Merge tag 'trace-fixes-v4.5-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git...
[deliverable/linux.git] / include / linux / trace_events.h
1
2 #ifndef _LINUX_TRACE_EVENT_H
3 #define _LINUX_TRACE_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 struct bpf_prog;
17
18 struct trace_print_flags {
19 unsigned long mask;
20 const char *name;
21 };
22
23 struct trace_print_flags_u64 {
24 unsigned long long mask;
25 const char *name;
26 };
27
28 const char *trace_print_flags_seq(struct trace_seq *p, const char *delim,
29 unsigned long flags,
30 const struct trace_print_flags *flag_array);
31
32 const char *trace_print_symbols_seq(struct trace_seq *p, unsigned long val,
33 const struct trace_print_flags *symbol_array);
34
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
39 *symbol_array);
40 #endif
41
42 const char *trace_print_bitmask_seq(struct trace_seq *p, void *bitmask_ptr,
43 unsigned int bitmask_size);
44
45 const char *trace_print_hex_seq(struct trace_seq *p,
46 const unsigned char *buf, int len);
47
48 const char *trace_print_array_seq(struct trace_seq *p,
49 const void *buf, int count,
50 size_t el_size);
51
52 struct trace_iterator;
53 struct trace_event;
54
55 int trace_raw_output_prep(struct trace_iterator *iter,
56 struct trace_event *event);
57
58 /*
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:
61 *
62 * bash-15816 [01] 235.197585: idle_cpu <- irq_enter
63 */
64 struct trace_entry {
65 unsigned short type;
66 unsigned char flags;
67 unsigned char preempt_count;
68 int pid;
69 };
70
71 #define TRACE_EVENT_TYPE_MAX \
72 ((1 << (sizeof(((struct trace_entry *)0)->type) * 8)) - 1)
73
74 /*
75 * Trace iterator - used by printout routines who present trace
76 * results to users and which routines might sleep, etc:
77 */
78 struct trace_iterator {
79 struct trace_array *tr;
80 struct tracer *trace;
81 struct trace_buffer *trace_buffer;
82 void *private;
83 int cpu_file;
84 struct mutex mutex;
85 struct ring_buffer_iter **buffer_iter;
86 unsigned long iter_flags;
87
88 /* trace_seq for __print_flags() and __print_symbolic() etc. */
89 struct trace_seq tmp_seq;
90
91 cpumask_var_t started;
92
93 /* it's true when current open file is snapshot */
94 bool snapshot;
95
96 /* The below is zeroed out in pipe_read */
97 struct trace_seq seq;
98 struct trace_entry *ent;
99 unsigned long lost_events;
100 int leftover;
101 int ent_size;
102 int cpu;
103 u64 ts;
104
105 loff_t pos;
106 long idx;
107
108 /* All new field here will be zeroed out in pipe_read */
109 };
110
111 enum trace_iter_flags {
112 TRACE_FILE_LAT_FMT = 1,
113 TRACE_FILE_ANNOTATE = 2,
114 TRACE_FILE_TIME_IN_NS = 4,
115 };
116
117
118 typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
119 int flags, struct trace_event *event);
120
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;
126 };
127
128 struct trace_event {
129 struct hlist_node node;
130 struct list_head list;
131 int type;
132 struct trace_event_functions *funcs;
133 };
134
135 extern int register_trace_event(struct trace_event *event);
136 extern int unregister_trace_event(struct trace_event *event);
137
138 /* Return values for print_line callback */
139 enum print_line_t {
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 */
144 };
145
146 /*
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.
150 */
151 static inline enum print_line_t trace_handle_return(struct trace_seq *s)
152 {
153 return trace_seq_has_overflowed(s) ?
