring_buffer_event
*event
,
156 unsigned long flags
, int pc
);
157 void trace_buffer_unlock_commit_regs(struct ring_buffer
*buffer
,
158 struct ring_buffer_event
*event
,
159 unsigned long flags
, int pc
,
160 struct pt_regs
*regs
);
161 void trace_current_buffer_discard_commit(struct ring_buffer
*buffer
,
162 struct ring_buffer_event
*event
);
164 void tracing_record_cmdline(struct task_struct
*tsk
);
170 TRACE_REG_UNREGISTER
,
171 #ifdef CONFIG_PERF_EVENTS
172 TRACE_REG_PERF_REGISTER
,
173 TRACE_REG_PERF_UNREGISTER
,
175 TRACE_REG_PERF_CLOSE
,
181 struct ftrace_event_call
;
183 struct ftrace_event_class
{
186 #ifdef CONFIG_PERF_EVENTS
189 int (*reg
)(struct ftrace_event_call
*event
,
190 enum trace_reg type
, void *data
);
191 int (*define_fields
)(struct ftrace_event_call
*);
192 struct list_head
*(*get_fields
)(struct ftrace_event_call
*);
193 struct list_head fields
;
194 int (*raw_init
)(struct ftrace_event_call
*);
197 extern int ftrace_event_reg(struct ftrace_event_call
*event
,
198 enum trace_reg type
, void *data
);
201 TRACE_EVENT_FL_FILTERED_BIT
,
202 TRACE_EVENT_FL_CAP_ANY_BIT
,
203 TRACE_EVENT_FL_NO_SET_FILTER_BIT
,
204 TRACE_EVENT_FL_IGNORE_ENABLE_BIT
,
205 TRACE_EVENT_FL_WAS_ENABLED_BIT
,
206 TRACE_EVENT_FL_USE_CALL_FILTER_BIT
,
211 * FILTERED - The event has a filter attached
212 * CAP_ANY - Any user can enable for perf
213 * NO_SET_FILTER - Set when filter has error and is to be ignored
214 * IGNORE_ENABLE - For ftrace internal events, do not enable with debugfs file
215 * WAS_ENABLED - Set and stays set when an event was ever enabled
216 * (used for module unloading, if a module event is enabled,
217 * it is best to clear the buffers that used it).
218 * USE_CALL_FILTER - For ftrace internal events, don't use file filter
221 TRACE_EVENT_FL_FILTERED
= (1 << TRACE_EVENT_FL_FILTERED_BIT
),
222 TRACE_EVENT_FL_CAP_ANY
= (1 << TRACE_EVENT_FL_CAP_ANY_BIT
),
223 TRACE_EVENT_FL_NO_SET_FILTER
= (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT
),
224 TRACE_EVENT_FL_IGNORE_ENABLE
= (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT
),
225 TRACE_EVENT_FL_WAS_ENABLED
= (1 << TRACE_EVENT_FL_WAS_ENABLED_BIT
),
226 TRACE_EVENT_FL_USE_CALL_FILTER
= (1 << TRACE_EVENT_FL_USE_CALL_FILTER_BIT
),
229 struct ftrace_event_call
{
230 struct list_head list
;
231 struct ftrace_event_class
*class;
233 struct trace_event event
;
234 const char *print_fmt
;
235 struct event_filter
*filter
;
236 struct list_head
*files
;
240 * bit 0: filter_active
241 * bit 1: allow trace by non root (cap any)
242 * bit 2: failed to apply filter
243 * bit 3: ftrace internal event (do not enable)
244 * bit 4: Event was enabled by module
245 * bit 5: use call filter rather than file filter
247 int flags
; /* static flags of different events */
249 #ifdef CONFIG_PERF_EVENTS
251 struct hlist_head __percpu
*perf_events
;
253 int (*perf_perm
)(struct ftrace_event_call
*,
254 struct perf_event
*);
259 struct ftrace_subsystem_dir
;
262 FTRACE_EVENT_FL_ENABLED_BIT
,
263 FTRACE_EVENT_FL_RECORDED_CMD_BIT
,
264 FTRACE_EVENT_FL_FILTERED_BIT
,
265 FTRACE_EVENT_FL_NO_SET_FILTER_BIT
,
266 FTRACE_EVENT_FL_SOFT_MODE_BIT
,
267 FTRACE_EVENT_FL_SOFT_DISABLED_BIT
,
268 FTRACE_EVENT_FL_TRIGGER_MODE_BIT
,
269 FTRACE_EVENT_FL_TRIGGER_COND_BIT
,
273 * Ftrace event file flags:
274 * ENABLED - The event is enabled
275 * RECORDED_CMD - The comms should be recorded at sched_switch
276 * FILTERED - The event has a filter attached
