1
2 #ifndef _LINUX_FTRACE_EVENT_H
3 #define _LINUX_FTRACE_EVENT_H
4
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>
11
12 struct trace_array;
13 struct trace_buffer;
14 struct tracer;
15 struct dentry;
16
17 struct trace_print_flags {
18 unsigned long mask;
19 const char *name;
20 };
21
22 struct trace_print_flags_u64 {
23 unsigned long long mask;
24 const char *name;
25 };
26
27 const char *ftrace_print_flags_seq(struct trace_seq *p, const char *delim,
28 unsigned long flags,
29 const struct trace_print_flags *flag_array);
30
31 const char *ftrace_print_symbols_seq(struct trace_seq *p, unsigned long val,
32 const struct trace_print_flags *symbol_array);
33
34 #if BITS_PER_LONG == 32
35 const char *ftrace_print_symbols_seq_u64(struct trace_seq *p,
36 unsigned long long val,
37 const struct trace_print_flags_u64
38 *symbol_array);
39 #endif
40
41 const char *ftrace_print_bitmask_seq(struct trace_seq *p, void *bitmask_ptr,
42 unsigned int bitmask_size);
43
44 const char *ftrace_print_hex_seq(struct trace_seq *p,
45 const unsigned char *buf, int len);
46
47 struct trace_iterator;
48 struct trace_event;
49
50 int ftrace_raw_output_prep(struct trace_iterator *iter,
51 struct trace_event *event);
52
53 /*
54 * The trace entry - the most basic unit of tracing. This is what
55 * is printed in the end as a single line in the trace output, such as:
56 *
57 * bash-15816 [01] 235.197585: idle_cpu <- irq_enter
58 */
59 struct trace_entry {
60 unsigned short type;
61 unsigned char flags;
62 unsigned char preempt_count;
63 int pid;
64 };
65
66 #define FTRACE_MAX_EVENT \
67 ((1 << (sizeof(((struct trace_entry *)0)->type) * 8)) - 1)
68
69 /*
70 * Trace iterator - used by printout routines who present trace
71 * results to users and which routines might sleep, etc:
72 */
73 struct trace_iterator {
74 struct trace_array *tr;
75 struct tracer *trace;
76 struct trace_buffer *trace_buffer;
77 void *private;
78 int cpu_file;
79 struct mutex mutex;
80 struct ring_buffer_iter **buffer_iter;
81 unsigned long iter_flags;
82
83 /* trace_seq for __print_flags() and __print_symbolic() etc. */
84 struct trace_seq tmp_seq;
85
86 cpumask_var_t started;
87
88 /* it's true when current open file is snapshot */
89 bool snapshot;
90
91 /* The below is zeroed out in pipe_read */
92 struct trace_seq seq;
93 struct trace_entry *ent;
94 unsigned long lost_events;
95 int leftover;
96 int ent_size;
97 int cpu;
98 u64 ts;
99
100 loff_t pos;
101 long idx;
102
103 /* All new field here will be zeroed out in pipe_read */
104 };
105
106 enum trace_iter_flags {
107 TRACE_FILE_LAT_FMT = 1,
108 TRACE_FILE_ANNOTATE = 2,
109 TRACE_FILE_TIME_IN_NS = 4,
110 };
111
112
113 typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
114 int flags, struct trace_event *event);
115
116 struct trace_event_functions {
117 trace_print_func trace;
118 trace_print_func raw;
119 trace_print_func hex;
120 trace_print_func binary;
121 };
122
123 struct trace_event {
124 struct hlist_node node;
125 struct list_head list;
126 int type;
127 struct trace_event_functions *funcs;
128 };
129
130 extern int register_ftrace_event(struct trace_event *event);
131 extern int unregister_ftrace_event(struct trace_event *event);
132
133 /* Return values for print_line callback */
134 enum print_line_t {
135 TRACE_TYPE_PARTIAL_LINE = 0, /* Retry after flushing the seq */
136 TRACE_TYPE_HANDLED = 1,
137 TRACE_TYPE_UNHANDLED = 2, /* Relay to other output functions */
138 TRACE_TYPE_NO_CONSUME = 3 /* Handled but ask to not consume */
139 };
140
141 void tracing_generic_entry_update(struct trace_entry *entry,
142 unsigned long flags,
