xref: /linux-6.15/include/linux/trace_events.h (revision b5734e99)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 
3 #ifndef _LINUX_TRACE_EVENT_H
4 #define _LINUX_TRACE_EVENT_H
5 
6 #include <linux/ring_buffer.h>
7 #include <linux/trace_seq.h>
8 #include <linux/percpu.h>
9 #include <linux/hardirq.h>
10 #include <linux/perf_event.h>
11 #include <linux/tracepoint.h>
12 
13 struct trace_array;
14 struct array_buffer;
15 struct tracer;
16 struct dentry;
17 struct bpf_prog;
18 
19 const char *trace_print_flags_seq(struct trace_seq *p, const char *delim,
20 				  unsigned long flags,
21 				  const struct trace_print_flags *flag_array);
22 
23 const char *trace_print_symbols_seq(struct trace_seq *p, unsigned long val,
24 				    const struct trace_print_flags *symbol_array);
25 
26 #if BITS_PER_LONG == 32
27 const char *trace_print_flags_seq_u64(struct trace_seq *p, const char *delim,
28 		      unsigned long long flags,
29 		      const struct trace_print_flags_u64 *flag_array);
30 
31 const char *trace_print_symbols_seq_u64(struct trace_seq *p,
32 					unsigned long long val,
33 					const struct trace_print_flags_u64
34 								 *symbol_array);
35 #endif
36 
37 const char *trace_print_bitmask_seq(struct trace_seq *p, void *bitmask_ptr,
38 				    unsigned int bitmask_size);
39 
40 const char *trace_print_hex_seq(struct trace_seq *p,
41 				const unsigned char *buf, int len,
42 				bool concatenate);
43 
44 const char *trace_print_array_seq(struct trace_seq *p,
45 				   const void *buf, int count,
46 				   size_t el_size);
47 
48 const char *
49 trace_print_hex_dump_seq(struct trace_seq *p, const char *prefix_str,
50 			 int prefix_type, int rowsize, int groupsize,
51 			 const void *buf, size_t len, bool ascii);
52 
53 struct trace_iterator;
54 struct trace_event;
55 
56 int trace_raw_output_prep(struct trace_iterator *iter,
57 			  struct trace_event *event);
58 
59 /*
60  * The trace entry - the most basic unit of tracing. This is what
61  * is printed in the end as a single line in the trace output, such as:
62  *
63  *     bash-15816 [01]   235.197585: idle_cpu <- irq_enter
64  */
65 struct trace_entry {
66 	unsigned short		type;
67 	unsigned char		flags;
68 	unsigned char		preempt_count;
69 	int			pid;
70 };
71 
72 #define TRACE_EVENT_TYPE_MAX						\
73 	((1 << (sizeof(((struct trace_entry *)0)->type) * 8)) - 1)
74 
75 /*
76  * Trace iterator - used by printout routines who present trace
77  * results to users and which routines might sleep, etc:
78  */
79 struct trace_iterator {
80 	struct trace_array	*tr;
81 	struct tracer		*trace;
82 	struct array_buffer	*array_buffer;
83 	void			*private;
84 	int			cpu_file;
85 	struct mutex		mutex;
86 	struct ring_buffer_iter	**buffer_iter;
87 	unsigned long		iter_flags;
88 	void			*temp;	/* temp holder */
89 	unsigned int		temp_size;
90 
91 	/* trace_seq for __print_flags() and __print_symbolic() etc. */
92 	struct trace_seq	tmp_seq;
93 
94 	cpumask_var_t		started;
95 
96 	/* it's true when current open file is snapshot */
97 	bool			snapshot;
98 
99 	/* The below is zeroed out in pipe_read */
100 	struct trace_seq	seq;
101 	struct trace_entry	*ent;
102 	unsigned long		lost_events;
103 	int			leftover;
104 	int			ent_size;
105 	int			cpu;
106 	u64			ts;
107 
108 	loff_t			pos;
109 	long			idx;
110 
111 	/* All new field here will be zeroed out in pipe_read */
112 };
113 
114 enum trace_iter_flags {
115 	TRACE_FILE_LAT_FMT	= 1,
116 	TRACE_FILE_ANNOTATE	= 2,
117 	TRACE_FILE_TIME_IN_NS	= 4,
118 };
119 
120 
121 typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
122 				      int flags, struct trace_event *event);
123 
124 struct trace_event_functions {
125 	trace_print_func	trace;