277 * NO_SET_FILTER - Set when filter has error and is to be ignored
278 * SOFT_MODE - The event is enabled/disabled by SOFT_DISABLED
279 * SOFT_DISABLED - When set, do not trace the event (even though its
280 * tracepoint may be enabled)
281 * TRIGGER_MODE - When set, invoke the triggers associated with the event
282 * TRIGGER_COND - When set, one or more triggers has an associated filter
285 FTRACE_EVENT_FL_ENABLED
= (1 << FTRACE_EVENT_FL_ENABLED_BIT
),
286 FTRACE_EVENT_FL_RECORDED_CMD
= (1 << FTRACE_EVENT_FL_RECORDED_CMD_BIT
),
287 FTRACE_EVENT_FL_FILTERED
= (1 << FTRACE_EVENT_FL_FILTERED_BIT
),
288 FTRACE_EVENT_FL_NO_SET_FILTER
= (1 << FTRACE_EVENT_FL_NO_SET_FILTER_BIT
),
289 FTRACE_EVENT_FL_SOFT_MODE
= (1 << FTRACE_EVENT_FL_SOFT_MODE_BIT
),
290 FTRACE_EVENT_FL_SOFT_DISABLED
= (1 << FTRACE_EVENT_FL_SOFT_DISABLED_BIT
),
291 FTRACE_EVENT_FL_TRIGGER_MODE
= (1 << FTRACE_EVENT_FL_TRIGGER_MODE_BIT
),
292 FTRACE_EVENT_FL_TRIGGER_COND
= (1 << FTRACE_EVENT_FL_TRIGGER_COND_BIT
),
295 struct ftrace_event_file
{
296 struct list_head list
;
297 struct ftrace_event_call
*event_call
;
298 struct event_filter
*filter
;
300 struct trace_array
*tr
;
301 struct ftrace_subsystem_dir
*system
;
302 struct list_head triggers
;
307 * bit 1: enabled cmd record
308 * bit 2: enable/disable with the soft disable bit
309 * bit 3: soft disabled
310 * bit 4: trigger enabled
312 * Note: The bits must be set atomically to prevent races
313 * from other writers. Reads of flags do not need to be in
314 * sync as they occur in critical sections. But the way flags
315 * is currently used, these changes do not affect the code
316 * except that when a change is made, it may have a slight
317 * delay in propagating the changes to other CPUs due to
318 * caching and such. Which is mostly OK ;-)
321 atomic_t sm_ref
; /* soft-mode reference counter */
322 atomic_t tm_ref
; /* trigger-mode reference counter */
325 #define __TRACE_EVENT_FLAGS(name, value) \
326 static int __init trace_init_flags_##name(void) \
328 event_##name.flags = value; \
331 early_initcall(trace_init_flags_##name);
333 #define __TRACE_EVENT_PERF_PERM(name, expr...) \
334 static int perf_perm_##name(struct ftrace_event_call *tp_event, \
335 struct perf_event *p_event) \
337 return ({ expr; }); \
339 static int __init trace_init_perf_perm_##name(void) \
341 event_##name.perf_perm = &perf_perm_##name; \
344 early_initcall(trace_init_perf_perm_##name);
346 #define PERF_MAX_TRACE_SIZE 2048
348 #define MAX_FILTER_STR_VAL 256 /* Should handle KSYM_SYMBOL_LEN */
350 enum event_trigger_type
{
352 ETT_TRACE_ONOFF
= (1 << 0),
353 ETT_SNAPSHOT
= (1 << 1),
354 ETT_STACKTRACE
= (1 << 2),
355 ETT_EVENT_ENABLE
= (1 << 3),
358 extern void destroy_preds(struct ftrace_event_file
*file
);
359 extern void destroy_call_preds(struct ftrace_event_call
*call
);
360 extern int filter_match_preds(struct event_filter
*filter
, void *rec
);
362 extern int filter_check_discard(struct ftrace_event_file
*file
, void *rec
,
363 struct ring_buffer
*buffer
,
364 struct ring_buffer_event
*event
);
365 extern int call_filter_check_discard(struct ftrace_event_call
*call
, void *rec
,
366 struct ring_buffer
*buffer
,
367 struct ring_buffer_event
*event
);
368 extern enum event_trigger_type
event_triggers_call(struct ftrace_event_file
*file
,
370 extern void event_triggers_post_call(struct ftrace_event_file
*file
,
371 enum event_trigger_type tt
);
374 * ftrace_trigger_soft_disabled - do triggers and test if soft disabled
