xref: /linux-6.15/include/linux/trace_events.h (revision be6d9942)
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 extern __printf(2, 3)
59 void trace_event_printf(struct trace_iterator *iter, const char *fmt, ...);
60 
61 /*
62  * The trace entry - the most basic unit of tracing. This is what
63  * is printed in the end as a single line in the trace output, such as:
64  *
65  *     bash-15816 [01]   235.197585: idle_cpu <- irq_enter
66  */
67 struct trace_entry {
68 	unsigned short		type;
69 	unsigned char		flags;
70 	unsigned char		preempt_count;
71 	int			pid;
72 };
73 
74 #define TRACE_EVENT_TYPE_MAX						\
75 	((1 << (sizeof(((struct trace_entry *)0)->type) * 8)) - 1)
76 
77 /*
78  * Trace iterator - used by printout routines who present trace
79  * results to users and which routines might sleep, etc:
80  */
81 struct trace_iterator {
82 	struct trace_array	*tr;
83 	struct tracer		*trace;
84 	struct array_buffer	*array_buffer;
85 	void			*private;
86 	int			cpu_file;
87 	struct mutex		mutex;
88 	struct ring_buffer_iter	**buffer_iter;
89 	unsigned long		iter_flags;
90 	void			*temp;	/* temp holder */
91 	unsigned int		temp_size;
92 	char			*fmt;	/* modified format holder */
93 	unsigned int		fmt_size;
94 
95 	/* trace_seq for __print_flags() and __print_symbolic() etc. */
96 	struct trace_seq	tmp_seq;
97 
98 	cpumask_var_t		started;
99 
100 	/* it's true when current open file is snapshot */
101 	bool			snapshot;
102 
103 	/* The below is zeroed out in pipe_read */
104 	struct trace_seq	seq;
105 	struct trace_entry	*ent;
106 	unsigned long		lost_events;
107 	int			leftover;
108 	int			ent_size;
109 	int			cpu;
110 	u64			ts;
111 
112 	loff_t			pos;
113 	long			idx;
114 
115 	/* All new field here will be zeroed out in pipe_read */
116 };
117 
118 enum trace_iter_flags {
119 	TRACE_FILE_LAT_FMT	= 1,
120 	TRACE_FILE_ANNOTATE	= 2,
121 	TRACE_FILE_TIME_IN_NS	= 4,
122 };
123 
124 
125 typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
126 				      int flags, struct trace_event *event);
127 
128 struct trace_event_functions {
129 	trace_print_func	trace;
130 	trace_print_func	raw;
131 	trace_print_func	hex;
132 	trace_print_func	binary;
133 };
134 
135 struct trace_event {
136 	struct hlist_node		node;
137 	struct list_head		list;
138 	int				type;
139 	struct trace_event_functions	*funcs;
140 };
141 
142 extern int register_trace_event(struct trace_event *event);
143 extern int unregister_trace_event(struct trace_event *event);
144 
145 /* Return values for print_line callback */
146 enum print_line_t {
147 	TRACE_TYPE_PARTIAL_LINE	= 0,	/* Retry after flushing the seq */
148 	TRACE_TYPE_HANDLED	= 1,
149 	TRACE_TYPE_UNHANDLED	= 2,	/* Relay to other output functions */
150 	TRACE_TYPE_NO_CONSUME	= 3	/* Handled but ask to not consume */
151 };
152 
153 enum print_line_t trace_handle_return(struct trace_seq *s);
154 
155 static inline void tracing_generic_entry_update(struct trace_entry *entry,
156 						unsigned short type,
157 						unsigned int trace_ctx)
158 {
159 	entry->preempt_count		= trace_ctx & 0xff;
160 	entry->pid			= current->pid;
161 	entry->type			= type;
162 	entry->flags =			trace_ctx >> 16;
163 }
164 
165 unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status);
166 
167 enum trace_flag_type {
168 	TRACE_FLAG_IRQS_OFF		= 0x01,
169 	TRACE_FLAG_IRQS_NOSUPPORT	= 0x02,
170 	TRACE_FLAG_NEED_RESCHED		= 0x04,
171 	TRACE_FLAG_HARDIRQ		= 0x08,
172 	TRACE_FLAG_SOFTIRQ		= 0x10,
173 	TRACE_FLAG_PREEMPT_RESCHED	= 0x20,
174 	TRACE_FLAG_NMI			= 0x40,
175 };
176 
177 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
178 static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
179 {
180 	unsigned int irq_status = irqs_disabled_flags(irqflags) ?