126 	trace_print_func	raw;
127 	trace_print_func	hex;
128 	trace_print_func	binary;
129 };
130 
131 struct trace_event {
132 	struct hlist_node		node;
133 	struct list_head		list;
134 	int				type;
135 	struct trace_event_functions	*funcs;
136 };
137 
138 extern int register_trace_event(struct trace_event *event);
139 extern int unregister_trace_event(struct trace_event *event);
140 
141 /* Return values for print_line callback */
142 enum print_line_t {
143 	TRACE_TYPE_PARTIAL_LINE	= 0,	/* Retry after flushing the seq */
144 	TRACE_TYPE_HANDLED	= 1,
145 	TRACE_TYPE_UNHANDLED	= 2,	/* Relay to other output functions */
146 	TRACE_TYPE_NO_CONSUME	= 3	/* Handled but ask to not consume */
147 };
148 
149 enum print_line_t trace_handle_return(struct trace_seq *s);
150 
151 static inline void tracing_generic_entry_update(struct trace_entry *entry,
152 						unsigned short type,
153 						unsigned int trace_ctx)
154 {
155 	entry->preempt_count		= trace_ctx & 0xff;
156 	entry->pid			= current->pid;
157 	entry->type			= type;
158 	entry->flags =			trace_ctx >> 16;
159 }
160 
161 unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status);
162 
163 enum trace_flag_type {
164 	TRACE_FLAG_IRQS_OFF		= 0x01,
165 	TRACE_FLAG_IRQS_NOSUPPORT	= 0x02,
166 	TRACE_FLAG_NEED_RESCHED		= 0x04,
167 	TRACE_FLAG_HARDIRQ		= 0x08,
168 	TRACE_FLAG_SOFTIRQ		= 0x10,
169 	TRACE_FLAG_PREEMPT_RESCHED	= 0x20,
170 	TRACE_FLAG_NMI			= 0x40,
171 };
172 
173 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
174 static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
175 {
176 	unsigned int irq_status = irqs_disabled_flags(irqflags) ?
177 		TRACE_FLAG_IRQS_OFF : 0;
178 	return tracing_gen_ctx_irq_test(irq_status);
179 }
180 static inline unsigned int tracing_gen_ctx(void)
181 {
182 	unsigned long irqflags;
183 
184 	local_save_flags(irqflags);
185 	return tracing_gen_ctx_flags(irqflags);
186 }
187 #else
188 
189 static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
190 {
191 	return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
192 }
193 static inline unsigned int tracing_gen_ctx(void)
194 {
195 	return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
196 }
197 #endif
198 
199 static inline unsigned int tracing_gen_ctx_dec(void)
200 {
201 	unsigned int trace_ctx;
202 
203 	trace_ctx = tracing_gen_ctx();
204 	/*
205 	 * Subtract one from the preeption counter if preemption is enabled,
206 	 * see trace_event_buffer_reserve()for details.
207 	 */
208 	if (IS_ENABLED(CONFIG_PREEMPTION))
209 		trace_ctx--;
210 	return trace_ctx;
211 }
212 
213 struct trace_event_file;
214 
215 struct ring_buffer_event *
216 trace_event_buffer_lock_reserve(struct trace_buffer **current_buffer,
217 				struct trace_event_file *trace_file,
218 				int type, unsigned long len,
219 				unsigned int trace_ctx);
220 
221 #define TRACE_RECORD_CMDLINE	BIT(0)
222 #define TRACE_RECORD_TGID	BIT(1)
223 
224 void tracing_record_taskinfo(struct task_struct *task, int flags);
225 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
226 					  struct task_struct *next, int flags);
227 
228 void tracing_record_cmdline(struct task_struct *task);
229 void tracing_record_tgid(struct task_struct *task);
230 
231 int trace_output_call(struct trace_iterator *iter, char *name, char *fmt, ...);
232 
233 struct event_filter;
234 
235 enum trace_reg {
236 	TRACE_REG_REGISTER,
237 	TRACE_REG_UNREGISTER,
238 #ifdef CONFIG_PERF_EVENTS
239 	TRACE_REG_PERF_REGISTER,
240 	TRACE_REG_PERF_UNREGISTER,
241 	TRACE_REG_PERF_OPEN,
242 	TRACE_REG_PERF_CLOSE,
243 	/*
244 	 * These (ADD/DEL) use a 'boolean' return value, where 1 (true) means a
245 	 * custom action was taken and the default action is not to be
246 	 * performed.