154 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
155 }
156
157 void tracing_generic_entry_update(struct trace_entry *entry,
158 unsigned long flags,
159 int pc);
160 struct trace_event_file;
161
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_buffer_unlock_commit(struct trace_array *tr,
172 struct ring_buffer *buffer,
173 struct ring_buffer_event *event,
174 unsigned long flags, int pc);
175 void trace_buffer_unlock_commit_regs(struct trace_array *tr,
176 struct ring_buffer *buffer,
177 struct ring_buffer_event *event,
178 unsigned long flags, int pc,
179 struct pt_regs *regs);
180 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
181 struct ring_buffer_event *event);
182
183 void tracing_record_cmdline(struct task_struct *tsk);
184
185 int trace_output_call(struct trace_iterator *iter, char *name, char *fmt, ...);
186
187 struct event_filter;
188
189 enum trace_reg {
190 TRACE_REG_REGISTER,
191 TRACE_REG_UNREGISTER,
192 #ifdef CONFIG_PERF_EVENTS
193 TRACE_REG_PERF_REGISTER,
194 TRACE_REG_PERF_UNREGISTER,
195 TRACE_REG_PERF_OPEN,
196 TRACE_REG_PERF_CLOSE,
197 TRACE_REG_PERF_ADD,
198 TRACE_REG_PERF_DEL,
199 #endif
200 };
201
202 struct trace_event_call;
203
204 struct trace_event_class {
205 const char *system;
206 void *probe;
207 #ifdef CONFIG_PERF_EVENTS
208 void *perf_probe;
209 #endif
210 int (*reg)(struct trace_event_call *event,
211 enum trace_reg type, void *data);
212 int (*define_fields)(struct trace_event_call *);
213 struct list_head *(*get_fields)(struct trace_event_call *);
214 struct list_head fields;
215 int (*raw_init)(struct trace_event_call *);
216 };
217
218 extern int trace_event_reg(struct trace_event_call *event,
219 enum trace_reg type, void *data);
220
221 struct trace_event_buffer {
222 struct ring_buffer *buffer;
223 struct ring_buffer_event *event;
224 struct trace_event_file *trace_file;
225 void *entry;
226 unsigned long flags;
227 int pc;
228 };
229
230 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
231 struct trace_event_file *trace_file,
232 unsigned long len);
233
234 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer);
235
236 enum {
237 TRACE_EVENT_FL_FILTERED_BIT,
238 TRACE_EVENT_FL_CAP_ANY_BIT,
239 TRACE_EVENT_FL_NO_SET_FILTER_BIT,
240 TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
241 TRACE_EVENT_FL_WAS_ENABLED_BIT,
242 TRACE_EVENT_FL_USE_CALL_FILTER_BIT,
243 TRACE_EVENT_FL_TRACEPOINT_BIT,
244 TRACE_EVENT_FL_KPROBE_BIT,
245 TRACE_EVENT_FL_UPROBE_BIT,
246 };
247
248 /*
249 * Event flags:
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
260 * UPROBE - Event is a uprobe
261 */
262 enum {
263 TRACE_EVENT_FL_FILTERED = (1 << TRACE_EVENT_FL_FILTERED_BIT),
264 TRACE_EVENT_FL_CAP_ANY = (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
265 TRACE_EVENT_FL_NO_SET_FILTER = (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
266 TRACE_EVENT_FL_IGNORE_ENABLE = (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
267 TRACE_EVENT_FL_WAS_ENABLED = (1 << TRACE_EVENT_FL_WAS_ENABLED_BIT),
268 TRACE_EVENT_FL_USE_CALL_FILTER = (1 << TRACE_EVENT_FL_USE_CALL_FILTER_BIT),
269 TRACE_EVENT_FL_TRACEPOINT = (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
270 TRACE_EVENT_FL_KPROBE = (1 << TRACE_EVENT_FL_KPROBE_BIT),
271 TRACE_EVENT_FL_UPROBE = (1 << TRACE_EVENT_FL_UPROBE_BIT),
272 };
273
274 #define TRACE_EVENT_FL_UKPROBE (TRACE_EVENT_FL_KPROBE | TRACE_EVENT_FL_UPROBE)
275
276 struct trace_event_call {
277 struct list_head list;
278 struct trace_event_class *class;
279 union {
280 char *name;
281 /* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
282 struct tracepoint *tp;
283 };
284 struct trace_event event;
285 char *print_fmt;
286 struct event_filter *filter;
287 void *mod;
288 void *data;
289 /*
290 * bit 0: filter_active
291 * bit 1: allow trace by non root (cap any)
292 * bit 2: failed to apply filter