181 		TRACE_FLAG_IRQS_OFF : 0;
182 	return tracing_gen_ctx_irq_test(irq_status);
183 }
184 static inline unsigned int tracing_gen_ctx(void)
185 {
186 	unsigned long irqflags;
187 
188 	local_save_flags(irqflags);
189 	return tracing_gen_ctx_flags(irqflags);
190 }
191 #else
192 
193 static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
194 {
195 	return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
196 }
197 static inline unsigned int tracing_gen_ctx(void)
198 {
199 	return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
200 }
201 #endif
202 
203 static inline unsigned int tracing_gen_ctx_dec(void)
204 {
205 	unsigned int trace_ctx;
206 
207 	trace_ctx = tracing_gen_ctx();
208 	/*
209 	 * Subtract one from the preeption counter if preemption is enabled,
210 	 * see trace_event_buffer_reserve()for details.
211 	 */
212 	if (IS_ENABLED(CONFIG_PREEMPTION))
213 		trace_ctx--;
214 	return trace_ctx;
215 }
216 
217 struct trace_event_file;
218 
219 struct ring_buffer_event *
220 trace_event_buffer_lock_reserve(struct trace_buffer **current_buffer,
221 				struct trace_event_file *trace_file,
222 				int type, unsigned long len,
223 				unsigned int trace_ctx);
224 
225 #define TRACE_RECORD_CMDLINE	BIT(0)
226 #define TRACE_RECORD_TGID	BIT(1)
227 
228 void tracing_record_taskinfo(struct task_struct *task, int flags);
229 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
230 					  struct task_struct *next, int flags);
231 
232 void tracing_record_cmdline(struct task_struct *task);
233 void tracing_record_tgid(struct task_struct *task);
234 
235 int trace_output_call(struct trace_iterator *iter, char *name, char *fmt, ...);
236 
237 struct event_filter;
238 
239 enum trace_reg {
240 	TRACE_REG_REGISTER,
241 	TRACE_REG_UNREGISTER,
242 #ifdef CONFIG_PERF_EVENTS
243 	TRACE_REG_PERF_REGISTER,
244 	TRACE_REG_PERF_UNREGISTER,
245 	TRACE_REG_PERF_OPEN,
246 	TRACE_REG_PERF_CLOSE,
247 	/*
248 	 * These (ADD/DEL) use a 'boolean' return value, where 1 (true) means a
249 	 * custom action was taken and the default action is not to be
250 	 * performed.
