]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - include/linux/ftrace_event.h
watchdog: introduce separate handlers for parameters in /proc/sys/kernel
[mirror_ubuntu-artful-kernel.git] / include / linux / ftrace_event.h
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 const char *ftrace_print_array_seq(struct trace_seq *p,
48 const void *buf, int buf_len,
49 size_t el_size);
50
51 struct trace_iterator;
52 struct trace_event;
53
54 int ftrace_raw_output_prep(struct trace_iterator *iter,
55 struct trace_event *event);
56
57 /*
58 * The trace entry - the most basic unit of tracing. This is what
59 * is printed in the end as a single line in the trace output, such as:
60 *
61 * bash-15816 [01] 235.197585: idle_cpu <- irq_enter
62 */
63 struct trace_entry {
64 unsigned short type;
65 unsigned char flags;
66 unsigned char preempt_count;
67 int pid;
68 };
69
70 #define FTRACE_MAX_EVENT \
71 ((1 << (sizeof(((struct trace_entry *)0)->type) * 8)) - 1)
72
73 /*
74 * Trace iterator - used by printout routines who present trace
75 * results to users and which routines might sleep, etc:
76 */
77 struct trace_iterator {
78 struct trace_array *tr;
79 struct tracer *trace;
80 struct trace_buffer *trace_buffer;
81 void *private;
82 int cpu_file;
83 struct mutex mutex;
84 struct ring_buffer_iter **buffer_iter;
85 unsigned long iter_flags;
86
87 /* trace_seq for __print_flags() and __print_symbolic() etc. */
88 struct trace_seq tmp_seq;
89
90 cpumask_var_t started;
91
92 /* it's true when current open file is snapshot */
93 bool snapshot;
94
95 /* The below is zeroed out in pipe_read */
96 struct trace_seq seq;
97 struct trace_entry *ent;
98 unsigned long lost_events;
99 int leftover;
100 int ent_size;
101 int cpu;
102 u64 ts;
103
104 loff_t pos;
105 long idx;
106
107 /* All new field here will be zeroed out in pipe_read */
108 };
109
110 enum trace_iter_flags {
111 TRACE_FILE_LAT_FMT = 1,
112 TRACE_FILE_ANNOTATE = 2,
113 TRACE_FILE_TIME_IN_NS = 4,
114 };
115
116
117 typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
118 int flags, struct trace_event *event);
119
120 struct trace_event_functions {
121 trace_print_func trace;
122 trace_print_func raw;
123 trace_print_func hex;
124 trace_print_func binary;
125 };
126
127 struct trace_event {
128 struct hlist_node node;
129 struct list_head list;
130 int type;
131 struct trace_event_functions *funcs;
132 };
133
134 extern int register_ftrace_event(struct trace_event *event);
135 extern int unregister_ftrace_event(struct trace_event *event);
136
137 /* Return values for print_line callback */
138 enum print_line_t {
139 TRACE_TYPE_PARTIAL_LINE = 0, /* Retry after flushing the seq */
140 TRACE_TYPE_HANDLED = 1,
141 TRACE_TYPE_UNHANDLED = 2, /* Relay to other output functions */
142 TRACE_TYPE_NO_CONSUME = 3 /* Handled but ask to not consume */
143 };
144
145 /*
146 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
147 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
148 * simplifies those functions and keeps them in sync.
149 */
150 static inline enum print_line_t trace_handle_return(struct trace_seq *s)
151 {
152 return trace_seq_has_overflowed(s) ?