375 * @file: The file pointer of the event to test
377 * If any triggers without filters are attached to this event, they
378 * will be called here. If the event is soft disabled and has no
379 * triggers that require testing the fields, it will return true,
383 ftrace_trigger_soft_disabled(struct ftrace_event_file
*file
)
385 unsigned long eflags
= file
->flags
;
387 if (!(eflags
& FTRACE_EVENT_FL_TRIGGER_COND
)) {
388 if (eflags
& FTRACE_EVENT_FL_TRIGGER_MODE
)
389 event_triggers_call(file
, NULL
);
390 if (eflags
& FTRACE_EVENT_FL_SOFT_DISABLED
)
397 * Helper function for event_trigger_unlock_commit{_regs}().
398 * If there are event triggers attached to this event that requires
399 * filtering against its fields, then they wil be called as the
400 * entry already holds the field information of the current event.
402 * It also checks if the event should be discarded or not.
403 * It is to be discarded if the event is soft disabled and the
404 * event was only recorded to process triggers, or if the event
405 * filter is active and this event did not match the filters.
407 * Returns true if the event is discarded, false otherwise.
410 __event_trigger_test_discard(struct ftrace_event_file
*file
,
411 struct ring_buffer
*buffer
,
412 struct ring_buffer_event
*event
,
414 enum event_trigger_type
*tt
)
416 unsigned long eflags
= file
->flags
;
418 if (eflags
& FTRACE_EVENT_FL_TRIGGER_COND
)
419 *tt
= event_triggers_call(file
, entry
);
421 if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT
, &file
->flags
))
422 ring_buffer_discard_commit(buffer
, event
);
423 else if (!filter_check_discard(file
, entry
, buffer
, event
))
430 * event_trigger_unlock_commit - handle triggers and finish event commit
431 * @file: The file pointer assoctiated to the event
432 * @buffer: The ring buffer that the event is being written to
433 * @event: The event meta data in the ring buffer
434 * @entry: The event itself
435 * @irq_flags: The state of the interrupts at the start of the event
436 * @pc: The state of the preempt count at the start of the event.
438 * This is a helper function to handle triggers that require data
439 * from the event itself. It also tests the event against filters and
440 * if the event is soft disabled and should be discarded.
443 event_trigger_unlock_commit(struct ftrace_event_file
*file
,
444 struct ring_buffer
*buffer
,
445 struct ring_buffer_event
*event
,
446 void *entry
, unsigned long irq_flags
, int pc
)
448 enum event_trigger_type tt
= ETT_NONE
;
450 if (!__event_trigger_test_discard(file
, buffer
, event
, entry
, &tt
))
451 trace_buffer_unlock_commit(buffer
, event
, irq_flags
, pc
);
454 event_triggers_post_call(file
, tt
);
458 * event_trigger_unlock_commit_regs - handle triggers and finish event commit
459 * @file: The file pointer assoctiated to the event
460 * @buffer: The ring buffer that the event is being written to
461 * @event: The event meta data in the ring buffer
462 * @entry: The event itself
463 * @irq_flags: The state of the interrupts at the start of the event
464 * @pc: The state of the preempt count at the start of the event.
466 * This is a helper function to handle triggers that require data
467 * from the event itself. It also tests the event against filters and
468 * if the event is soft disabled and should be discarded.
470 * Same as event_trigger_unlock_commit() but calls
471 * trace_buffer_unlock_commit_regs() instead of trace_buffer_unlock_commit().