293 * bit 3: trace internal event (do not enable)
294 * bit 4: Event was enabled by module
295 * bit 5: use call filter rather than file filter
296 * bit 6: Event is a tracepoint
297 */
298 int flags; /* static flags of different events */
299
300 #ifdef CONFIG_PERF_EVENTS
301 int perf_refcount;
302 struct hlist_head __percpu *perf_events;
303 struct bpf_prog *prog;
304
305 int (*perf_perm)(struct trace_event_call *,
306 struct perf_event *);
307 #endif
308 };
309
310 static inline const char *
311 trace_event_name(struct trace_event_call *call)
312 {
313 if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
314 return call->tp ? call->tp->name : NULL;
315 else
316 return call->name;
317 }
318
319 struct trace_array;
320 struct trace_subsystem_dir;
321
322 enum {
323 EVENT_FILE_FL_ENABLED_BIT,
324 EVENT_FILE_FL_RECORDED_CMD_BIT,
325 EVENT_FILE_FL_FILTERED_BIT,
326 EVENT_FILE_FL_NO_SET_FILTER_BIT,
327 EVENT_FILE_FL_SOFT_MODE_BIT,
328 EVENT_FILE_FL_SOFT_DISABLED_BIT,
329 EVENT_FILE_FL_TRIGGER_MODE_BIT,
330 EVENT_FILE_FL_TRIGGER_COND_BIT,
331 EVENT_FILE_FL_PID_FILTER_BIT,
332 };
333
334 /*
335 * Event file flags:
336 * ENABLED - The event is enabled
337 * RECORDED_CMD - The comms should be recorded at sched_switch
338 * FILTERED - The event has a filter attached
339 * NO_SET_FILTER - Set when filter has error and is to be ignored
340 * SOFT_MODE - The event is enabled/disabled by SOFT_DISABLED
341 * SOFT_DISABLED - When set, do not trace the event (even though its
342 * tracepoint may be enabled)
343 * TRIGGER_MODE - When set, invoke the triggers associated with the event
344 * TRIGGER_COND - When set, one or more triggers has an associated filter
345 * PID_FILTER - When set, the event is filtered based on pid
346 */
347 enum {
348 EVENT_FILE_FL_ENABLED = (1 << EVENT_FILE_FL_ENABLED_BIT),
349 EVENT_FILE_FL_RECORDED_CMD = (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
350 EVENT_FILE_FL_FILTERED = (1 << EVENT_FILE_FL_FILTERED_BIT),
351 EVENT_FILE_FL_NO_SET_FILTER = (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
352 EVENT_FILE_FL_SOFT_MODE = (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
353 EVENT_FILE_FL_SOFT_DISABLED = (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
354 EVENT_FILE_FL_TRIGGER_MODE = (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
355 EVENT_FILE_FL_TRIGGER_COND = (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
356 EVENT_FILE_FL_PID_FILTER = (1 << EVENT_FILE_FL_PID_FILTER_BIT),
357 };
358
359 struct trace_event_file {
360 struct list_head list;
361 struct trace_event_call *event_call;
362 struct event_filter *filter;
363 struct dentry *dir;
364 struct trace_array *tr;
365 struct trace_subsystem_dir *system;
366 struct list_head triggers;
367
368 /*
369 * 32 bit flags:
370 * bit 0: enabled
371 * bit 1: enabled cmd record
372 * bit 2: enable/disable with the soft disable bit
373 * bit 3: soft disabled
374 * bit 4: trigger enabled
375 *
376 * Note: The bits must be set atomically to prevent races
377 * from other writers. Reads of flags do not need to be in
378 * sync as they occur in critical sections. But the way flags
379 * is currently used, these changes do not affect the code
380 * except that when a change is made, it may have a slight
381 * delay in propagating the changes to other CPUs due to
382 * caching and such. Which is mostly OK ;-)
383 */
384 unsigned long flags;
385 atomic_t sm_ref; /* soft-mode reference counter */
386 atomic_t tm_ref; /* trigger-mode reference counter */
387 };
388
389 #define __TRACE_EVENT_FLAGS(name, value) \
390 static int __init trace_init_flags_##name(void) \
391 { \
392 event_##name.flags |= value; \
393 return 0; \
394 } \
395 early_initcall(trace_init_flags_##name);
396
397 #define __TRACE_EVENT_PERF_PERM(name, expr...) \
398 static int perf_perm_##name(struct trace_event_call *tp_event, \
399 struct perf_event *p_event) \