143 int pc);
144 struct ftrace_event_file;
145
146 struct ring_buffer_event *
147 trace_event_buffer_lock_reserve(struct ring_buffer **current_buffer,
148 struct ftrace_event_file *ftrace_file,
149 int type, unsigned long len,
150 unsigned long flags, int pc);
151 struct ring_buffer_event *
152 trace_current_buffer_lock_reserve(struct ring_buffer **current_buffer,
153 int type, unsigned long len,
154 unsigned long flags, int pc);
155 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
156 struct ring_buffer_event *event,
157 unsigned long flags, int pc);
158 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
159 struct ring_buffer_event *event,
160 unsigned long flags, int pc);
161 void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer,
162 struct ring_buffer_event *event,
163 unsigned long flags, int pc,
164 struct pt_regs *regs);
165 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
166 struct ring_buffer_event *event);
167
168 void tracing_record_cmdline(struct task_struct *tsk);
169
170 int ftrace_output_call(struct trace_iterator *iter, char *name, char *fmt, ...);
171
172 struct event_filter;
173
174 enum trace_reg {
175 TRACE_REG_REGISTER,
176 TRACE_REG_UNREGISTER,
177 #ifdef CONFIG_PERF_EVENTS
178 TRACE_REG_PERF_REGISTER,
179 TRACE_REG_PERF_UNREGISTER,
180 TRACE_REG_PERF_OPEN,
181 TRACE_REG_PERF_CLOSE,
182 TRACE_REG_PERF_ADD,
183 TRACE_REG_PERF_DEL,
184 #endif
185 };
186
187 struct ftrace_event_call;
188
189 struct ftrace_event_class {
190 char *system;
191 void *probe;
192 #ifdef CONFIG_PERF_EVENTS
193 void *perf_probe;
194 #endif
195 int (*reg)(struct ftrace_event_call *event,
196 enum trace_reg type, void *data);
197 int (*define_fields)(struct ftrace_event_call *);
198 struct list_head *(*get_fields)(struct ftrace_event_call *);
199 struct list_head fields;
200 int (*raw_init)(struct ftrace_event_call *);
201 };
202
203 extern int ftrace_event_reg(struct ftrace_event_call *event,
204 enum trace_reg type, void *data);
205
206 int ftrace_output_event(struct trace_iterator *iter, struct ftrace_event_call *event,
207 char *fmt, ...);
208
209 int ftrace_event_define_field(struct ftrace_event_call *call,
210 char *type, int len, char *item, int offset,
211 int field_size, int sign, int filter);
212
213 struct ftrace_event_buffer {
214 struct ring_buffer *buffer;
215 struct ring_buffer_event *event;
216 struct ftrace_event_file *ftrace_file;
217 void *entry;
218 unsigned long flags;
219 int pc;
220 };
221
222 void *ftrace_event_buffer_reserve(struct ftrace_event_buffer *fbuffer,
223 struct ftrace_event_file *ftrace_file,
224 unsigned long len);
225
226 void ftrace_event_buffer_commit(struct ftrace_event_buffer *fbuffer);
227
228 int ftrace_event_define_field(struct ftrace_event_call *call,
229 char *type, int len, char *item, int offset,
230 int field_size, int sign, int filter);
231
232 enum {
233 TRACE_EVENT_FL_FILTERED_BIT,
234 TRACE_EVENT_FL_CAP_ANY_BIT,
235 TRACE_EVENT_FL_NO_SET_FILTER_BIT,
236 TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
237 TRACE_EVENT_FL_WAS_ENABLED_BIT,
238 TRACE_EVENT_FL_USE_CALL_FILTER_BIT,
239 TRACE_EVENT_FL_TRACEPOINT_BIT,
240 };
241
242 /*
243 * Event flags:
244 * FILTERED - The event has a filter attached
245 * CAP_ANY - Any user can enable for perf
246 * NO_SET_FILTER - Set when filter has error and is to be ignored
247 * IGNORE_ENABLE - For ftrace internal events, do not enable with debugfs file
248 * WAS_ENABLED - Set and stays set when an event was ever enabled
249 * (used for module unloading, if a module event is enabled,
250 * it is best to clear the buffers that used it).