247 	 */
248 	TRACE_REG_PERF_ADD,
249 	TRACE_REG_PERF_DEL,
250 #endif
251 };
252 
253 struct trace_event_call;
254 
255 #define TRACE_FUNCTION_TYPE ((const char *)~0UL)
256 
257 struct trace_event_fields {
258 	const char *type;
259 	union {
260 		struct {
261 			const char *name;
262 			const int  size;
263 			const int  align;
264 			const int  is_signed;
265 			const int  filter_type;
266 		};
267 		int (*define_fields)(struct trace_event_call *);
268 	};
269 };
270 
271 struct trace_event_class {
272 	const char		*system;
273 	void			*probe;
274 #ifdef CONFIG_PERF_EVENTS
275 	void			*perf_probe;
276 #endif
277 	int			(*reg)(struct trace_event_call *event,
278 				       enum trace_reg type, void *data);
279 	struct trace_event_fields *fields_array;
280 	struct list_head	*(*get_fields)(struct trace_event_call *);
281 	struct list_head	fields;
282 	int			(*raw_init)(struct trace_event_call *);
283 };
284 
285 extern int trace_event_reg(struct trace_event_call *event,
286 			    enum trace_reg type, void *data);
287 
288 struct trace_event_buffer {
289 	struct trace_buffer		*buffer;
290 	struct ring_buffer_event	*event;
291 	struct trace_event_file		*trace_file;
292 	void				*entry;
293 	unsigned int			trace_ctx;
294 	struct pt_regs			*regs;
295 };
296 
297 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
298 				  struct trace_event_file *trace_file,
299 				  unsigned long len);
300 
301 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer);
302 
303 enum {
304 	TRACE_EVENT_FL_FILTERED_BIT,
305 	TRACE_EVENT_FL_CAP_ANY_BIT,
306 	TRACE_EVENT_FL_NO_SET_FILTER_BIT,
307 	TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
308 	TRACE_EVENT_FL_TRACEPOINT_BIT,
309 	TRACE_EVENT_FL_KPROBE_BIT,
310 	TRACE_EVENT_FL_UPROBE_BIT,
311 };
312 
313 /*
314  * Event flags:
315  *  FILTERED	  - The event has a filter attached
316  *  CAP_ANY	  - Any user can enable for perf
317  *  NO_SET_FILTER - Set when filter has error and is to be ignored
318  *  IGNORE_ENABLE - For trace internal events, do not enable with debugfs file
319  *  TRACEPOINT    - Event is a tracepoint
320  *  KPROBE        - Event is a kprobe
321  *  UPROBE        - Event is a uprobe
322  */
323 enum {
324 	TRACE_EVENT_FL_FILTERED		= (1 << TRACE_EVENT_FL_FILTERED_BIT),
325 	TRACE_EVENT_FL_CAP_ANY		= (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
326 	TRACE_EVENT_FL_NO_SET_FILTER	= (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
327 	TRACE_EVENT_FL_IGNORE_ENABLE	= (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
328 	TRACE_EVENT_FL_TRACEPOINT	= (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
329 	TRACE_EVENT_FL_KPROBE		= (1 << TRACE_EVENT_FL_KPROBE_BIT),
330 	TRACE_EVENT_FL_UPROBE		= (1 << TRACE_EVENT_FL_UPROBE_BIT),
331 };
332 
333 #define TRACE_EVENT_FL_UKPROBE (TRACE_EVENT_FL_KPROBE | TRACE_EVENT_FL_UPROBE)
334 
335 struct trace_event_call {
336 	struct list_head	list;
337 	struct trace_event_class *class;
338 	union {
339 		char			*name;
340 		/* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
341 		struct tracepoint	*tp;
342 	};
343 	struct trace_event	event;
344 	char			*print_fmt;
345 	struct event_filter	*filter;
346 	void			*mod;
347 	void			*data;
348 	/*
349 	 *   bit 0:		filter_active
350 	 *   bit 1:		allow trace by non root (cap any)
351 	 *   bit 2:		failed to apply filter
352 	 *   bit 3:		trace internal event (do not enable)
353 	 *   bit 4:		Event was enabled by module
354 	 *   bit 5:		use call filter rather than file filter
355 	 *   bit 6:		Event is a tracepoint
356 	 */
357 	int			flags; /* static flags of different events */
358 
359 #ifdef CONFIG_PERF_EVENTS
360 	int				perf_refcount;
361 	struct hlist_head __percpu	*perf_events;
362 	struct bpf_prog_array __rcu	*prog_array;
363 
364 	int	(*perf_perm)(struct trace_event_call *,
365 			     struct perf_event *);
366 #endif
367 };
368 
369 #ifdef CONFIG_PERF_EVENTS
370 static inline bool bpf_prog_array_valid(struct trace_event_call *call)
371 {
372 	/*
373 	 * This inline function checks whether call->prog_array
374 	 * is valid or not. The function is called in various places,
375 	 * outside rcu_read_lock/unlock, as a heuristic to speed up execution.