251 	 */
252 	TRACE_REG_PERF_ADD,
253 	TRACE_REG_PERF_DEL,
254 #endif
255 };
256 
257 struct trace_event_call;
258 
259 #define TRACE_FUNCTION_TYPE ((const char *)~0UL)
260 
261 struct trace_event_fields {
262 	const char *type;
263 	union {
264 		struct {
265 			const char *name;
266 			const int  size;
267 			const int  align;
268 			const int  is_signed;
269 			const int  filter_type;
270 		};
271 		int (*define_fields)(struct trace_event_call *);
272 	};
273 };
274 
275 struct trace_event_class {
276 	const char		*system;
277 	void			*probe;
278 #ifdef CONFIG_PERF_EVENTS
279 	void			*perf_probe;
280 #endif
281 	int			(*reg)(struct trace_event_call *event,
282 				       enum trace_reg type, void *data);
283 	struct trace_event_fields *fields_array;
284 	struct list_head	*(*get_fields)(struct trace_event_call *);
285 	struct list_head	fields;
286 	int			(*raw_init)(struct trace_event_call *);
287 };
288 
289 extern int trace_event_reg(struct trace_event_call *event,
290 			    enum trace_reg type, void *data);
291 
292 struct trace_event_buffer {
293 	struct trace_buffer		*buffer;
294 	struct ring_buffer_event	*event;
295 	struct trace_event_file		*trace_file;
296 	void				*entry;
297 	unsigned int			trace_ctx;
298 	struct pt_regs			*regs;
299 };
300 
301 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
302 				  struct trace_event_file *trace_file,
303 				  unsigned long len);
304 
305 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer);
306 
307 enum {
308 	TRACE_EVENT_FL_FILTERED_BIT,
309 	TRACE_EVENT_FL_CAP_ANY_BIT,
310 	TRACE_EVENT_FL_NO_SET_FILTER_BIT,
311 	TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
312 	TRACE_EVENT_FL_TRACEPOINT_BIT,
313 	TRACE_EVENT_FL_KPROBE_BIT,
314 	TRACE_EVENT_FL_UPROBE_BIT,
315 };
316 
317 /*
318  * Event flags:
319  *  FILTERED	  - The event has a filter attached
320  *  CAP_ANY	  - Any user can enable for perf
321  *  NO_SET_FILTER - Set when filter has error and is to be ignored
322  *  IGNORE_ENABLE - For trace internal events, do not enable with debugfs file
323  *  TRACEPOINT    - Event is a tracepoint
324  *  KPROBE        - Event is a kprobe
325  *  UPROBE        - Event is a uprobe
326  */
327 enum {
328 	TRACE_EVENT_FL_FILTERED		= (1 << TRACE_EVENT_FL_FILTERED_BIT),
329 	TRACE_EVENT_FL_CAP_ANY		= (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
330 	TRACE_EVENT_FL_NO_SET_FILTER	= (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
331 	TRACE_EVENT_FL_IGNORE_ENABLE	= (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
332 	TRACE_EVENT_FL_TRACEPOINT	= (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
333 	TRACE_EVENT_FL_KPROBE		= (1 << TRACE_EVENT_FL_KPROBE_BIT),
334 	TRACE_EVENT_FL_UPROBE		= (1 << TRACE_EVENT_FL_UPROBE_BIT),
335 };
336 
337 #define TRACE_EVENT_FL_UKPROBE (TRACE_EVENT_FL_KPROBE | TRACE_EVENT_FL_UPROBE)
338 
339 struct trace_event_call {
340 	struct list_head	list;
341 	struct trace_event_class *class;
342 	union {
343 		char			*name;
344 		/* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
345 		struct tracepoint	*tp;
346 	};
347 	struct trace_event	event;
348 	char			*print_fmt;
349 	struct event_filter	*filter;
350 	void			*mod;
351 	void			*data;
352 	/*
353 	 *   bit 0:		filter_active
354 	 *   bit 1:		allow trace by non root (cap any)
355 	 *   bit 2:		failed to apply filter
356 	 *   bit 3:		trace internal event (do not enable)
357 	 *   bit 4:		Event was enabled by module
358 	 *   bit 5:		use call filter rather than file filter
359 	 *   bit 6:		Event is a tracepoint
360 	 */
361 	int			flags; /* static flags of different events */
362 
363 #ifdef CONFIG_PERF_EVENTS
364 	int				perf_refcount;
365 	struct hlist_head __percpu	*perf_events;
366 	struct bpf_prog_array __rcu	*prog_array;
367 
368 	int	(*perf_perm)(struct trace_event_call *,
369 			     struct perf_event *);
370 #endif
371 };
372 
373 #ifdef CONFIG_PERF_EVENTS
374 static inline bool bpf_prog_array_valid(struct trace_event_call *call)
375 {
376 	/*
377 	 * This inline function checks whether call->prog_array
378 	 * is valid or not. The function is called in various places,
379 	 * outside rcu_read_lock/unlock, as a heuristic to speed up execution.
380 	 *
381 	 * If this function returns true, and later call->prog_array
382 	 * becomes false inside rcu_read_lock/unlock region,
383 	 * we bail out then. If this function return false,
384 	 * there is a risk that we might miss a few events if the checking
385 	 * were delayed until inside rcu_read_lock/unlock region and
386 	 * call->prog_array happened to become non-NULL then.
