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
);
166 int ftrace_output_call(struct trace_iterator
*iter
, char *name
, char *fmt
, ...);
172 TRACE_REG_UNREGISTER
,
173 #ifdef CONFIG_PERF_EVENTS
174 TRACE_REG_PERF_REGISTER
,
175 TRACE_REG_PERF_UNREGISTER
,
177 TRACE_REG_PERF_CLOSE
,
183 struct ftrace_event_call
;
185 struct ftrace_event_class
{
188 #ifdef CONFIG_PERF_EVENTS
191 int (*reg
)(struct ftrace_event_call
*event
,
192 enum trace_reg type
, void *data
);
193 int (*define_fields
)(struct ftrace_event_call
*);
194 struct list_head
*(*get_fields
)(struct ftrace_event_call
*);
195 struct list_head fields
;
196 int (*raw_init
)(struct ftrace_event_call
*);
199 extern int ftrace_event_reg(struct ftrace_event_call
*event
,
200 enum trace_reg type
, void *data
);
202 int ftrace_output_event(struct trace_iterator
*iter
, struct ftrace_event_call
*event
,
205 int ftrace_event_define_field(struct ftrace_event_call
*call
,
206 char *type
, int len
, char *item
, int offset
,
207 int field_size
, int sign
, int filter
);
209 struct ftrace_event_buffer
{
210 struct ring_buffer
*buffer
;
211 struct ring_buffer_event
*event
;
212 struct ftrace_event_file
*ftrace_file
;
218 void *ftrace_event_buffer_reserve(struct ftrace_event_buffer
*fbuffer
,
219 struct ftrace_event_file
*ftrace_file
,
222 void ftrace_event_buffer_commit(struct ftrace_event_buffer
*fbuffer
);
225 TRACE_EVENT_FL_FILTERED_BIT
,
226 TRACE_EVENT_FL_CAP_ANY_BIT
,
227 TRACE_EVENT_FL_NO_SET_FILTER_BIT
,
228 TRACE_EVENT_FL_IGNORE_ENABLE_BIT
,
229 TRACE_EVENT_FL_WAS_ENABLED_BIT
,
230 TRACE_EVENT_FL_USE_CALL_FILTER_BIT
,
235 * FILTERED - The event has a filter attached
236 * CAP_ANY - Any user can enable for perf
237 * NO_SET_FILTER - Set when filter has error and is to be ignored
238 * IGNORE_ENABLE - For ftrace internal events, do not enable with debugfs file
239 * WAS_ENABLED - Set and stays set when an event was ever enabled
240 * (used for module unloading, if a module event is enabled,
241 * it is best to clear the buffers that used it).
242 * USE_CALL_FILTER - For ftrace internal events, don't use file filter
245 TRACE_EVENT_FL_FILTERED
= (1 << TRACE_EVENT_FL_FILTERED_BIT
),
246 TRACE_EVENT_FL_CAP_ANY
= (1 << TRACE_EVENT_FL_CAP_ANY_BIT
),
247 TRACE_EVENT_FL_NO_SET_FILTER
= (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT
),
248 TRACE_EVENT_FL_IGNORE_ENABLE
= (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT
),
249 TRACE_EVENT_FL_WAS_ENABLED
= (1 << TRACE_EVENT_FL_WAS_ENABLED_BIT
),
250 TRACE_EVENT_FL_USE_CALL_FILTER
= (1 << TRACE_EVENT_FL_USE_CALL_FILTER_BIT
),
253 struct ftrace_event_call
{
254 struct list_head list
;
255 struct ftrace_event_class
*class;
257 struct trace_event event
;
258 const char *print_fmt
;
259 struct event_filter
*filter
;
260 struct list_head
*files
;
264 * bit 0: filter_active
265 * bit 1: allow trace by non root (cap any)
266 * bit 2: failed to apply filter
267 * bit 3: ftrace internal event (do not enable)
268 * bit 4: Event was enabled by module
269 * bit 5: use call filter rather than file filter
271 int flags
; /* static flags of different events */
273 #ifdef CONFIG_PERF_EVENTS
275 struct hlist_head __percpu
*perf_events
;
277 int (*perf_perm
)(struct ftrace_event_call
*,
278 struct perf_event
*);
283 struct ftrace_subsystem_dir
;
286 FTRACE_EVENT_FL_ENABLED_BIT
,
287 FTRACE_EVENT_FL_RECORDED_CMD_BIT
,
288 FTRACE_EVENT_FL_FILTERED_BIT
,
289 FTRACE_EVENT_FL_NO_SET_FILTER_BIT
,
290 FTRACE_EVENT_FL_SOFT_MODE_BIT
,
291 FTRACE_EVENT_FL_SOFT_DISABLED_BIT
,
292 FTRACE_EVENT_FL_TRIGGER_MODE_BIT
,