153 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
154 }
155
156 void tracing_generic_entry_update(struct trace_entry *entry,
157 unsigned long flags,
158 int pc);
159 struct ftrace_event_file;
160
161 struct ring_buffer_event *
162 trace_event_buffer_lock_reserve(struct ring_buffer **current_buffer,
163 struct ftrace_event_file *ftrace_file,
164 int type, unsigned long len,
165 unsigned long flags, int pc);
166 struct ring_buffer_event *
167 trace_current_buffer_lock_reserve(struct ring_buffer **current_buffer,
168 int type, unsigned long len,
169 unsigned long flags, int pc);
170 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
171 struct ring_buffer_event *event,
172 unsigned long flags, int pc);
173 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
174 struct ring_buffer_event *event,
175 unsigned long flags, int pc);
176 void trace_buffer_unlock_commit_regs(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 ftrace_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 ftrace_event_call;
203
204 struct ftrace_event_class {
205 char *system;
206 void *probe;
207 #ifdef CONFIG_PERF_EVENTS
208 void *perf_probe;
209 #endif
210 int (*reg)(struct ftrace_event_call *event,
211 enum trace_reg type, void *data);
212 int (*define_fields)(struct ftrace_event_call *);
213 struct list_head *(*get_fields)(struct ftrace_event_call *);
214 struct list_head fields;
215 int (*raw_init)(struct ftrace_event_call *);
216 };
217
218 extern int ftrace_event_reg(struct ftrace_event_call *event,
219 enum trace_reg type, void *data);
220
221 int ftrace_output_event(struct trace_iterator *iter, struct ftrace_event_call *event,
222 char *fmt, ...);
223
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);
227
228 struct ftrace_event_buffer {
229 struct ring_buffer *buffer;
230 struct ring_buffer_event *event;
231 struct ftrace_event_file *ftrace_file;
232 void *entry;
233 unsigned long flags;
234 int pc;
235 };
236
237 void *ftrace_event_buffer_reserve(struct ftrace_event_buffer *fbuffer,
238 struct ftrace_event_file *ftrace_file,
239 unsigned long len);
240
241 void ftrace_event_buffer_commit(struct ftrace_event_buffer *fbuffer);
242
243 int ftrace_event_define_field(struct ftrace_event_call *call,
244 char *type, int len, char *item, int offset,
245 int field_size, int sign, int filter);
246
247 enum {
248 TRACE_EVENT_FL_FILTERED_BIT,
249 TRACE_EVENT_FL_CAP_ANY_BIT,
250 TRACE_EVENT_FL_NO_SET_FILTER_BIT,
251 TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
252 TRACE_EVENT_FL_WAS_ENABLED_BIT,
253 TRACE_EVENT_FL_USE_CALL_FILTER_BIT,
254 TRACE_EVENT_FL_TRACEPOINT_BIT,
255 };
256
257 /*
258 * Event flags:
259 * FILTERED - The event has a filter attached
260 * CAP_ANY - Any user can enable for perf
261 * NO_SET_FILTER - Set when filter has error and is to be ignored
262 * IGNORE_ENABLE - For ftrace internal events, do not enable with debugfs file
263 * WAS_ENABLED - Set and stays set when an event was ever enabled
264 * (used for module unloading, if a module event is enabled,
265 * it is best to clear the buffers that used it).
266 * USE_CALL_FILTER - For ftrace internal events, don't use file filter
267 * TRACEPOINT - Event is a tracepoint
268 */
269 enum {
270 TRACE_EVENT_FL_FILTERED = (1 << TRACE_EVENT_FL_FILTERED_BIT),
271 TRACE_EVENT_FL_CAP_ANY = (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
272 TRACE_EVENT_FL_NO_SET_FILTER = (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
273 TRACE_EVENT_FL_IGNORE_ENABLE = (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
274 TRACE_EVENT_FL_WAS_ENABLED = (1 << TRACE_EVENT_FL_WAS_ENABLED_BIT),
275 TRACE_EVENT_FL_USE_CALL_FILTER = (1 << TRACE_EVENT_FL_USE_CALL_FILTER_BIT),
276 TRACE_EVENT_FL_TRACEPOINT = (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
277 };
278
279 struct ftrace_event_call {
280 struct list_head list;
281 struct ftrace_event_class *class;