387 	 *
388 	 * Here, READ_ONCE() is used instead of rcu_access_pointer().
389 	 * rcu_access_pointer() requires the actual definition of
390 	 * "struct bpf_prog_array" while READ_ONCE() only needs
391 	 * a declaration of the same type.
392 	 */
393 	return !!READ_ONCE(call->prog_array);
394 }
395 #endif
396 
397 static inline const char *
398 trace_event_name(struct trace_event_call *call)
399 {
400 	if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
401 		return call->tp ? call->tp->name : NULL;
402 	else
403 		return call->name;
404 }
405 
406 static inline struct list_head *
407 trace_get_fields(struct trace_event_call *event_call)
408 {
409 	if (!event_call->class->get_fields)
410 		return &event_call->class->fields;
411 	return event_call->class->get_fields(event_call);
412 }
413 
414 struct trace_array;
415 struct trace_subsystem_dir;
416 
417 enum {
418 	EVENT_FILE_FL_ENABLED_BIT,
419 	EVENT_FILE_FL_RECORDED_CMD_BIT,
420 	EVENT_FILE_FL_RECORDED_TGID_BIT,
421 	EVENT_FILE_FL_FILTERED_BIT,
422 	EVENT_FILE_FL_NO_SET_FILTER_BIT,
423 	EVENT_FILE_FL_SOFT_MODE_BIT,
424 	EVENT_FILE_FL_SOFT_DISABLED_BIT,
425 	EVENT_FILE_FL_TRIGGER_MODE_BIT,
426 	EVENT_FILE_FL_TRIGGER_COND_BIT,
427 	EVENT_FILE_FL_PID_FILTER_BIT,
428 	EVENT_FILE_FL_WAS_ENABLED_BIT,
429 };
430 
431 extern struct trace_event_file *trace_get_event_file(const char *instance,
432 						     const char *system,
433 						     const char *event);
434 extern void trace_put_event_file(struct trace_event_file *file);
435 
436 #define MAX_DYNEVENT_CMD_LEN	(2048)
437 
438 enum dynevent_type {
439 	DYNEVENT_TYPE_SYNTH = 1,
440 	DYNEVENT_TYPE_KPROBE,
441 	DYNEVENT_TYPE_NONE,
442 };
443 
444 struct dynevent_cmd;
445 
446 typedef int (*dynevent_create_fn_t)(struct dynevent_cmd *cmd);
447 
448 struct dynevent_cmd {
449 	struct seq_buf		seq;
450 	const char		*event_name;
451 	unsigned int		n_fields;
452 	enum dynevent_type	type;
453 	dynevent_create_fn_t	run_command;
454 	void			*private_data;
455 };
456 
457 extern int dynevent_create(struct dynevent_cmd *cmd);
458 
459 extern int synth_event_delete(const char *name);
460 
461 extern void synth_event_cmd_init(struct dynevent_cmd *cmd,
462 				 char *buf, int maxlen);
463 
464 extern int __synth_event_gen_cmd_start(struct dynevent_cmd *cmd,
465 				       const char *name,
466 				       struct module *mod, ...);
467 
468 #define synth_event_gen_cmd_start(cmd, name, mod, ...)	\
469 	__synth_event_gen_cmd_start(cmd, name, mod, ## __VA_ARGS__, NULL)
470 
471 struct synth_field_desc {
472 	const char *type;
473 	const char *name;
474 };
475 
476 extern int synth_event_gen_cmd_array_start(struct dynevent_cmd *cmd,
477 					   const char *name,
478 					   struct module *mod,
479 					   struct synth_field_desc *fields,
480 					   unsigned int n_fields);
481 extern int synth_event_create(const char *name,
482 			      struct synth_field_desc *fields,
483 			      unsigned int n_fields, struct module *mod);
484 
485 extern int synth_event_add_field(struct dynevent_cmd *cmd,
486 				 const char *type,
487 				 const char *name);
488 extern int synth_event_add_field_str(struct dynevent_cmd *cmd,
489 				     const char *type_name);
490 extern int synth_event_add_fields(struct dynevent_cmd *cmd,
491 				  struct synth_field_desc *fields,
492 				  unsigned int n_fields);
493 
494 #define synth_event_gen_cmd_end(cmd)	\
495 	dynevent_create(cmd)