293 FTRACE_EVENT_FL_TRIGGER_COND_BIT
,
297 * Ftrace event file flags:
298 * ENABLED - The event is enabled
299 * RECORDED_CMD - The comms should be recorded at sched_switch
300 * FILTERED - The event has a filter attached
301 * NO_SET_FILTER - Set when filter has error and is to be ignored
302 * SOFT_MODE - The event is enabled/disabled by SOFT_DISABLED
303 * SOFT_DISABLED - When set, do not trace the event (even though its
304 * tracepoint may be enabled)
305 * TRIGGER_MODE - When set, invoke the triggers associated with the event
306 * TRIGGER_COND - When set, one or more triggers has an associated filter
309 FTRACE_EVENT_FL_ENABLED
= (1 << FTRACE_EVENT_FL_ENABLED_BIT
),
310 FTRACE_EVENT_FL_RECORDED_CMD
= (1 << FTRACE_EVENT_FL_RECORDED_CMD_BIT
),
311 FTRACE_EVENT_FL_FILTERED
= (1 << FTRACE_EVENT_FL_FILTERED_BIT
),
312 FTRACE_EVENT_FL_NO_SET_FILTER
= (1 << FTRACE_EVENT_FL_NO_SET_FILTER_BIT
),
313 FTRACE_EVENT_FL_SOFT_MODE
= (1 << FTRACE_EVENT_FL_SOFT_MODE_BIT
),
314 FTRACE_EVENT_FL_SOFT_DISABLED
= (1 << FTRACE_EVENT_FL_SOFT_DISABLED_BIT
),
315 FTRACE_EVENT_FL_TRIGGER_MODE
= (1 << FTRACE_EVENT_FL_TRIGGER_MODE_BIT
),
316 FTRACE_EVENT_FL_TRIGGER_COND
= (1 << FTRACE_EVENT_FL_TRIGGER_COND_BIT
),
319 struct ftrace_event_file
{
320 struct list_head list
;
321 struct ftrace_event_call
*event_call
;
322 struct event_filter
*filter
;
324 struct trace_array
*tr
;
325 struct ftrace_subsystem_dir
*system
;
326 struct list_head triggers
;
331 * bit 1: enabled cmd record
332 * bit 2: enable/disable with the soft disable bit
333 * bit 3: soft disabled
334 * bit 4: trigger enabled
336 * Note: The bits must be set atomically to prevent races
337 * from other writers. Reads of flags do not need to be in
338 * sync as they occur in critical sections. But the way flags
339 * is currently used, these changes do not affect the code
340 * except that when a change is made, it may have a slight
341 * delay in propagating the changes to other CPUs due to
342 * caching and such. Which is mostly OK ;-)
345 atomic_t sm_ref
; /* soft-mode reference counter */
346 atomic_t tm_ref
; /* trigger-mode reference counter */
349 #define __TRACE_EVENT_FLAGS(name, value) \
350 static int __init trace_init_flags_##name(void) \
352 event_##name.flags = value; \
355 early_initcall(trace_init_flags_##name);
357 #define __TRACE_EVENT_PERF_PERM(name, expr...) \
358 static int perf_perm_##name(struct ftrace_event_call *tp_event, \
359 struct perf_event *p_event) \
361 return ({ expr; }); \
363 static int __init trace_init_perf_perm_##name(void) \
365 event_##name.perf_perm = &perf_perm_##name; \
368 early_initcall(trace_init_perf_perm_##name);
370 #define PERF_MAX_TRACE_SIZE 2048
372 #define MAX_FILTER_STR_VAL 256 /* Should handle KSYM_SYMBOL_LEN */
374 enum event_trigger_type
{
376 ETT_TRACE_ONOFF
= (1 << 0),
377 ETT_SNAPSHOT
= (1 << 1),
378 ETT_STACKTRACE
= (1 << 2),
379 ETT_EVENT_ENABLE
= (1 << 3),
382 extern void destroy_preds(struct ftrace_event_file
*file
);
383 extern void destroy_call_preds(struct ftrace_event_call
*call
);
384 extern int filter_match_preds(struct event_filter
*filter
, void *rec
);
386 extern int filter_check_discard(struct ftrace_event_file
*file
, void *rec
,
387 struct ring_buffer
*buffer
,
388 struct ring_buffer_event
*event
);
389 extern int call_filter_check_discard(struct ftrace_event_call
*call
, void *rec
,
390 struct ring_buffer
*buffer
,
391 struct ring_buffer_event
*event
);
392 extern enum event_trigger_type
event_triggers_call(struct ftrace_event_file
*file
,
394 extern void event_triggers_post_call(struct ftrace_event_file
*file