400 { \
401 return ({ expr; }); \
402 } \
403 static int __init trace_init_perf_perm_##name(void) \
404 { \
405 event_##name.perf_perm = &perf_perm_##name; \
406 return 0; \
407 } \
408 early_initcall(trace_init_perf_perm_##name);
409
410 #define PERF_MAX_TRACE_SIZE 2048
411
412 #define MAX_FILTER_STR_VAL 256 /* Should handle KSYM_SYMBOL_LEN */
413
414 enum event_trigger_type {
415 ETT_NONE = (0),
416 ETT_TRACE_ONOFF = (1 << 0),
417 ETT_SNAPSHOT = (1 << 1),
418 ETT_STACKTRACE = (1 << 2),
419 ETT_EVENT_ENABLE = (1 << 3),
420 };
421
422 extern int filter_match_preds(struct event_filter *filter, void *rec);
423
424 extern int filter_check_discard(struct trace_event_file *file, void *rec,
425 struct ring_buffer *buffer,
426 struct ring_buffer_event *event);
427 extern int call_filter_check_discard(struct trace_event_call *call, void *rec,
428 struct ring_buffer *buffer,
429 struct ring_buffer_event *event);
430 extern enum event_trigger_type event_triggers_call(struct trace_event_file *file,
431 void *rec);
432 extern void event_triggers_post_call(struct trace_event_file *file,
433 enum event_trigger_type tt);
434
435 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file);
436
437 /**
438 * trace_trigger_soft_disabled - do triggers and test if soft disabled
439 * @file: The file pointer of the event to test
440 *
441 * If any triggers without filters are attached to this event, they
442 * will be called here. If the event is soft disabled and has no
443 * triggers that require testing the fields, it will return true,
444 * otherwise false.
445 */
446 static inline bool
447 trace_trigger_soft_disabled(struct trace_event_file *file)
448 {
449 unsigned long eflags = file->flags;
450
451 if (!(eflags & EVENT_FILE_FL_TRIGGER_COND)) {
452 if (eflags & EVENT_FILE_FL_TRIGGER_MODE)
453 event_triggers_call(file, NULL);
454 if (eflags & EVENT_FILE_FL_SOFT_DISABLED)
455 return true;
456 if (eflags & EVENT_FILE_FL_PID_FILTER)
457 return trace_event_ignore_this_pid(file);
458 }
459 return false;
460 }
461
462 /*
463 * Helper function for event_trigger_unlock_commit{_regs}().
464 * If there are event triggers attached to this event that requires
465 * filtering against its fields, then they wil be called as the
466 * entry already holds the field information of the current event.
467 *
468 * It also checks if the event should be discarded or not.
469 * It is to be discarded if the event is soft disabled and the
470 * event was only recorded to process triggers, or if the event
471 * filter is active and this event did not match the filters.
472 *
473 * Returns true if the event is discarded, false otherwise.
474 */
475 static inline bool
476 __event_trigger_test_discard(struct trace_event_file *file,
477 struct ring_buffer *buffer,
478 struct ring_buffer_event *event,
479 void *entry,
480 enum event_trigger_type *tt)
481 {
482 unsigned long eflags = file->flags;
483
484 if (eflags & EVENT_FILE_FL_TRIGGER_COND)
485 *tt = event_triggers_call(file, entry);
486
487 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags))
488 ring_buffer_discard_commit(buffer, event);
489 else if (!filter_check_discard(file, entry, buffer, event))
490 return false;
491
492 return true;
493 }
494
495 /**
496 * event_trigger_unlock_commit - handle triggers and finish event commit
497 * @file: The file pointer assoctiated to the event
498 * @buffer: The ring buffer that the event is being written to
499 * @event: The event meta data in the ring buffer
500 * @entry: The event itself
501 * @irq_flags: The state of the interrupts at the start of the event
502 * @pc: The state of the preempt count at the start of the event.
503 *
504 * This is a helper function to handle triggers that require data
505 * from the event itself. It also tests the event against filters and
506 * if the event is soft disabled and should be discarded.