282 union {
283 char *name;
284 /* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
285 struct tracepoint *tp;
286 };
287 struct trace_event event;
288 const char *print_fmt;
289 struct event_filter *filter;
290 void *mod;
291 void *data;
292 /*
293 * bit 0: filter_active
294 * bit 1: allow trace by non root (cap any)
295 * bit 2: failed to apply filter
296 * bit 3: ftrace internal event (do not enable)
297 * bit 4: Event was enabled by module
298 * bit 5: use call filter rather than file filter
299 * bit 6: Event is a tracepoint
300 */
301 int flags; /* static flags of different events */
302
303 #ifdef CONFIG_PERF_EVENTS
304 int perf_refcount;
305 struct hlist_head __percpu *perf_events;
306
307 int (*perf_perm)(struct ftrace_event_call *,
308 struct perf_event *);
309 #endif
310 };
311
312 static inline const char *
313 ftrace_event_name(struct ftrace_event_call *call)
314 {
315 if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
316 return call->tp ? call->tp->name : NULL;
317 else
318 return call->name;
319 }
320
321 struct trace_array;
322 struct ftrace_subsystem_dir;
323
324 enum {
325 FTRACE_EVENT_FL_ENABLED_BIT,
326 FTRACE_EVENT_FL_RECORDED_CMD_BIT,
327 FTRACE_EVENT_FL_FILTERED_BIT,
328 FTRACE_EVENT_FL_NO_SET_FILTER_BIT,
329 FTRACE_EVENT_FL_SOFT_MODE_BIT,
330 FTRACE_EVENT_FL_SOFT_DISABLED_BIT,
331 FTRACE_EVENT_FL_TRIGGER_MODE_BIT,
332 FTRACE_EVENT_FL_TRIGGER_COND_BIT,
333 };
334
335 /*
336 * Ftrace event file flags:
337 * ENABLED - The event is enabled
338 * RECORDED_CMD - The comms should be recorded at sched_switch
339 * FILTERED - The event has a filter attached
340 * NO_SET_FILTER - Set when filter has error and is to be ignored
341 * SOFT_MODE - The event is enabled/disabled by SOFT_DISABLED
342 * SOFT_DISABLED - When set, do not trace the event (even though its
343 * tracepoint may be enabled)
344 * TRIGGER_MODE - When set, invoke the triggers associated with the event
345 * TRIGGER_COND - When set, one or more triggers has an associated filter
346 */
347 enum {
348 FTRACE_EVENT_FL_ENABLED = (1 << FTRACE_EVENT_FL_ENABLED_BIT),
349 FTRACE_EVENT_FL_RECORDED_CMD = (1 << FTRACE_EVENT_FL_RECORDED_CMD_BIT),
350 FTRACE_EVENT_FL_FILTERED = (1 << FTRACE_EVENT_FL_FILTERED_BIT),
351 FTRACE_EVENT_FL_NO_SET_FILTER = (1 << FTRACE_EVENT_FL_NO_SET_FILTER_BIT),
352 FTRACE_EVENT_FL_SOFT_MODE = (1 << FTRACE_EVENT_FL_SOFT_MODE_BIT),
353 FTRACE_EVENT_FL_SOFT_DISABLED = (1 << FTRACE_EVENT_FL_SOFT_DISABLED_BIT),
354 FTRACE_EVENT_FL_TRIGGER_MODE = (1 << FTRACE_EVENT_FL_TRIGGER_MODE_BIT),
355 FTRACE_EVENT_FL_TRIGGER_COND = (1 << FTRACE_EVENT_FL_TRIGGER_COND_BIT),
356 };
357
358 struct ftrace_event_file {
359 struct list_head list;
360 struct ftrace_event_call *event_call;
361 struct event_filter *filter;
362 struct dentry *dir;
363 struct trace_array *tr;
364 struct ftrace_subsystem_dir *system;
365 struct list_head triggers;
366
367 /*
368 * 32 bit flags:
369 * bit 0: enabled
370 * bit 1: enabled cmd record
371 * bit 2: enable/disable with the soft disable bit
372 * bit 3: soft disabled
373 * bit 4: trigger enabled
374 *
375 * Note: The bits must be set atomically to prevent races
376 * from other writers. Reads of flags do not need to be in
377 * sync as they occur in critical sections. But the way flags
378 * is currently used, these changes do not affect the code
379 * except that when a change is made, it may have a slight
380 * delay in propagating the changes to other CPUs due to
381 * caching and such. Which is mostly OK ;-)
382 */
383 unsigned long flags;
384 atomic_t sm_ref; /* soft-mode reference counter */
385 atomic_t tm_ref; /* trigger-mode reference counter */
386 };
387
388 #define __TRACE_EVENT_FLAGS(name, value) \
389 static int __init trace_init_flags_##name(void) \
390 { \
391 event_##name.flags |= value; \
392 return 0; \
393 } \
394 early_initcall(trace_init_flags_##name);
395