,
395 enum event_trigger_type tt
);
398 * ftrace_trigger_soft_disabled - do triggers and test if soft disabled
399 * @file: The file pointer of the event to test
401 * If any triggers without filters are attached to this event, they
402 * will be called here. If the event is soft disabled and has no
403 * triggers that require testing the fields, it will return true,
407 ftrace_trigger_soft_disabled(struct ftrace_event_file
*file
)
409 unsigned long eflags
= file
->flags
;
411 if (!(eflags
& FTRACE_EVENT_FL_TRIGGER_COND
)) {
412 if (eflags
& FTRACE_EVENT_FL_TRIGGER_MODE
)
413 event_triggers_call(file
, NULL
);
414 if (eflags
& FTRACE_EVENT_FL_SOFT_DISABLED
)
421 * Helper function for event_trigger_unlock_commit{_regs}().
422 * If there are event triggers attached to this event that requires
423 * filtering against its fields, then they wil be called as the
424 * entry already holds the field information of the current event.
426 * It also checks if the event should be discarded or not.
427 * It is to be discarded if the event is soft disabled and the
428 * event was only recorded to process triggers, or if the event
429 * filter is active and this event did not match the filters.
431 * Returns true if the event is discarded, false otherwise.
434 __event_trigger_test_discard(struct ftrace_event_file
*file
,
435 struct ring_buffer
*buffer
,
436 struct ring_buffer_event
*event
,
438 enum event_trigger_type
*tt
)
440 unsigned long eflags
= file
->flags
;
442 if (eflags
& FTRACE_EVENT_FL_TRIGGER_COND
)
443 *tt
= event_triggers_call(file
, entry
);
445 if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT
, &file
->flags
))
446 ring_buffer_discard_commit(buffer
, event
);
447 else if (!filter_check_discard(file
, entry
, buffer
, event
))
454 * event_trigger_unlock_commit - handle triggers and finish event commit
455 * @file: The file pointer assoctiated to the event
456 * @buffer: The ring buffer that the event is being written to
457 * @event: The event meta data in the ring buffer
458 * @entry: The event itself
459 * @irq_flags: The state of the interrupts at the start of the event
460 * @pc: The state of the preempt count at the start of the event.
462 * This is a helper function to handle triggers that require data
463 * from the event itself. It also tests the event against filters and
464 * if the event is soft disabled and should be discarded.
467 event_trigger_unlock_commit(struct ftrace_event_file
*file
,
468 struct ring_buffer
*buffer
,
469 struct ring_buffer_event
*event
,
470 void *entry
, unsigned long irq_flags
, int pc
)
472 enum event_trigger_type tt
= ETT_NONE
;
474 if (!__event_trigger_test_discard(file
, buffer
, event
, entry
, &tt
))
475 trace_buffer_unlock_commit(buffer
, event
, irq_flags
, pc
);
478 event_triggers_post_call(file
, tt
);
482 * event_trigger_unlock_commit_regs - handle triggers and finish event commit
483 * @file: The file pointer assoctiated to the event
484 * @buffer: The ring buffer that the event is being written to
485 * @event: The event meta data in the ring buffer
486 * @entry: The event itself
487 * @irq_flags: The state of the interrupts at the start of the event
488 * @pc: The state of the preempt count at the start of the event.
490 * This is a helper function to handle triggers that require data
491 * from the event itself. It also tests the event against filters and
492 * if the event is soft disabled and should be discarded.
494 * Same as event_trigger_unlock_commit() but calls
495 * trace_buffer_unlock_commit_regs() instead of trace_buffer_unlock_commit().
