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);
224 int ftrace_event_define_field(struct ftrace_event_call *call,
225 char *type, int len, char *item, int offset,
226 int field_size, int sign, int filter);
229 TRACE_EVENT_FL_FILTERED_BIT,
230 TRACE_EVENT_FL_CAP_ANY_BIT,
231 TRACE_EVENT_FL_NO_SET_FILTER_BIT,
232 TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
233 TRACE_EVENT_FL_WAS_ENABLED_BIT,
234 TRACE_EVENT_FL_USE_CALL_FILTER_BIT,
239 * FILTERED - The event has a filter attached
240 * CAP_ANY - Any user can enable for perf
241 * NO_SET_FILTER - Set when filter has error and is to be ignored
242 * IGNORE_ENABLE - For ftrace internal events, do not enable with debugfs file
243 * WAS_ENABLED - Set and stays set when an event was ever enabled
244 * (used for module unloading, if a module event is enabled,
245 * it is best to clear the buffers that used it).
246 * USE_CALL_FILTER - For ftrace internal events, don't use file filter
249 TRACE_EVENT_FL_FILTERED = (1 << TRACE_EVENT_FL_FILTERED_BIT),
250 TRACE_EVENT_FL_CAP_ANY = (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
251 TRACE_EVENT_FL_NO_SET_FILTER = (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
252 TRACE_EVENT_FL_IGNORE_ENABLE = (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
253 TRACE_EVENT_FL_WAS_ENABLED = (1 << TRACE_EVENT_FL_WAS_ENABLED_BIT),
254 TRACE_EVENT_FL_USE_CALL_FILTER = (1 << TRACE_EVENT_FL_USE_CALL_FILTER_BIT),
257 struct ftrace_event_call {
258 struct list_head list;
259 struct ftrace_event_class *class;
261 struct trace_event event;
262 const char *print_fmt;
263 struct event_filter *filter;
264 struct list_head *files;
268 * bit 0: filter_active
269 * bit 1: allow trace by non root (cap any)
270 * bit 2: failed to apply filter
271 * bit 3: ftrace internal event (do not enable)
272 * bit 4: Event was enabled by module
273 * bit 5: use call filter rather than file filter
275 int flags; /* static flags of different events */
277 #ifdef CONFIG_PERF_EVENTS
279 struct hlist_head __percpu *perf_events;
281 int (*perf_perm)(struct ftrace_event_call *,
282 struct perf_event *);
287 struct ftrace_subsystem_dir;
290 FTRACE_EVENT_FL_ENABLED_BIT,
291 FTRACE_EVENT_FL_RECORDED_CMD_BIT,
292 FTRACE_EVENT_FL_FILTERED_BIT,
293 FTRACE_EVENT_FL_NO_SET_FILTER_BIT,
294 FTRACE_EVENT_FL_SOFT_MODE_BIT,
295 FTRACE_EVENT_FL_SOFT_DISABLED_BIT,
296 FTRACE_EVENT_FL_TRIGGER_MODE_BIT,
297 FTRACE_EVENT_FL_TRIGGER_COND_BIT,
301 * Ftrace event file flags:
302 * ENABLED - The event is enabled
303 * RECORDED_CMD - The comms should be recorded at sched_switch
304 * FILTERED - The event has a filter attached
305 * NO_SET_FILTER - Set when filter has error and is to be ignored
306 * SOFT_MODE - The event is enabled/disabled by SOFT_DISABLED
307 * SOFT_DISABLED - When set, do not trace the event (even though its
308 * tracepoint may be enabled)
309 * TRIGGER_MODE - When set, invoke the triggers associated with the event
310 * TRIGGER_COND - When set, one or more triggers has an associated filter
313 FTRACE_EVENT_FL_ENABLED = (1 << FTRACE_EVENT_FL_ENABLED_BIT),
314 FTRACE_EVENT_FL_RECORDED_CMD = (1 << FTRACE_EVENT_FL_RECORDED_CMD_BIT),
315 FTRACE_EVENT_FL_FILTERED = (1 << FTRACE_EVENT_FL_FILTERED_BIT),
316 FTRACE_EVENT_FL_NO_SET_FILTER = (1 << FTRACE_EVENT_FL_NO_SET_FILTER_BIT),
317 FTRACE_EVENT_FL_SOFT_MODE = (1 << FTRACE_EVENT_FL_SOFT_MODE_BIT),
318 FTRACE_EVENT_FL_SOFT_DISABLED = (1 << FTRACE_EVENT_FL_SOFT_DISABLED_BIT),
319 FTRACE_EVENT_FL_TRIGGER_MODE = (1 << FTRACE_EVENT_FL_TRIGGER_MODE_BIT),
320 FTRACE_EVENT_FL_TRIGGER_COND = (1 << FTRACE_EVENT_FL_TRIGGER_COND_BIT),
323 struct ftrace_event_file {
324 struct list_head list;
325 struct ftrace_event_call *event_call;
326 struct event_filter *filter;
328 struct trace_array *tr;
329 struct ftrace_subsystem_dir *system;
330 struct list_head triggers;
335 * bit 1: enabled cmd record