507 */
508 static inline void
509 event_trigger_unlock_commit(struct trace_event_file *file,
510 struct ring_buffer *buffer,
511 struct ring_buffer_event *event,
512 void *entry, unsigned long irq_flags, int pc)
513 {
514 enum event_trigger_type tt = ETT_NONE;
515
516 if (!__event_trigger_test_discard(file, buffer, event, entry, &tt))
517 trace_buffer_unlock_commit(file->tr, buffer, event, irq_flags, pc);
518
519 if (tt)
520 event_triggers_post_call(file, tt);
521 }
522
523 /**
524 * event_trigger_unlock_commit_regs - handle triggers and finish event commit
525 * @file: The file pointer assoctiated to the event
526 * @buffer: The ring buffer that the event is being written to
527 * @event: The event meta data in the ring buffer
528 * @entry: The event itself
529 * @irq_flags: The state of the interrupts at the start of the event
530 * @pc: The state of the preempt count at the start of the event.
531 *
532 * This is a helper function to handle triggers that require data
533 * from the event itself. It also tests the event against filters and
534 * if the event is soft disabled and should be discarded.
535 *
536 * Same as event_trigger_unlock_commit() but calls
537 * trace_buffer_unlock_commit_regs() instead of trace_buffer_unlock_commit().
538 */
539 static inline void
540 event_trigger_unlock_commit_regs(struct trace_event_file *file,
541 struct ring_buffer *buffer,
542 struct ring_buffer_event *event,
543 void *entry, unsigned long irq_flags, int pc,
544 struct pt_regs *regs)
545 {
546 enum event_trigger_type tt = ETT_NONE;
547
548 if (!__event_trigger_test_discard(file, buffer, event, entry, &tt))
549 trace_buffer_unlock_commit_regs(file->tr, buffer, event,
550 irq_flags, pc, regs);
551
552 if (tt)
553 event_triggers_post_call(file, tt);
554 }
555
556 #ifdef CONFIG_BPF_EVENTS
557 unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx);
558 #else
559 static inline unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx)
560 {
561 return 1;
562 }
563 #endif
564
565 enum {
566 FILTER_OTHER = 0,
567 FILTER_STATIC_STRING,
568 FILTER_DYN_STRING,
569 FILTER_PTR_STRING,
570 FILTER_TRACE_FN,
571 };
572
573 extern int trace_event_raw_init(struct trace_event_call *call);
574 extern int trace_define_field(struct trace_event_call *call, const char *type,
575 const char *name, int offset, int size,
576 int is_signed, int filter_type);
577 extern int trace_add_event_call(struct trace_event_call *call);
578 extern int trace_remove_event_call(struct trace_event_call *call);
579
580 #define is_signed_type(type) (((type)(-1)) < (type)1)
581
582 int trace_set_clr_event(const char *system, const char *event, int set);
583
584 /*
585 * The double __builtin_constant_p is because gcc will give us an error
586 * if we try to allocate the static variable to fmt if it is not a
587 * constant. Even with the outer if statement optimizing out.
588 */
589 #define event_trace_printk(ip, fmt, args...) \
590 do { \
591 __trace_printk_check_format(fmt, ##args); \
592 tracing_record_cmdline(current); \
593 if (__builtin_constant_p(fmt)) { \
594 static const char *trace_printk_fmt \
595 __attribute__((section("__trace_printk_fmt"))) = \
596 __builtin_constant_p(fmt) ? fmt : NULL; \
597 \
598 __trace_bprintk(ip, trace_printk_fmt, ##args); \
599 } else \
600 __trace_printk(ip, fmt, ##args); \
601 } while (0)
602
603 #ifdef CONFIG_PERF_EVENTS
604 struct perf_event;
605
606 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
607
608 extern int perf_trace_init(struct perf_event *event);
609 extern void perf_trace_destroy(struct perf_event *event);
610 extern int perf_trace_add(struct perf_event *event, int flags);
611 extern void perf_trace_del(struct perf_event *event, int flags);
612 extern int ftrace_profile_set_filter(struct perf_event *event, int event_id,
613 char *filter_str);
614 extern void ftrace_profile_free_filter(struct perf_event *event);
615 extern void *perf_trace_buf_prepare(int size, unsigned short type,
616 struct pt_regs **regs, int *rctxp);
617
618 static inline void
619 perf_trace_buf_submit(void *raw_data, int size, int rctx, u64 addr,
620 u64 count, struct pt_regs *regs, void *head,
621 struct task_struct *task)
622 {
623 perf_tp_event(addr, count, raw_data, size, regs, head, rctx, task);
624 }
625 #endif
626
627 #endif /* _LINUX_TRACE_EVENT_H */
This page took 0.044267 seconds and 5 git commands to generate.