496 
497 struct synth_event;
498 
499 struct synth_event_trace_state {
500 	struct trace_event_buffer fbuffer;
501 	struct synth_trace_event *entry;
502 	struct trace_buffer *buffer;
503 	struct synth_event *event;
504 	unsigned int cur_field;
505 	unsigned int n_u64;
506 	bool disabled;
507 	bool add_next;
508 	bool add_name;
509 };
510 
511 extern int synth_event_trace(struct trace_event_file *file,
512 			     unsigned int n_vals, ...);
513 extern int synth_event_trace_array(struct trace_event_file *file, u64 *vals,
514 				   unsigned int n_vals);
515 extern int synth_event_trace_start(struct trace_event_file *file,
516 				   struct synth_event_trace_state *trace_state);
517 extern int synth_event_add_next_val(u64 val,
518 				    struct synth_event_trace_state *trace_state);
519 extern int synth_event_add_val(const char *field_name, u64 val,
520 			       struct synth_event_trace_state *trace_state);
521 extern int synth_event_trace_end(struct synth_event_trace_state *trace_state);
522 
523 extern int kprobe_event_delete(const char *name);
524 
525 extern void kprobe_event_cmd_init(struct dynevent_cmd *cmd,
526 				  char *buf, int maxlen);
527 
528 #define kprobe_event_gen_cmd_start(cmd, name, loc, ...)			\
529 	__kprobe_event_gen_cmd_start(cmd, false, name, loc, ## __VA_ARGS__, NULL)
530 
531 #define kretprobe_event_gen_cmd_start(cmd, name, loc, ...)		\
532 	__kprobe_event_gen_cmd_start(cmd, true, name, loc, ## __VA_ARGS__, NULL)
533 
534 extern int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd,
535 					bool kretprobe,
536 					const char *name,
537 					const char *loc, ...);
538 
539 #define kprobe_event_add_fields(cmd, ...)	\
540 	__kprobe_event_add_fields(cmd, ## __VA_ARGS__, NULL)
541 
542 #define kprobe_event_add_field(cmd, field)	\
543 	__kprobe_event_add_fields(cmd, field, NULL)
544 
545 extern int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...);
546 
547 #define kprobe_event_gen_cmd_end(cmd)		\
548 	dynevent_create(cmd)
549 
550 #define kretprobe_event_gen_cmd_end(cmd)	\
551 	dynevent_create(cmd)
552 
553 /*
554  * Event file flags:
555  *  ENABLED	  - The event is enabled
556  *  RECORDED_CMD  - The comms should be recorded at sched_switch
557  *  RECORDED_TGID - The tgids should be recorded at sched_switch
558  *  FILTERED	  - The event has a filter attached
559  *  NO_SET_FILTER - Set when filter has error and is to be ignored
560  *  SOFT_MODE     - The event is enabled/disabled by SOFT_DISABLED
561  *  SOFT_DISABLED - When set, do not trace the event (even though its
562  *                   tracepoint may be enabled)
563  *  TRIGGER_MODE  - When set, invoke the triggers associated with the event
564  *  TRIGGER_COND  - When set, one or more triggers has an associated filter
565  *  PID_FILTER    - When set, the event is filtered based on pid
566  *  WAS_ENABLED   - Set when enabled to know to clear trace on module removal
567  */
568 enum {
569 	EVENT_FILE_FL_ENABLED		= (1 << EVENT_FILE_FL_ENABLED_BIT),
570 	EVENT_FILE_FL_RECORDED_CMD	= (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
571 	EVENT_FILE_FL_RECORDED_TGID	= (1 << EVENT_FILE_FL_RECORDED_TGID_BIT),
572 	EVENT_FILE_FL_FILTERED		= (1 << EVENT_FILE_FL_FILTERED_BIT),
573 	EVENT_FILE_FL_NO_SET_FILTER	= (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
574 	EVENT_FILE_FL_SOFT_MODE		= (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