336 * bit 2: enable/disable with the soft disable bit
337 * bit 3: soft disabled
338 * bit 4: trigger enabled
340 * Note: The bits must be set atomically to prevent races
341 * from other writers. Reads of flags do not need to be in
342 * sync as they occur in critical sections. But the way flags
343 * is currently used, these changes do not affect the code
344 * except that when a change is made, it may have a slight
345 * delay in propagating the changes to other CPUs due to
346 * caching and such. Which is mostly OK ;-)
349 atomic_t sm_ref; /* soft-mode reference counter */
350 atomic_t tm_ref; /* trigger-mode reference counter */
353 #define __TRACE_EVENT_FLAGS(name, value) \
354 static int __init trace_init_flags_##name(void) \
356 event_##name.flags = value; \
359 early_initcall(trace_init_flags_##name);
361 #define __TRACE_EVENT_PERF_PERM(name, expr...) \
362 static int perf_perm_##name(struct ftrace_event_call *tp_event, \
363 struct perf_event *p_event) \
365 return ({ expr; }); \
367 static int __init trace_init_perf_perm_##name(void) \
369 event_##name.perf_perm = &perf_perm_##name; \
372 early_initcall(trace_init_perf_perm_##name);
374 #define PERF_MAX_TRACE_SIZE 2048
376 #define MAX_FILTER_STR_VAL 256 /* Should handle KSYM_SYMBOL_LEN */
378 enum event_trigger_type {
380 ETT_TRACE_ONOFF = (1 << 0),
381 ETT_SNAPSHOT = (1 << 1),
382 ETT_STACKTRACE = (1 << 2),
383 ETT_EVENT_ENABLE = (1 << 3),
386 extern void destroy_preds(struct ftrace_event_file *file);
387 extern void destroy_call_preds(struct ftrace_event_call *call);
388 extern int filter_match_preds(struct event_filter *filter, void *rec);
390 extern int filter_check_discard(struct ftrace_event_file *file, void *rec,
391 struct ring_buffer *buffer,
392 struct ring_buffer_event *event);
393 extern int call_filter_check_discard(struct ftrace_event_call *call, void *rec,
394 struct ring_buffer *buffer,
395 struct ring_buffer_event *event);
396 extern enum event_trigger_type event_triggers_call(struct ftrace_event_file *file,
398 extern void event_triggers_post_call(struct ftrace_event_file *file,
399 enum event_trigger_type tt);
402 * ftrace_trigger_soft_disabled - do triggers and test if soft disabled
403 * @file: The file pointer of the event to test
405 * If any triggers without filters are attached to this event, they
406 * will be called here. If the event is soft disabled and has no
407 * triggers that require testing the fields, it will return true,
411 ftrace_trigger_soft_disabled(struct ftrace_event_file *file)
413 unsigned long eflags = file->flags;
415 if (!(eflags & FTRACE_EVENT_FL_TRIGGER_COND)) {
416 if (eflags & FTRACE_EVENT_FL_TRIGGER_MODE)
417 event_triggers_call(file, NULL);
418 if (eflags & FTRACE_EVENT_FL_SOFT_DISABLED)
425 * Helper function for event_trigger_unlock_commit{_regs}().
426 * If there are event triggers attached to this event that requires
427 * filtering against its fields, then they wil be called as the
428 * entry already holds the field information of the current event.
430 * It also checks if the event should be discarded or not.
431 * It is to be discarded if the event is soft disabled and the
432 * event was only recorded to process triggers, or if the event
433 * filter is active and this event did not match the filters.
435 * Returns true if the event is discarded, false otherwise.
438 __event_trigger_test_discard(struct ftrace_event_file *file,
439 struct ring_buffer *buffer,
440 struct ring_buffer_event *event,
442 enum event_trigger_type *tt)
444 unsigned long eflags = file->flags;
446 if (eflags & FTRACE_EVENT_FL_TRIGGER_COND)
447 *tt = event_triggers_call(file, entry);
449 if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags))
450 ring_buffer_discard_commit(buffer, event);
451 else if (!filter_check_discard(file, entry, buffer, event))
458 * event_trigger_unlock_commit - 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.