396 #define __TRACE_EVENT_PERF_PERM(name, expr...) \
397 static int perf_perm_##name(struct ftrace_event_call *tp_event, \
398 struct perf_event *p_event) \
399 { \
400 return ({ expr; }); \
401 } \
402 static int __init trace_init_perf_perm_##name(void) \
403 { \
404 event_##name.perf_perm = &perf_perm_##name; \
405 return 0; \
406 } \
407 early_initcall(trace_init_perf_perm_##name);
408
409 #define PERF_MAX_TRACE_SIZE 2048
410
411 #define MAX_FILTER_STR_VAL 256 /* Should handle KSYM_SYMBOL_LEN */
412
413 enum event_trigger_type {
414 ETT_NONE = (0),
415 ETT_TRACE_ONOFF = (1 << 0),
416 ETT_SNAPSHOT = (1 << 1),
417 ETT_STACKTRACE = (1 << 2),
418 ETT_EVENT_ENABLE = (1 << 3),
419 };
420
421 extern int filter_match_preds(struct event_filter *filter, void *rec);
422
423 extern int filter_check_discard(struct ftrace_event_file *file, void *rec,
424 struct ring_buffer *buffer,
425 struct ring_buffer_event *event);
426 extern int call_filter_check_discard(struct ftrace_event_call *call, void *rec,
427 struct ring_buffer *buffer,
428 struct ring_buffer_event *event);
429 extern enum event_trigger_type event_triggers_call(struct ftrace_event_file *file,
430 void *rec);
431 extern void event_triggers_post_call(struct ftrace_event_file *file,
432 enum event_trigger_type tt);
433
434 /**
435 * ftrace_trigger_soft_disabled - do triggers and test if soft disabled
436 * @file: The file pointer of the event to test
437 *
438 * If any triggers without filters are attached to this event, they
439 * will be called here. If the event is soft disabled and has no
440 * triggers that require testing the fields, it will return true,
441 * otherwise false.
442 */
443 static inline bool
444 ftrace_trigger_soft_disabled(struct ftrace_event_file *file)
445 {
446 unsigned long eflags = file->flags;
447
448 if (!(eflags & FTRACE_EVENT_FL_TRIGGER_COND)) {
449 if (eflags & FTRACE_EVENT_FL_TRIGGER_MODE)
450 event_triggers_call(file, NULL);
451 if (eflags & FTRACE_EVENT_FL_SOFT_DISABLED)
452 return true;
453 }
454 return false;
455 }
456
457 /*
458 * Helper function for event_trigger_unlock_commit{_regs}().
459 * If there are event triggers attached to this event that requires
460 * filtering against its fields, then they wil be called as the
461 * entry already holds the field information of the current event.
462 *
463 * It also checks if the event should be discarded or not.
464 * It is to be discarded if the event is soft disabled and the
465 * event was only recorded to process triggers, or if the event
466 * filter is active and this event did not match the filters.
467 *
468 * Returns true if the event is discarded, false otherwise.
469 */
470 static inline bool
471 __event_trigger_test_discard(struct ftrace_event_file *file,
472 struct ring_buffer *buffer,
473 struct ring_buffer_event *event,
474 void *entry,
475 enum event_trigger_type *tt)
476 {
477 unsigned long eflags = file->flags;
478
479 if (eflags & FTRACE_EVENT_FL_TRIGGER_COND)
480 *tt = event_triggers_call(file, entry);
481
482 if (test_bit(FTRACE_EVENT_FL_SOFT_DISABLED_BIT, &file->flags))
483 ring_buffer_discard_commit(buffer, event);
484 else if (!filter_check_discard(file, entry, buffer, event))
485 return false;
486
487 return true;
488 }
489
490 /**
491 * event_trigger_unlock_commit - handle triggers and finish event commit
492 * @file: The file pointer assoctiated to the event
493 * @buffer: The ring buffer that the event is being written to
494 * @event: The event meta data in the ring buffer
495 * @entry: The event itself
496 * @irq_flags: The state of the interrupts at the start of the event
497 * @pc: The state of the preempt count at the start of the event.
498 *
499 * This is a helper function to handle triggers that require data
500 * from the event itself. It also tests the event against filters and
501 * if the event is soft disabled and should be discarded.