498 event_trigger_unlock_commit_regs(struct ftrace_event_file
*file
,
499 struct ring_buffer
*buffer
,
500 struct ring_buffer_event
*event
,
501 void *entry
, unsigned long irq_flags
, int pc
,
502 struct pt_regs
*regs
)
504 enum event_trigger_type tt
= ETT_NONE
;
506 if (!__event_trigger_test_discard(file
, buffer
, event
, entry
, &tt
))
507 trace_buffer_unlock_commit_regs(buffer
, event
,
508 irq_flags
, pc
, regs
);
511 event_triggers_post_call(file
, tt
);
516 FILTER_STATIC_STRING
,
522 extern int trace_event_raw_init(struct ftrace_event_call
*call
);
523 extern int trace_define_field(struct ftrace_event_call
*call
, const char *type
,
524 const char *name
, int offset
, int size
,
525 int is_signed
, int filter_type
);
526 extern int trace_add_event_call(struct ftrace_event_call
*call
);
527 extern int trace_remove_event_call(struct ftrace_event_call
*call
);
529 #define is_signed_type(type) (((type)(-1)) < (type)1)
531 int trace_set_clr_event(const char *system
, const char *event
, int set
);
534 * The double __builtin_constant_p is because gcc will give us an error
535 * if we try to allocate the static variable to fmt if it is not a
536 * constant. Even with the outer if statement optimizing out.
538 #define event_trace_printk(ip, fmt, args...) \
540 __trace_printk_check_format(fmt, ##args); \
541 tracing_record_cmdline(current); \
542 if (__builtin_constant_p(fmt)) { \
543 static const char *trace_printk_fmt \
544 __attribute__((section("__trace_printk_fmt"))) = \
545 __builtin_constant_p(fmt) ? fmt : NULL; \
547 __trace_bprintk(ip, trace_printk_fmt, ##args); \
549 __trace_printk(ip, fmt, ##args); \
553 * tracepoint_string - register constant persistent string to trace system
554 * @str - a constant persistent string that will be referenced in tracepoints
556 * If constant strings are being used in tracepoints, it is faster and
557 * more efficient to just save the pointer to the string and reference
558 * that with a printf "%s" instead of saving the string in the ring buffer
559 * and wasting space and time.
561 * The problem with the above approach is that userspace tools that read
562 * the binary output of the trace buffers do not have access to the string.
563 * Instead they just show the address of the string which is not very
566 * With tracepoint_string(), the string will be registered to the tracing
567 * system and exported to userspace via the debugfs/tracing/printk_formats
568 * file that maps the string address to the string text. This way userspace
569 * tools that read the binary buffers have a way to map the pointers to
570 * the ASCII strings they represent.
572 * The @str used must be a constant string and persistent as it would not
573 * make sense to show a string that no longer exists. But it is still fine
574 * to be used with modules, because when modules are unloaded, if they
575 * had tracepoints, the ring buffers are cleared too. As long as the string
576 * does not change during the life of the module, it is fine to use
577 * tracepoint_string() within a module.
579 #define tracepoint_string(str) \
581 static const char *___tp_str __tracepoint_string = str; \
584 #define __tracepoint_string __attribute__((section("__tracepoint_str")))
586 #ifdef CONFIG_PERF_EVENTS
589 DECLARE_PER_CPU(struct pt_regs
, perf_trace_regs
);
591 extern int perf_trace_init(struct perf_event
*event
);
592 extern void perf_trace_destroy(struct perf_event
*event
);
593 extern int perf_trace_add(struct perf_event
*event
, int flags
);
594 extern void perf_trace_del(struct perf_event
*event
, int flags
);
595 extern int ftrace_profile_set_filter(struct perf_event
*event
, int event_id
,
597 extern void ftrace_profile_free_filter(struct perf_event
*event
);
598 extern void *perf_trace_buf_prepare(int size
, unsigned short type
,
599 struct pt_regs
*regs
, int *rctxp
);
602 perf_trace_buf_submit(void *raw_data
, int size
, int rctx
, u64 addr
,
603 u64 count
, struct pt_regs
*regs
, void *head
,
604 struct task_struct
*task
)
606 perf_tp_event(addr
, count
, raw_data
, size
, regs
, head
, rctx
, task
);
610 #endif /* _LINUX_FTRACE_EVENT_H */