575 	EVENT_FILE_FL_SOFT_DISABLED	= (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
576 	EVENT_FILE_FL_TRIGGER_MODE	= (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
577 	EVENT_FILE_FL_TRIGGER_COND	= (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
578 	EVENT_FILE_FL_PID_FILTER	= (1 << EVENT_FILE_FL_PID_FILTER_BIT),
579 	EVENT_FILE_FL_WAS_ENABLED	= (1 << EVENT_FILE_FL_WAS_ENABLED_BIT),
580 };
581 
582 struct trace_event_file {
583 	struct list_head		list;
584 	struct trace_event_call		*event_call;
585 	struct event_filter __rcu	*filter;
586 	struct dentry			*dir;
587 	struct trace_array		*tr;
588 	struct trace_subsystem_dir	*system;
589 	struct list_head		triggers;
590 
591 	/*
592 	 * 32 bit flags:
593 	 *   bit 0:		enabled
594 	 *   bit 1:		enabled cmd record
595 	 *   bit 2:		enable/disable with the soft disable bit
596 	 *   bit 3:		soft disabled
597 	 *   bit 4:		trigger enabled
598 	 *
599 	 * Note: The bits must be set atomically to prevent races
600 	 * from other writers. Reads of flags do not need to be in
601 	 * sync as they occur in critical sections. But the way flags
602 	 * is currently used, these changes do not affect the code
603 	 * except that when a change is made, it may have a slight
604 	 * delay in propagating the changes to other CPUs due to
605 	 * caching and such. Which is mostly OK ;-)
606 	 */
607 	unsigned long		flags;
608 	atomic_t		sm_ref;	/* soft-mode reference counter */
609 	atomic_t		tm_ref;	/* trigger-mode reference counter */
610 };
611 
612 #define __TRACE_EVENT_FLAGS(name, value)				\
613 	static int __init trace_init_flags_##name(void)			\
614 	{								\
615 		event_##name.flags |= value;				\
616 		return 0;						\
617 	}								\
618 	early_initcall(trace_init_flags_##name);
619 
620 #define __TRACE_EVENT_PERF_PERM(name, expr...)				\
621 	static int perf_perm_##name(struct trace_event_call *tp_event, \
622 				    struct perf_event *p_event)		\
623 	{								\
624 		return ({ expr; });					\
625 	}								\
626 	static int __init trace_init_perf_perm_##name(void)		\
627 	{								\
628 		event_##name.perf_perm = &perf_perm_##name;		\
629 		return 0;						\
630 	}								\
631 	early_initcall(trace_init_perf_perm_##name);
632 
633 #define PERF_MAX_TRACE_SIZE	2048
634 
635 #define MAX_FILTER_STR_VAL	256	/* Should handle KSYM_SYMBOL_LEN */
636 
637 enum event_trigger_type {
638 	ETT_NONE		= (0),
639 	ETT_TRACE_ONOFF		= (1 << 0),
640 	ETT_SNAPSHOT		= (1 << 1),
641 	ETT_STACKTRACE		= (1 << 2),
642 	ETT_EVENT_ENABLE	= (1 << 3),
643 	ETT_EVENT_HIST		= (1 << 4),
644 	ETT_HIST_ENABLE		= (1 << 5),
645 };
646 
647 extern int filter_match_preds(struct event_filter *filter, void *rec);
648 
649 extern enum event_trigger_type
650 event_triggers_call(struct trace_event_file *file, void *rec,
651 		    struct ring_buffer_event *event);
652 extern void
653 event_triggers_post_call(struct trace_event_file *file,
654 			 enum event_trigger_type tt);
655 
656 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file);
657 
658 /**
659  * trace_trigger_soft_disabled - do triggers and test if soft disabled
660  * @file: The file pointer of the event to test
661  *
662  * If any triggers without filters are attached to this event, they
663  * will be called here. If the event is soft disabled and has no
664  * triggers that require testing the fields, it will return true,
665  * otherwise false.