376 	 *
377 	 * If this function returns true, and later call->prog_array
378 	 * becomes false inside rcu_read_lock/unlock region,
379 	 * we bail out then. If this function return false,
380 	 * there is a risk that we might miss a few events if the checking
381 	 * were delayed until inside rcu_read_lock/unlock region and
382 	 * call->prog_array happened to become non-NULL then.
383 	 *
384 	 * Here, READ_ONCE() is used instead of rcu_access_pointer().
385 	 * rcu_access_pointer() requires the actual definition of
386 	 * "struct bpf_prog_array" while READ_ONCE() only needs
387 	 * a declaration of the same type.
388 	 */
389 	return !!READ_ONCE(call->prog_array);
390 }
391 #endif
392 
393 static inline const char *
394 trace_event_name(struct trace_event_call *call)
395 {
396 	if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
397 		return call->tp ? call->tp->name : NULL;
398 	else
399 		return call->name;
400 }
401 
402 static inline struct list_head *
403 trace_get_fields(struct trace_event_call *event_call)
404 {
405 	if (!event_call->class->get_fields)
406 		return &event_call->class->fields;
407 	return event_call->class->get_fields(event_call);
408 }
409 
410 struct trace_array;
411 struct trace_subsystem_dir;
412 
413 enum {
414 	EVENT_FILE_FL_ENABLED_BIT,
415 	EVENT_FILE_FL_RECORDED_CMD_BIT,
416 	EVENT_FILE_FL_RECORDED_TGID_BIT,
417 	EVENT_FILE_FL_FILTERED_BIT,
418 	EVENT_FILE_FL_NO_SET_FILTER_BIT,
419 	EVENT_FILE_FL_SOFT_MODE_BIT,
420 	EVENT_FILE_FL_SOFT_DISABLED_BIT,
421 	EVENT_FILE_FL_TRIGGER_MODE_BIT,
422 	EVENT_FILE_FL_TRIGGER_COND_BIT,
423 	EVENT_FILE_FL_PID_FILTER_BIT,
424 	EVENT_FILE_FL_WAS_ENABLED_BIT,
425 };
426 
427 extern struct trace_event_file *trace_get_event_file(const char *instance,
428 						     const char *system,
429 						     const char *event);
430 extern void trace_put_event_file(struct trace_event_file *file);
431 
432 #define MAX_DYNEVENT_CMD_LEN	(2048)
433 
434 enum dynevent_type {
435 	DYNEVENT_TYPE_SYNTH = 1,
436 	DYNEVENT_TYPE_KPROBE,
437 	DYNEVENT_TYPE_NONE,
438 };
439 
440 struct dynevent_cmd;
441 
442 typedef int (*dynevent_create_fn_t)(struct dynevent_cmd *cmd);
443 
444 struct dynevent_cmd {
445 	struct seq_buf		seq;
446 	const char		*event_name;
447 	unsigned int		n_fields;
448 	enum dynevent_type	type;
449 	dynevent_create_fn_t	run_command;
450 	void			*private_data;
451 };
452 
453 extern int dynevent_create(struct dynevent_cmd *cmd);
454 
455 extern int synth_event_delete(const char *name);
456 
457 extern void synth_event_cmd_init(struct dynevent_cmd *cmd,
458 				 char *buf, int maxlen);
459 
460 extern int __synth_event_gen_cmd_start(struct dynevent_cmd *cmd,
461 				       const char *name,
462 				       struct module *mod, ...);
463 
464 #define synth_event_gen_cmd_start(cmd, name, mod, ...)	\
465 	__synth_event_gen_cmd_start(cmd, name, mod, ## __VA_ARGS__, NULL)
466 
467 struct synth_field_desc {
468 	const char *type;
469 	const char *name;
470 };
471 
472 extern int synth_event_gen_cmd_array_start(struct dynevent_cmd *cmd,
473 					   const char *name,
474 					   struct module *mod,
475 					   struct synth_field_desc *fields,
476 					   unsigned int n_fields);
477 extern int synth_event_create(const char *name,
478 			      struct synth_field_desc *fields,
479 			      unsigned int n_fields, struct module *mod);
480 
481 extern int synth_event_add_field(struct dynevent_cmd *cmd,
482 				 const char *type,
483 				 const char *name);
484 extern int synth_event_add_field_str(struct dynevent_cmd *cmd,