502 */
503 static inline void
504 event_trigger_unlock_commit(struct ftrace_event_file *file,
505 struct ring_buffer *buffer,
506 struct ring_buffer_event *event,
507 void *entry, unsigned long irq_flags, int pc)
508 {
509 enum event_trigger_type tt = ETT_NONE;
510
511 if (!__event_trigger_test_discard(file, buffer, event, entry, &tt))
512 trace_buffer_unlock_commit(buffer, event, irq_flags, pc);
513
514 if (tt)
515 event_triggers_post_call(file, tt);
516 }
517
518 /**
519 * event_trigger_unlock_commit_regs - handle triggers and finish event commit
520 * @file: The file pointer assoctiated to the event
521 * @buffer: The ring buffer that the event is being written to
522 * @event: The event meta data in the ring buffer
523 * @entry: The event itself
524 * @irq_flags: The state of the interrupts at the start of the event
525 * @pc: The state of the preempt count at the start of the event.
526 *
527 * This is a helper function to handle triggers that require data
528 * from the event itself. It also tests the event against filters and
529 * if the event is soft disabled and should be discarded.
530 *
531 * Same as event_trigger_unlock_commit() but calls
532 * trace_buffer_unlock_commit_regs() instead of trace_buffer_unlock_commit().
533 */
534 static inline void
535 event_trigger_unlock_commit_regs(struct ftrace_event_file *file,
536 struct ring_buffer *buffer,
537 struct ring_buffer_event *event,
538 void *entry, unsigned long irq_flags, int pc,
539 struct pt_regs *regs)
540 {
541 enum event_trigger_type tt = ETT_NONE;
542
543 if (!__event_trigger_test_discard(file, buffer, event, entry, &tt))
544 trace_buffer_unlock_commit_regs(buffer, event,
545 irq_flags, pc, regs);
546
547 if (tt)
548 event_triggers_post_call(file, tt);
549 }
550
551 enum {
552 FILTER_OTHER = 0,
553 FILTER_STATIC_STRING,
554 FILTER_DYN_STRING,
555 FILTER_PTR_STRING,
556 FILTER_TRACE_FN,
557 };
558
559 extern int trace_event_raw_init(struct ftrace_event_call *call);
560 extern int trace_define_field(struct ftrace_event_call *call, const char *type,
561 const char *name, int offset, int size,
562 int is_signed, int filter_type);
563 extern int trace_add_event_call(struct ftrace_event_call *call);
564 extern int trace_remove_event_call(struct ftrace_event_call *call);
565
566 #define is_signed_type(type) (((type)(-1)) < (type)1)
567
568 int trace_set_clr_event(const char *system, const char *event, int set);
569
570 /*
571 * The double __builtin_constant_p is because gcc will give us an error
572 * if we try to allocate the static variable to fmt if it is not a
573 * constant. Even with the outer if statement optimizing out.
574 */
575 #define event_trace_printk(ip, fmt, args...) \
576 do { \
577 __trace_printk_check_format(fmt, ##args); \
578 tracing_record_cmdline(current); \
579 if (__builtin_constant_p(fmt)) { \
580 static const char *trace_printk_fmt \
581 __attribute__((section("__trace_printk_fmt"))) = \
582 __builtin_constant_p(fmt) ? fmt : NULL; \
583 \
584 __trace_bprintk(ip, trace_printk_fmt, ##args); \
585 } else \
586 __trace_printk(ip, fmt, ##args); \
587 } while (0)
588
589 #ifdef CONFIG_PERF_EVENTS
590 struct perf_event;
591
592 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
593
594 extern int perf_trace_init(struct perf_event *event);
595 extern void perf_trace_destroy(struct perf_event *event);
596 extern int perf_trace_add(struct perf_event *event, int flags);
597 extern void perf_trace_del(struct perf_event *event, int flags);
598 extern int ftrace_profile_set_filter(struct perf_event *event, int event_id,
599 char *filter_str);
600 extern void ftrace_profile_free_filter(struct perf_event *event);
601 extern void *perf_trace_buf_prepare(int size, unsigned short type,
602 struct pt_regs **regs, int *rctxp);
603
604 static inline void
605 perf_trace_buf_submit(void *raw_data, int size, int rctx, u64 addr,
606 u64 count, struct pt_regs *regs, void *head,
607 struct task_struct *task)
608 {
609 perf_tp_event(addr, count, raw_data, size, regs, head, rctx, task);
610 }
611 #endif
612
613 #endif /* _LINUX_FTRACE_EVENT_H */