666  */
667 static inline bool
668 trace_trigger_soft_disabled(struct trace_event_file *file)
669 {
670 	unsigned long eflags = file->flags;
671 
672 	if (!(eflags & EVENT_FILE_FL_TRIGGER_COND)) {
673 		if (eflags & EVENT_FILE_FL_TRIGGER_MODE)
674 			event_triggers_call(file, NULL, NULL);
675 		if (eflags & EVENT_FILE_FL_SOFT_DISABLED)
676 			return true;
677 		if (eflags & EVENT_FILE_FL_PID_FILTER)
678 			return trace_event_ignore_this_pid(file);
679 	}
680 	return false;
681 }
682 
683 #ifdef CONFIG_BPF_EVENTS
684 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx);
685 int perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog);
686 void perf_event_detach_bpf_prog(struct perf_event *event);
687 int perf_event_query_prog_array(struct perf_event *event, void __user *info);
688 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
689 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
690 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name);
691 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp);
692 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
693 			    u32 *fd_type, const char **buf,
694 			    u64 *probe_offset, u64 *probe_addr);
695 #else
696 static inline unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
697 {
698 	return 1;
699 }
700 
701 static inline int
702 perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog)
703 {
704 	return -EOPNOTSUPP;
705 }
706 
707 static inline void perf_event_detach_bpf_prog(struct perf_event *event) { }
708 
709 static inline int
710 perf_event_query_prog_array(struct perf_event *event, void __user *info)
711 {
712 	return -EOPNOTSUPP;
713 }
714 static inline int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *p)
715 {
716 	return -EOPNOTSUPP;
717 }
718 static inline int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *p)
719 {
720 	return -EOPNOTSUPP;
721 }
722 static inline struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
723 {
724 	return NULL;
725 }
726 static inline void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
727 {
728 }
729 static inline int bpf_get_perf_event_info(const struct perf_event *event,
730 					  u32 *prog_id, u32 *fd_type,
731 					  const char **buf, u64 *probe_offset,
732 					  u64 *probe_addr)
733 {
734 	return -EOPNOTSUPP;
735 }
736 #endif
737 
738 enum {
739 	FILTER_OTHER = 0,
740 	FILTER_STATIC_STRING,
741 	FILTER_DYN_STRING,
742 	FILTER_PTR_STRING,
743 	FILTER_TRACE_FN,
744 	FILTER_COMM,
745 	FILTER_CPU,
746 };
747 
748 extern int trace_event_raw_init(struct trace_event_call *call);
749 extern int trace_define_field(struct trace_event_call *call, const char *type,
750 			      const char *name, int offset, int size,
751 			      int is_signed, int filter_type);
752 extern int trace_add_event_call(struct trace_event_call *call);
753 extern int trace_remove_event_call(struct trace_event_call *call);
754 extern int trace_event_get_offsets(struct trace_event_call *call);
755 
756 #define is_signed_type(type)	(((type)(-1)) < (type)1)
757 
758 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set);
759 int trace_set_clr_event(const char *system, const char *event, int set);
760 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
761 		const char *event, bool enable);
762 /*
763  * The double __builtin_constant_p is because gcc will give us an error
764  * if we try to allocate the static variable to fmt if it is not a
765  * constant. Even with the outer if statement optimizing out.