485 				     const char *type_name);
486 extern int synth_event_add_fields(struct dynevent_cmd *cmd,
487 				  struct synth_field_desc *fields,
488 				  unsigned int n_fields);
489 
490 #define synth_event_gen_cmd_end(cmd)	\
491 	dynevent_create(cmd)
492 
493 struct synth_event;
494 
495 struct synth_event_trace_state {
496 	struct trace_event_buffer fbuffer;
497 	struct synth_trace_event *entry;
498 	struct trace_buffer *buffer;
499 	struct synth_event *event;
500 	unsigned int cur_field;
501 	unsigned int n_u64;
502 	bool disabled;
503 	bool add_next;
504 	bool add_name;
505 };
506 
507 extern int synth_event_trace(struct trace_event_file *file,
508 			     unsigned int n_vals, ...);
509 extern int synth_event_trace_array(struct trace_event_file *file, u64 *vals,
510 				   unsigned int n_vals);
511 extern int synth_event_trace_start(struct trace_event_file *file,
512 				   struct synth_event_trace_state *trace_state);
513 extern int synth_event_add_next_val(u64 val,
514 				    struct synth_event_trace_state *trace_state);
515 extern int synth_event_add_val(const char *field_name, u64 val,
516 			       struct synth_event_trace_state *trace_state);
517 extern int synth_event_trace_end(struct synth_event_trace_state *trace_state);
518 
519 extern int kprobe_event_delete(const char *name);
520 
521 extern void kprobe_event_cmd_init(struct dynevent_cmd *cmd,
522 				  char *buf, int maxlen);
523 
524 #define kprobe_event_gen_cmd_start(cmd, name, loc, ...)			\
525 	__kprobe_event_gen_cmd_start(cmd, false, name, loc, ## __VA_ARGS__, NULL)
526 
527 #define kretprobe_event_gen_cmd_start(cmd, name, loc, ...)		\
528 	__kprobe_event_gen_cmd_start(cmd, true, name, loc, ## __VA_ARGS__, NULL)
529 
530 extern int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd,
531 					bool kretprobe,
532 					const char *name,
533 					const char *loc, ...);
534 
535 #define kprobe_event_add_fields(cmd, ...)	\
536 	__kprobe_event_add_fields(cmd, ## __VA_ARGS__, NULL)
537 
538 #define kprobe_event_add_field(cmd, field)	\
539 	__kprobe_event_add_fields(cmd, field, NULL)
540 
541 extern int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...);
542 
543 #define kprobe_event_gen_cmd_end(cmd)		\
544 	dynevent_create(cmd)
545 
546 #define kretprobe_event_gen_cmd_end(cmd)	\
547 	dynevent_create(cmd)
548 
549 /*
550  * Event file flags:
551  *  ENABLED	  - The event is enabled
552  *  RECORDED_CMD  - The comms should be recorded at sched_switch
553  *  RECORDED_TGID - The tgids should be recorded at sched_switch
554  *  FILTERED	  - The event has a filter attached
555  *  NO_SET_FILTER - Set when filter has error and is to be ignored
556  *  SOFT_MODE     - The event is enabled/disabled by SOFT_DISABLED
557  *  SOFT_DISABLED - When set, do not trace the event (even though its
558  *                   tracepoint may be enabled)
559  *  TRIGGER_MODE  - When set, invoke the triggers associated with the event
560  *  TRIGGER_COND  - When set, one or more triggers has an associated filter
561  *  PID_FILTER    - When set, the event is filtered based on pid
562  *  WAS_ENABLED   - Set when enabled to know to clear trace on module removal
563  */
564 enum {
565 	EVENT_FILE_FL_ENABLED		= (1 << EVENT_FILE_FL_ENABLED_BIT),
566 	EVENT_FILE_FL_RECORDED_CMD	= (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
567 	EVENT_FILE_FL_RECORDED_TGID	= (1 << EVENT_FILE_FL_RECORDED_TGID_BIT),
568 	EVENT_FILE_FL_FILTERED		= (1 << EVENT_FILE_FL_FILTERED_BIT),
569 	EVENT_FILE_FL_NO_SET_FILTER	= (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