474 event_trigger_unlock_commit_regs(struct ftrace_event_file
*file
,
475 struct ring_buffer
*buffer
,
476 struct ring_buffer_event
*event
,
477 void *entry
, unsigned long irq_flags
, int pc
,
478 struct pt_regs
*regs
)
480 enum event_trigger_type tt
= ETT_NONE
;
482 if (!__event_trigger_test_discard(file
, buffer
, event
, entry
, &tt
))
483 trace_buffer_unlock_commit_regs(buffer
, event
,
484 irq_flags
, pc
, regs
);
487 event_triggers_post_call(file
, tt
);
492 FILTER_STATIC_STRING
,
498 #define EVENT_STORAGE_SIZE 128
499 extern struct mutex event_storage_mutex
;
500 extern char event_storage
[EVENT_STORAGE_SIZE
];
502 extern int trace_event_raw_init(struct ftrace_event_call
*call
);
503 extern int trace_define_field(struct ftrace_event_call
*call
, const char *type
,
504 const char *name
, int offset
, int size
,
505 int is_signed
, int filter_type
);
506 extern int trace_add_event_call(struct ftrace_event_call
*call
);
507 extern int trace_remove_event_call(struct ftrace_event_call
*call
);
509 #define is_signed_type(type) (((type)(-1)) < (type)1)
511 int trace_set_clr_event(const char *system
, const char *event
, int set
);
514 * The double __builtin_constant_p is because gcc will give us an error
515 * if we try to allocate the static variable to fmt if it is not a
516 * constant. Even with the outer if statement optimizing out.
518 #define event_trace_printk(ip, fmt, args...) \
520 __trace_printk_check_format(fmt, ##args); \
521 tracing_record_cmdline(current); \
522 if (__builtin_constant_p(fmt)) { \
523 static const char *trace_printk_fmt \
524 __attribute__((section("__trace_printk_fmt"))) = \
525 __builtin_constant_p(fmt) ? fmt : NULL; \
527 __trace_bprintk(ip, trace_printk_fmt, ##args); \
529 __trace_printk(ip, fmt, ##args); \
533 * tracepoint_string - register constant persistent string to trace system
534 * @str - a constant persistent string that will be referenced in tracepoints
536 * If constant strings are being used in tracepoints, it is faster and
537 * more efficient to just save the pointer to the string and reference
538 * that with a printf "%s" instead of saving the string in the ring buffer
539 * and wasting space and time.
541 * The problem with the above approach is that userspace tools that read
542 * the binary output of the trace buffers do not have access to the string.
543 * Instead they just show the address of the string which is not very
546 * With tracepoint_string(), the string will be registered to the tracing
547 * system and exported to userspace via the debugfs/tracing/printk_formats
548 * file that maps the string address to the string text. This way userspace
549 * tools that read the binary buffers have a way to map the pointers to
550 * the ASCII strings they represent.
552 * The @str used must be a constant string and persistent as it would not
553 * make sense to show a string that no longer exists. But it is still fine
554 * to be used with modules, because when modules are unloaded, if they
555 * had tracepoints, the ring buffers are cleared too. As long as the string
556 * does not change during the life of the module, it is fine to use
557 * tracepoint_string() within a module.
559 #define tracepoint_string(str) \
561 static const char *___tp_str __tracepoint_string = str; \
564 #define __tracepoint_string __attribute__((section("__tracepoint_str")))
566 #ifdef CONFIG_PERF_EVENTS
569 DECLARE_PER_CPU(struct pt_regs
, perf_trace_regs
);
571 extern int perf_trace_init(struct perf_event
*event
);
572 extern void perf_trace_destroy(struct perf_event
*event
);
573 extern int perf_trace_add(struct perf_event
*event
, int flags
);
574 extern void perf_trace_del(struct perf_event
*event
, int flags
);
575 extern int ftrace_profile_set_filter(struct perf_event
*event
, int event_id
,
577 extern void ftrace_profile_free_filter(struct perf_event
*event
);
578 extern void *perf_trace_buf_prepare(int size
, unsigned short type
,
579 struct pt_regs
*regs
, int *rctxp
);
582 perf_trace_buf_submit(void *raw_data
, int size
, int rctx
, u64 addr
,
583 u64 count
, struct pt_regs
*regs
, void *head
,
584 struct task_struct
*task
)
586 perf_tp_event(addr
, count
, raw_data
, size
, regs
, head
, rctx
, task
);
590 #endif /* _LINUX_FTRACE_EVENT_H */