766  */
767 #define event_trace_printk(ip, fmt, args...)				\
768 do {									\
769 	__trace_printk_check_format(fmt, ##args);			\
770 	tracing_record_cmdline(current);				\
771 	if (__builtin_constant_p(fmt)) {				\
772 		static const char *trace_printk_fmt			\
773 		  __section("__trace_printk_fmt") =			\
774 			__builtin_constant_p(fmt) ? fmt : NULL;		\
775 									\
776 		__trace_bprintk(ip, trace_printk_fmt, ##args);		\
777 	} else								\
778 		__trace_printk(ip, fmt, ##args);			\
779 } while (0)
780 
781 #ifdef CONFIG_PERF_EVENTS
782 struct perf_event;
783 
784 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
785 DECLARE_PER_CPU(int, bpf_kprobe_override);
786 
787 extern int  perf_trace_init(struct perf_event *event);
788 extern void perf_trace_destroy(struct perf_event *event);
789 extern int  perf_trace_add(struct perf_event *event, int flags);
790 extern void perf_trace_del(struct perf_event *event, int flags);
791 #ifdef CONFIG_KPROBE_EVENTS
792 extern int  perf_kprobe_init(struct perf_event *event, bool is_retprobe);
793 extern void perf_kprobe_destroy(struct perf_event *event);
794 extern int bpf_get_kprobe_info(const struct perf_event *event,
795 			       u32 *fd_type, const char **symbol,
796 			       u64 *probe_offset, u64 *probe_addr,
797 			       bool perf_type_tracepoint);
798 #endif
799 #ifdef CONFIG_UPROBE_EVENTS
800 extern int  perf_uprobe_init(struct perf_event *event,
801 			     unsigned long ref_ctr_offset, bool is_retprobe);
802 extern void perf_uprobe_destroy(struct perf_event *event);
803 extern int bpf_get_uprobe_info(const struct perf_event *event,
804 			       u32 *fd_type, const char **filename,
805 			       u64 *probe_offset, bool perf_type_tracepoint);
806 #endif
807 extern int  ftrace_profile_set_filter(struct perf_event *event, int event_id,
808 				     char *filter_str);
809 extern void ftrace_profile_free_filter(struct perf_event *event);
810 void perf_trace_buf_update(void *record, u16 type);
811 void *perf_trace_buf_alloc(int size, struct pt_regs **regs, int *rctxp);
812 
813 void bpf_trace_run1(struct bpf_prog *prog, u64 arg1);
814 void bpf_trace_run2(struct bpf_prog *prog, u64 arg1, u64 arg2);
815 void bpf_trace_run3(struct bpf_prog *prog, u64 arg1, u64 arg2,
816 		    u64 arg3);
817 void bpf_trace_run4(struct bpf_prog *prog, u64 arg1, u64 arg2,
818 		    u64 arg3, u64 arg4);
819 void bpf_trace_run5(struct bpf_prog *prog, u64 arg1, u64 arg2,
820 		    u64 arg3, u64 arg4, u64 arg5);
821 void bpf_trace_run6(struct bpf_prog *prog, u64 arg1, u64 arg2,
822 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6);
823 void bpf_trace_run7(struct bpf_prog *prog, u64 arg1, u64 arg2,
824 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7);
825 void bpf_trace_run8(struct bpf_prog *prog, u64 arg1, u64 arg2,
826 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
827 		    u64 arg8);
828 void bpf_trace_run9(struct bpf_prog *prog, u64 arg1, u64 arg2,
829 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
830 		    u64 arg8, u64 arg9);
831 void bpf_trace_run10(struct bpf_prog *prog, u64 arg1, u64 arg2,
832 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
833 		     u64 arg8, u64 arg9, u64 arg10);
834 void bpf_trace_run11(struct bpf_prog *prog, u64 arg1, u64 arg2,
835 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
836 		     u64 arg8, u64 arg9, u64 arg10, u64 arg11);
837 void bpf_trace_run12(struct bpf_prog *prog, u64 arg1, u64 arg2,
838 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
839 		     u64 arg8, u64 arg9, u64 arg10, u64 arg11, u64 arg12);
840 void perf_trace_run_bpf_submit(void *raw_data, int size, int rctx,
841 			       struct trace_event_call *call, u64 count,
842 			       struct pt_regs *regs, struct hlist_head *head,
843 			       struct task_struct *task);
844 
845 static inline void
846 perf_trace_buf_submit(void *raw_data, int size, int rctx, u16 type,
847 		       u64 count, struct pt_regs *regs, void *head,
848 		       struct task_struct *task)
849 {
850 	perf_tp_event(type, count, raw_data, size, regs, head, rctx, task);
851 }
852 
853 #endif
854 
855 #endif /* _LINUX_TRACE_EVENT_H */
856