570 	EVENT_FILE_FL_SOFT_MODE		= (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
571 	EVENT_FILE_FL_SOFT_DISABLED	= (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
572 	EVENT_FILE_FL_TRIGGER_MODE	= (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
573 	EVENT_FILE_FL_TRIGGER_COND	= (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
574 	EVENT_FILE_FL_PID_FILTER	= (1 << EVENT_FILE_FL_PID_FILTER_BIT),
575 	EVENT_FILE_FL_WAS_ENABLED	= (1 << EVENT_FILE_FL_WAS_ENABLED_BIT),
576 };
577 
578 struct trace_event_file {
579 	struct list_head		list;
580 	struct trace_event_call		*event_call;
581 	struct event_filter __rcu	*filter;
582 	struct dentry			*dir;
583 	struct trace_array		*tr;
584 	struct trace_subsystem_dir	*system;
585 	struct list_head		triggers;
586 
587 	/*
588 	 * 32 bit flags:
589 	 *   bit 0:		enabled
590 	 *   bit 1:		enabled cmd record
591 	 *   bit 2:		enable/disable with the soft disable bit
592 	 *   bit 3:		soft disabled
593 	 *   bit 4:		trigger enabled
594 	 *
595 	 * Note: The bits must be set atomically to prevent races
596 	 * from other writers. Reads of flags do not need to be in
597 	 * sync as they occur in critical sections. But the way flags
598 	 * is currently used, these changes do not affect the code
599 	 * except that when a change is made, it may have a slight
600 	 * delay in propagating the changes to other CPUs due to
601 	 * caching and such. Which is mostly OK ;-)
602 	 */
603 	unsigned long		flags;
604 	atomic_t		sm_ref;	/* soft-mode reference counter */
605 	atomic_t		tm_ref;	/* trigger-mode reference counter */
606 };
607 
608 #define __TRACE_EVENT_FLAGS(name, value)				\
609 	static int __init trace_init_flags_##name(void)			\
610 	{								\
611 		event_##name.flags |= value;				\
612 		return 0;						\
613 	}								\
614 	early_initcall(trace_init_flags_##name);
615 
616 #define __TRACE_EVENT_PERF_PERM(name, expr...)				\
617 	static int perf_perm_##name(struct trace_event_call *tp_event, \
618 				    struct perf_event *p_event)		\
619 	{								\
620 		return ({ expr; });					\
621 	}								\
622 	static int __init trace_init_perf_perm_##name(void)		\
623 	{								\
624 		event_##name.perf_perm = &perf_perm_##name;		\
625 		return 0;						\
626 	}								\
627 	early_initcall(trace_init_perf_perm_##name);
628 
629 #define PERF_MAX_TRACE_SIZE	2048
630 
631 #define MAX_FILTER_STR_VAL	256	/* Should handle KSYM_SYMBOL_LEN */
632 
633 enum event_trigger_type {
634 	ETT_NONE		= (0),
635 	ETT_TRACE_ONOFF		= (1 << 0),
636 	ETT_SNAPSHOT		= (1 << 1),
637 	ETT_STACKTRACE		= (1 << 2),
638 	ETT_EVENT_ENABLE	= (1 << 3),
639 	ETT_EVENT_HIST		= (1 << 4),
640 	ETT_HIST_ENABLE		= (1 << 5),
641 };
642 
643 extern int filter_match_preds(struct event_filter *filter, void *rec);
644 
645 extern enum event_trigger_type
646 event_triggers_call(struct trace_event_file *file, void *rec,
647 		    struct ring_buffer_event *event);
648 extern void
649 event_triggers_post_call(struct trace_event_file *file,
650 			 enum event_trigger_type tt);
651 
652 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file);
653 
654 /**
655  * trace_trigger_soft_disabled - do triggers and test if soft disabled
656  * @file: The file pointer of the event to test
657  *
658  * If any triggers without filters are attached to this event, they
659  * will be called here. If the event is soft disabled and has no
660  * triggers that require testing the fields, it will return true,
661  * otherwise false.