251 * USE_CALL_FILTER - For ftrace internal events, don't use file filter
252 * TRACEPOINT - Event is a tracepoint
253 */
254 enum {
255 TRACE_EVENT_FL_FILTERED = (1 << TRACE_EVENT_FL_FILTERED_BIT),
256 TRACE_EVENT_FL_CAP_ANY = (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
257 TRACE_EVENT_FL_NO_SET_FILTER = (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
258 TRACE_EVENT_FL_IGNORE_ENABLE = (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
259 TRACE_EVENT_FL_WAS_ENABLED = (1 << TRACE_EVENT_FL_WAS_ENABLED_BIT),
260 TRACE_EVENT_FL_USE_CALL_FILTER = (1 << TRACE_EVENT_FL_USE_CALL_FILTER_BIT),
261 TRACE_EVENT_FL_TRACEPOINT = (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
262 };
263
264 struct ftrace_event_call {
265 struct list_head list;
266 struct ftrace_event_class *class;
267 union {
268 char *name;
269 /* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
270 struct tracepoint *tp;
271 };
272 struct trace_event event;
273 const char *print_fmt;
274 struct event_filter *filter;
275 void *mod;
276 void *data;
277 /*
278 * bit 0: filter_active
279 * bit 1: allow trace by non root (cap any)
280 * bit 2: failed to apply filter
281 * bit 3: ftrace internal event (do not enable)
282 * bit 4: Event was enabled by module
283 * bit 5: use call filter rather than file filter
284 * bit 6: Event is a tracepoint
285 */
286 int flags; /* static flags of different events */
287
288 #ifdef CONFIG_PERF_EVENTS
289 int perf_refcount;
290 struct hlist_head __percpu *perf_events;
291
292 int (*perf_perm)(struct ftrace_event_call *,
293 struct perf_event *);
294 #endif
295 };
296
297 static inline const char *
ftrace_event_name(struct ftrace_event_call * call)298 ftrace_event_name(struct ftrace_event_call *call)
299 {
300 if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
301 return call->tp ? call->tp->name : NULL;
302 else
303 return call->name;
304 }
305
306 struct trace_array;
307 struct ftrace_subsystem_dir;
308
309 enum {
310 FTRACE_EVENT_FL_ENABLED_BIT,
311 FTRACE_EVENT_FL_RECORDED_CMD_BIT,
312 FTRACE_EVENT_FL_FILTERED_BIT,
313 FTRACE_EVENT_FL_NO_SET_FILTER_BIT,
314 FTRACE_EVENT_FL_SOFT_MODE_BIT,
315 FTRACE_EVENT_FL_SOFT_DISABLED_BIT,
316 FTRACE_EVENT_FL_TRIGGER_MODE_BIT,
317 FTRACE_EVENT_FL_TRIGGER_COND_BIT,
318 };
319
320 /*
321 * Ftrace event file flags:
322 * ENABLED - The event is enabled
323 * RECORDED_CMD - The comms should be recorded at sched_switch
324 * FILTERED - The event has a filter attached
325 * NO_SET_FILTER - Set when filter has error and is to be ignored
326 * SOFT_MODE - The event is enabled/disabled by SOFT_DISABLED
327 * SOFT_DISABLED - When set, do not trace the event (even though its
328 * tracepoint may be enabled)
329 * TRIGGER_MODE - When set, invoke the triggers associated with the event
330 * TRIGGER_COND - When set, one or more triggers has an associated filter
331 */
332 enum {
333 FTRACE_EVENT_FL_ENABLED = (1 << FTRACE_EVENT_FL_ENABLED_BIT),
334 FTRACE_EVENT_FL_RECORDED_CMD = (1 << FTRACE_EVENT_FL_RECORDED_CMD_BIT),
335 FTRACE_EVENT_FL_FILTERED = (1 << FTRACE_EVENT_FL_FILTERED_BIT),
336 FTRACE_EVENT_FL_NO_SET_FILTER = (1 << FTRACE_EVENT_FL_NO_SET_FILTER_BIT),