471 event_trigger_unlock_commit(struct ftrace_event_file *file,
472 struct ring_buffer *buffer,
473 struct ring_buffer_event *event,
474 void *entry, unsigned long irq_flags, int pc)
476 enum event_trigger_type tt = ETT_NONE;
478 if (!__event_trigger_test_discard(file, buffer, event, entry, &tt))
479 trace_buffer_unlock_commit(buffer, event, irq_flags, pc);
482 event_triggers_post_call(file, tt);
486 * event_trigger_unlock_commit_regs - handle triggers and finish event commit
487 * @file: The file pointer assoctiated to the event
488 * @buffer: The ring buffer that the event is being written to
489 * @event: The event meta data in the ring buffer
490 * @entry: The event itself
491 * @irq_flags: The state of the interrupts at the start of the event
492 * @pc: The state of the preempt count at the start of the event.
494 * This is a helper function to handle triggers that require data
495 * from the event itself. It also tests the event against filters and
496 * if the event is soft disabled and should be discarded.
498 * Same as event_trigger_unlock_commit() but calls
499 * trace_buffer_unlock_commit_regs() instead of trace_buffer_unlock_commit().
502 event_trigger_unlock_commit_regs(struct ftrace_event_file *file,
503 struct ring_buffer *buffer,
504 struct ring_buffer_event *event,
505 void *entry, unsigned long irq_flags, int pc,
506 struct pt_regs *regs)
508 enum event_trigger_type tt = ETT_NONE;
510 if (!__event_trigger_test_discard(file, buffer, event, entry, &tt))
511 trace_buffer_unlock_commit_regs(buffer, event,
512 irq_flags, pc, regs);
515 event_triggers_post_call(file, tt);
520 FILTER_STATIC_STRING,
526 extern int trace_event_raw_init(struct ftrace_event_call *call);
527 extern int trace_define_field(struct ftrace_event_call *call, const char *type,
528 const char *name, int offset, int size,
529 int is_signed, int filter_type);
530 extern int trace_add_event_call(struct ftrace_event_call *call);
531 extern int trace_remove_event_call(struct ftrace_event_call *call);
533 #define is_signed_type(type) (((type)(-1)) < (type)1)
535 int trace_set_clr_event(const char *system, const char *event, int set);
538 * The double __builtin_constant_p is because gcc will give us an error
539 * if we try to allocate the static variable to fmt if it is not a
540 * constant. Even with the outer if statement optimizing out.
542 #define event_trace_printk(ip, fmt, args...) \
544 __trace_printk_check_format(fmt, ##args); \
545 tracing_record_cmdline(current); \
546 if (__builtin_constant_p(fmt)) { \
547 static const char *trace_printk_fmt \
548 __attribute__((section("__trace_printk_fmt"))) = \
549 __builtin_constant_p(fmt) ? fmt : NULL; \
551 __trace_bprintk(ip, trace_printk_fmt, ##args); \
553 __trace_printk(ip, fmt, ##args); \
557 * tracepoint_string - register constant persistent string to trace system
558 * @str - a constant persistent string that will be referenced in tracepoints
560 * If constant strings are being used in tracepoints, it is faster and
561 * more efficient to just save the pointer to the string and reference
562 * that with a printf "%s" instead of saving the string in the ring buffer
563 * and wasting space and time.
565 * The problem with the above approach is that userspace tools that read
566 * the binary output of the trace buffers do not have access to the string.
567 * Instead they just show the address of the string which is not very
570 * With tracepoint_string(), the string will be registered to the tracing
571 * system and exported to userspace via the debugfs/tracing/printk_formats
572 * file that maps the string address to the string text. This way userspace
573 * tools that read the binary buffers have a way to map the pointers to
574 * the ASCII strings they represent.
576 * The @str used must be a constant string and persistent as it would not
577 * make sense to show a string that no longer exists. But it is still fine
578 * to be used with modules, because when modules are unloaded, if they
579 * had tracepoints, the ring buffers are cleared too. As long as the string
580 * does not change during the life of the module, it is fine to use
581 * tracepoint_string() within a module.
583 #define tracepoint_string(str) \
585 static const char *___tp_str __tracepoint_string = str; \
588 #define __tracepoint_string __attribute__((section("__tracepoint_str")))
590 #ifdef CONFIG_PERF_EVENTS
593 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
595 extern int perf_trace_init(struct perf_event *event);
596 extern void perf_trace_destroy(struct perf_event *event);
597 extern int perf_trace_add(struct perf_event *event, int flags);
598 extern void perf_trace_del(struct perf_event *event, int flags);
599 extern int ftrace_profile_set_filter(struct perf_event *event, int event_id,
601 extern void ftrace_profile_free_filter(struct perf_event *event);
602 extern void *perf_trace_buf_prepare(int size, unsigned short type,
603 struct pt_regs *regs, int *rctxp);
606 perf_trace_buf_submit(void *raw_data, int size, int rctx, u64 addr,
607 u64 count, struct pt_regs *regs, void *head,
608 struct task_struct *task)
610 perf_tp_event(addr, count, raw_data, size, regs, head, rctx, task);
614 #endif /* _LINUX_FTRACE_EVENT_H */