662  */
663 static inline bool
664 trace_trigger_soft_disabled(struct trace_event_file *file)
665 {
666 	unsigned long eflags = file->flags;
667 
668 	if (!(eflags & EVENT_FILE_FL_TRIGGER_COND)) {
669 		if (eflags & EVENT_FILE_FL_TRIGGER_MODE)
670 			event_triggers_call(file, NULL, NULL);
671 		if (eflags & EVENT_FILE_FL_SOFT_DISABLED)
672 			return true;
673 		if (eflags & EVENT_FILE_FL_PID_FILTER)
674 			return trace_event_ignore_this_pid(file);
675 	}
676 	return false;
677 }
678 
679 #ifdef CONFIG_BPF_EVENTS
680 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx);
681 int perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog);
682 void perf_event_detach_bpf_prog(struct perf_event *event);
683 int perf_event_query_prog_array(struct perf_event *event, void __user *info);
684 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
685 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
686 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name);
687 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp);
688 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
689 			    u32 *fd_type, const char **buf,
690 			    u64 *probe_offset, u64 *probe_addr);
691 #else
692 static inline unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
693 {
694 	return 1;
695 }
696 
697 static inline int
698 perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog)
699 {
700 	return -EOPNOTSUPP;
701 }
702 
703 static inline void perf_event_detach_bpf_prog(struct perf_event *event) { }
704 
705 static inline int
706 perf_event_query_prog_array(struct perf_event *event, void __user *info)
707 {
708 	return -EOPNOTSUPP;
709 }
710 static inline int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *p)
711 {
712 	return -EOPNOTSUPP;
713 }
714 static inline int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *p)
715 {
716 	return -EOPNOTSUPP;
717 }
718 static inline struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
719 {
720 	return NULL;
721 }
722 static inline void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
723 {
724 }
725 static inline int bpf_get_perf_event_info(const struct perf_event *event,
726 					  u32 *prog_id, u32 *fd_type,
727 					  const char **buf, u64 *probe_offset,
728 					  u64 *probe_addr)
729 {
730 	return -EOPNOTSUPP;
731 }
732 #endif
733 
734 enum {
735 	FILTER_OTHER = 0,
736 	FILTER_STATIC_STRING,
737 	FILTER_DYN_STRING,
738 	FILTER_PTR_STRING,
739 	FILTER_TRACE_FN,
740 	FILTER_COMM,
741 	FILTER_CPU,
742 };
743 
744 extern int trace_event_raw_init(struct trace_event_call *call);
745 extern int trace_define_field(struct trace_event_call *call, const char *type,
746 			      const char *name, int offset, int size,
747 			      int is_signed, int filter_type);
748 extern int trace_add_event_call(struct trace_event_call *call);
749 extern int trace_remove_event_call(struct trace_event_call *call);
750 extern int trace_event_get_offsets(struct trace_event_call *call);
751 
752 #define is_signed_type(type)	(((type)(-1)) < (type)1)
753 
754 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set);
755 int trace_set_clr_event(const char *system, const char *event, int set);
756 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
757 		const char *event, bool enable);
758 /*
759  * The double __builtin_constant_p is because gcc will give us an error
760  * if we try to allocate the static variable to fmt if it is not a
761  * constant. Even with the outer if statement optimizing out.