337 FTRACE_EVENT_FL_SOFT_MODE = (1 << FTRACE_EVENT_FL_SOFT_MODE_BIT),
338 FTRACE_EVENT_FL_SOFT_DISABLED = (1 << FTRACE_EVENT_FL_SOFT_DISABLED_BIT),
339 FTRACE_EVENT_FL_TRIGGER_MODE = (1 << FTRACE_EVENT_FL_TRIGGER_MODE_BIT),
340 FTRACE_EVENT_FL_TRIGGER_COND = (1 << FTRACE_EVENT_FL_TRIGGER_COND_BIT),
341 };
342
343 struct ftrace_event_file {
344 struct list_head list;
345 struct ftrace_event_call *event_call;
346 struct event_filter *filter;
347 struct dentry *dir;
348 struct trace_array *tr;
349 struct ftrace_subsystem_dir *system;
350 struct list_head triggers;
351
352 /*
353 * 32 bit flags:
354 * bit 0: enabled
355 * bit 1: enabled cmd record
356 * bit 2: enable/disable with the soft disable bit
357 * bit 3: soft disabled
358 * bit 4: trigger enabled
359 *
360 * Note: The bits must be set atomically to prevent races
361 * from other writers. Reads of flags do not need to be in
362 * sync as they occur in critical sections. But the way flags
363 * is currently used, these changes do not affect the code
364 * except that when a change is made, it may have a slight
365 * delay in propagating the changes to other CPUs due to
366 * caching and such. Which is mostly OK ;-)
367 */
368 unsigned long flags;
369 atomic_t sm_ref; /* soft-mode reference counter */
370 atomic_t tm_ref; /* trigger-mode reference counter */
371 };
372
373 #define __TRACE_EVENT_FLAGS(name, value) \
374 static int __init trace_init_flags_##name(void) \
375 { \
376 event_##name.flags |= value; \
377 return 0; \
378 } \
379 early_initcall(trace_init_flags_##name);
380
381 #define __TRACE_EVENT_PERF_PERM(name, expr...) \
382 static int perf_perm_##name(struct ftrace_event_call *tp_event, \
383 struct perf_event *p_event) \
384 { \
385 return ({ expr; }); \
386 } \
387 static int __init trace_init_perf_perm_##name(void) \
388 { \
389 event_##name.perf_perm = &perf_perm_##name; \
390 return 0; \
391 } \
392 early_initcall(trace_init_perf_perm_##name);
393
394 #define PERF_MAX_TRACE_SIZE 2048
395
396 #define MAX_FILTER_STR_VAL 256 /* Should handle KSYM_SYMBOL_LEN */
397
398 enum event_trigger_type {
399 ETT_NONE = (0),
400 ETT_TRACE_ONOFF = (1 << 0),
401 ETT_SNAPSHOT = (1 << 1),
402 ETT_STACKTRACE = (1 << 2),
403 ETT_EVENT_ENABLE = (1 << 3),
404 };
405
406 extern int filter_match_preds(struct event_filter *filter, void *rec);
407
408 extern int filter_check_discard(struct ftrace_event_file *file, void *rec,
409 struct ring_buffer *buffer,
410 struct ring_buffer_event *event);
411 extern int call_filter_check_discard(struct ftrace_event_call *call, void *rec,
412 struct ring_buffer *buffer,
413 struct ring_buffer_event *event);
414 extern enum event_trigger_type event_triggers_call(struct ftrace_event_file *file,
415 void *rec);
416 extern void event_triggers_post_call(struct ftrace_event_file *file,
417 enum event_trigger_type tt);
418
419 /**
420 * ftrace_trigger_soft_disabled - do triggers and test if soft disabled
421 * @file: The file pointer of the event to test
422 *
423 * If any triggers without filters are attached to this event, they
424 * will be called here. If the event is soft disabled and has no
425 * triggers that require testing the fields, it will return true,
426 * otherwise false.
427 */
428 static inline bool
ftrace_trigger_soft_disabled(struct ftrace_event_file * file)429 ftrace_trigger_soft_disabled(struct ftrace_event_file *file)
430 {
431 unsigned long eflags = file->flags;
432
433 if (!(eflags & FTRACE_EVENT_FL_TRIGGER_COND)) {
434 if (eflags & FTRACE_EVENT_FL_TRIGGER_MODE)
435 event_triggers_call(file, NULL);
436 if (eflags & FTRACE_EVENT_FL_SOFT_DISABLED)
437 return true;
438 }
439 return false;
440 }
441
442 /*
443 * Helper function for event_trigger_unlock_commit{_regs}().
444 * If there are event triggers attached to this event that requires
445 * filtering against its fields, then they wil be called as the
446 * entry already holds the field information of the current event.
447 *
448 * It also checks if the event should be discarded or not.
449 * It is to be discarded if the event is soft disabled and the
450 * event was only recorded to process triggers, or if the event
451 * filter is active and this event did not match the filters.
452 *
453 * Returns true if the event is discarded, false otherwise.
454 */
455 static inline bool
__event_trigger_test_discard(struct ftrace_event_file * file,struct ring_buffer * buffer,struct ring_buffer_event * event,void * entry,enum event_trigger_type * tt)456 __event_trigger_test_discard(struct ftrace_event_file *file,
457 struct ring_buffer *buffer,
458 struct ring_buffer_event *event,
459 void *entry,
460 enum event_trigger_type *tt)
461 {
462 unsigned long eflags = file->flags;
463
464 if (eflags & FTRACE_EVENT_FL_TRIGGER_COND)
465 *tt = event_triggers_call(file, entry);
466
467 if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags))
468 ring_buffer_discard_commit(buffer, event);
469 else if (!filter_check_discard(file, entry, buffer, event))
470 return false;
471
472 return true;
473 }
474
475 /**
476 * event_trigger_unlock_commit - handle triggers and finish event commit
477 * @file: The file pointer assoctiated to the event
478 * @buffer: The ring buffer that the event is being written to
479 * @event: The event meta data in the ring buffer
480 * @entry: The event itself
481 * @irq_flags: The state of the interrupts at the start of the event
482 * @pc: The state of the preempt count at the start of the event.
483 *
484 * This is a helper function to handle triggers that require data
485 * from the event itself. It also tests the event against filters and
486 * if the event is soft disabled and should be discarded.
487 */
488 static inline void
event_trigger_unlock_commit(struct ftrace_event_file * file,struct ring_buffer * buffer,struct ring_buffer_event * event,void * entry,unsigned long irq_flags,int pc)489 event_trigger_unlock_commit(struct ftrace_event_file *file,
490 struct ring_buffer *buffer,
491 struct ring_buffer_event *event,
492 void *entry, unsigned long irq_flags, int pc)
493 {
494 enum event_trigger_type tt = ETT_NONE;
495
496 if (!__event_trigger_test_discard(file, buffer, event, entry, &tt))
497 trace_buffer_unlock_commit(buffer, event, irq_flags, pc);
498
499 if (tt)
500 event_triggers_post_call(file, tt);
501 }
502
503 /**
504 * event_trigger_unlock_commit_regs - handle triggers and finish event commit
505 * @file: The file pointer assoctiated to the event
506 * @buffer: The ring buffer that the event is being written to
507 * @event: The event meta data in the ring buffer
508 * @entry: The event itself
509 * @irq_flags: The state of the interrupts at the start of the event
510 * @pc: The state of the preempt count at the start of the event.