762  */
763 #define event_trace_printk(ip, fmt, args...)				\
764 do {									\
765 	__trace_printk_check_format(fmt, ##args);			\
766 	tracing_record_cmdline(current);				\
767 	if (__builtin_constant_p(fmt)) {				\
768 		static const char *trace_printk_fmt			\
769 		  __section("__trace_printk_fmt") =			\
770 			__builtin_constant_p(fmt) ? fmt : NULL;		\
771 									\
772 		__trace_bprintk(ip, trace_printk_fmt, ##args);		\
773 	} else								\
774 		__trace_printk(ip, fmt, ##args);			\
775 } while (0)
776 
777 #ifdef CONFIG_PERF_EVENTS
778 struct perf_event;
779 
780 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
781 DECLARE_PER_CPU(int, bpf_kprobe_override);
782 
783 extern int  perf_trace_init(struct perf_event *event);
784 extern void perf_trace_destroy(struct perf_event *event);
785 extern int  perf_trace_add(struct perf_event *event, int flags);
786 extern void perf_trace_del(struct perf_event *event, int flags);
787 #ifdef CONFIG_KPROBE_EVENTS
788 extern int  perf_kprobe_init(struct perf_event *event, bool is_retprobe);
789 extern void perf_kprobe_destroy(struct perf_event *event);
790 extern int bpf_get_kprobe_info(const struct perf_event *event,
791 			       u32 *fd_type, const char **symbol,
792 			       u64 *probe_offset, u64 *probe_addr,
793 			       bool perf_type_tracepoint);
794 #endif
795 #ifdef CONFIG_UPROBE_EVENTS
796 extern int  perf_uprobe_init(struct perf_event *event,
797 			     unsigned long ref_ctr_offset, bool is_retprobe);
798 extern void perf_uprobe_destroy(struct perf_event *event);
799 extern int bpf_get_uprobe_info(const struct perf_event *event,
800 			       u32 *fd_type, const char **filename,
801 			       u64 *probe_offset, bool perf_type_tracepoint);
802 #endif
803 extern int  ftrace_profile_set_filter(struct perf_event *event, int event_id,
804 				     char *filter_str);
805 extern void ftrace_profile_free_filter(struct perf_event *event);
806 void perf_trace_buf_update(void *record, u16 type);
807 void *perf_trace_buf_alloc(int size, struct pt_regs **regs, int *rctxp);
808 
809 void bpf_trace_run1(struct bpf_prog *prog, u64 arg1);
810 void bpf_trace_run2(struct bpf_prog *prog, u64 arg1, u64 arg2);
811 void bpf_trace_run3(struct bpf_prog *prog, u64 arg1, u64 arg2,
812 		    u64 arg3);
813 void bpf_trace_run4(struct bpf_prog *prog, u64 arg1, u64 arg2,
814 		    u64 arg3, u64 arg4);
815 void bpf_trace_run5(struct bpf_prog *prog, u64 arg1, u64 arg2,
816 		    u64 arg3, u64 arg4, u64 arg5);
817 void bpf_trace_run6(struct bpf_prog *prog, u64 arg1, u64 arg2,
818 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6);
819 void bpf_trace_run7(struct bpf_prog *prog, u64 arg1, u64 arg2,
820 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7);
821 void bpf_trace_run8(struct bpf_prog *prog, u64 arg1, u64 arg2,
822 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
823 		    u64 arg8);
824 void bpf_trace_run9(struct bpf_prog *prog, u64 arg1, u64 arg2,
825 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
826 		    u64 arg8, u64 arg9);
827 void bpf_trace_run10(struct bpf_prog *prog, u64 arg1, u64 arg2,
828 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
829 		     u64 arg8, u64 arg9, u64 arg10);
830 void bpf_trace_run11(struct bpf_prog *prog, u64 arg1, u64 arg2,
831 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
832 		     u64 arg8, u64 arg9, u64 arg10, u64 arg11);
833 void bpf_trace_run12(struct bpf_prog *prog, u64 arg1, u64 arg2,
834 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
835 		     u64 arg8, u64 arg9, u64 arg10, u64 arg11, u64 arg12);
836 void perf_trace_run_bpf_submit(void *raw_data, int size, int rctx,
837 			       struct trace_event_call *call, u64 count,
838 			       struct pt_regs *regs, struct hlist_head *head,
839 			       struct task_struct *task);
840 
841 static inline void
842 perf_trace_buf_submit(void *raw_data, int size, int rctx, u16 type,
843 		       u64 count, struct pt_regs *regs, void *head,
844 		       struct task_struct *task)
845 {
846 	perf_tp_event(type, count, raw_data, size, regs, head, rctx, task);
847 }
848 
849 #endif
850 
851 #endif /* _LINUX_TRACE_EVENT_H */
852