511 *
512 * This is a helper function to handle triggers that require data
513 * from the event itself. It also tests the event against filters and
514 * if the event is soft disabled and should be discarded.
515 *
516 * Same as event_trigger_unlock_commit() but calls
517 * trace_buffer_unlock_commit_regs() instead of trace_buffer_unlock_commit().
518 */
519 static inline void
event_trigger_unlock_commit_regs(struct ftrace_event_file * file,struct ring_buffer * buffer,struct ring_buffer_event * event,void * entry,unsigned long irq_flags,int pc,struct pt_regs * regs)520 event_trigger_unlock_commit_regs(struct ftrace_event_file *file,
521 struct ring_buffer *buffer,
522 struct ring_buffer_event *event,
523 void *entry, unsigned long irq_flags, int pc,
524 struct pt_regs *regs)
525 {
526 enum event_trigger_type tt = ETT_NONE;
527
528 if (!__event_trigger_test_discard(file, buffer, event, entry, &tt))
529 trace_buffer_unlock_commit_regs(buffer, event,
530 irq_flags, pc, regs);
531
532 if (tt)
533 event_triggers_post_call(file, tt);
534 }
535
536 enum {
537 FILTER_OTHER = 0,
538 FILTER_STATIC_STRING,
539 FILTER_DYN_STRING,
540 FILTER_PTR_STRING,
541 FILTER_TRACE_FN,
542 };
543
544 extern int trace_event_raw_init(struct ftrace_event_call *call);
545 extern int trace_define_field(struct ftrace_event_call *call, const char *type,
546 const char *name, int offset, int size,
547 int is_signed, int filter_type);
548 extern int trace_add_event_call(struct ftrace_event_call *call);
549 extern int trace_remove_event_call(struct ftrace_event_call *call);
550
551 #define is_signed_type(type) (((type)(-1)) < (type)1)
552
553 int trace_set_clr_event(const char *system, const char *event, int set);
554
555 /*
556 * The double __builtin_constant_p is because gcc will give us an error
557 * if we try to allocate the static variable to fmt if it is not a
558 * constant. Even with the outer if statement optimizing out.
559 */
560 #define event_trace_printk(ip, fmt, args...) \
561 do { \
562 __trace_printk_check_format(fmt, ##args); \
563 tracing_record_cmdline(current); \
564 if (__builtin_constant_p(fmt)) { \
565 static const char *trace_printk_fmt \
566 __attribute__((section("__trace_printk_fmt"))) = \
567 __builtin_constant_p(fmt) ? fmt : NULL; \
568 \
569 __trace_bprintk(ip, trace_printk_fmt, ##args); \
570 } else \
571 __trace_printk(ip, fmt, ##args); \
572 } while (0)
573
574 #ifdef CONFIG_PERF_EVENTS
575 struct perf_event;
576
577 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
578
579 extern int perf_trace_init(struct perf_event *event);
580 extern void perf_trace_destroy(struct perf_event *event);
581 extern int perf_trace_add(struct perf_event *event, int flags);
582 extern void perf_trace_del(struct perf_event *event, int flags);
583 extern int ftrace_profile_set_filter(struct perf_event *event, int event_id,
584 char *filter_str);
585 extern void ftrace_profile_free_filter(struct perf_event *event);
586 extern void *perf_trace_buf_prepare(int size, unsigned short type,
587 struct pt_regs **regs, int *rctxp);
588
589 static inline void
perf_trace_buf_submit(void * raw_data,int size,int rctx,u64 addr,u64 count,struct pt_regs * regs,void * head,struct task_struct * task)590 perf_trace_buf_submit(void *raw_data, int size, int rctx, u64 addr,
591 u64 count, struct pt_regs *regs, void *head,
592 struct task_struct *task)
593 {
594 perf_tp_event(addr, count, raw_data, size, regs, head, rctx, task);
595 }
596 #endif
597
598 #endif /* _LINUX_FTRACE_EVENT_H */
599