xref: /openbmc/linux/include/linux/trace_events.h (revision 99bf10e9)
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 union bpf_attr;
19 
20 const char *trace_print_flags_seq(struct trace_seq *p, const char *delim,
21 				  unsigned long flags,
22 				  const struct trace_print_flags *flag_array);
23 
24 const char *trace_print_symbols_seq(struct trace_seq *p, unsigned long val,
25 				    const struct trace_print_flags *symbol_array);
26 
27 #if BITS_PER_LONG == 32
28 const char *trace_print_flags_seq_u64(struct trace_seq *p, const char *delim,
29 		      unsigned long long flags,
30 		      const struct trace_print_flags_u64 *flag_array);
31 
32 const char *trace_print_symbols_seq_u64(struct trace_seq *p,
33 					unsigned long long val,
34 					const struct trace_print_flags_u64
35 								 *symbol_array);
36 #endif
37 
38 const char *trace_print_bitmask_seq(struct trace_seq *p, void *bitmask_ptr,
39 				    unsigned int bitmask_size);
40 
41 const char *trace_print_hex_seq(struct trace_seq *p,
42 				const unsigned char *buf, int len,
43 				bool concatenate);
44 
45 const char *trace_print_array_seq(struct trace_seq *p,
46 				   const void *buf, int count,
47 				   size_t el_size);
48 
49 const char *
50 trace_print_hex_dump_seq(struct trace_seq *p, const char *prefix_str,
51 			 int prefix_type, int rowsize, int groupsize,
52 			 const void *buf, size_t len, bool ascii);
53 
54 struct trace_iterator;
55 struct trace_event;
56 
57 int trace_raw_output_prep(struct trace_iterator *iter,
58 			  struct trace_event *event);
59 extern __printf(2, 3)
60 void trace_event_printf(struct trace_iterator *iter, const char *fmt, ...);
61 
62 /* Used to find the offset and length of dynamic fields in trace events */
63 struct trace_dynamic_info {
64 #ifdef CONFIG_CPU_BIG_ENDIAN
65 	u16	len;
66 	u16	offset;
67 #else
68 	u16	offset;
69 	u16	len;
70 #endif
71 } __packed;
72 
73 /*
74  * The trace entry - the most basic unit of tracing. This is what
75  * is printed in the end as a single line in the trace output, such as:
76  *
77  *     bash-15816 [01]   235.197585: idle_cpu <- irq_enter
78  */
79 struct trace_entry {
80 	unsigned short		type;
81 	unsigned char		flags;
82 	unsigned char		preempt_count;
83 	int			pid;
84 };
85 
86 #define TRACE_EVENT_TYPE_MAX						\
87 	((1 << (sizeof(((struct trace_entry *)0)->type) * 8)) - 1)
88 
89 /*
90  * Trace iterator - used by printout routines who present trace
91  * results to users and which routines might sleep, etc:
92  */
93 struct trace_iterator {
94 	struct trace_array	*tr;
95 	struct tracer		*trace;
96 	struct array_buffer	*array_buffer;
97 	void			*private;
98 	int			cpu_file;
99 	struct mutex		mutex;
100 	struct ring_buffer_iter	**buffer_iter;
101 	unsigned long		iter_flags;
102 	void			*temp;	/* temp holder */
103 	unsigned int		temp_size;
104 	char			*fmt;	/* modified format holder */
105 	unsigned int		fmt_size;
106 	long			wait_index;
107 
108 	/* trace_seq for __print_flags() and __print_symbolic() etc. */
109 	struct trace_seq	tmp_seq;
110 
111 	cpumask_var_t		started;
112 
113 	/* it's true when current open file is snapshot */
114 	bool			snapshot;
115 
116 	/* The below is zeroed out in pipe_read */
117 	struct trace_seq	seq;
118 	struct trace_entry	*ent;
119 	unsigned long		lost_events;
120 	int			leftover;
121 	int			ent_size;
122 	int			cpu;
123 	u64			ts;
124 
125 	loff_t			pos;
126 	long			idx;
127 
128 	/* All new field here will be zeroed out in pipe_read */
129 };
130 
131 enum trace_iter_flags {
132 	TRACE_FILE_LAT_FMT	= 1,
133 	TRACE_FILE_ANNOTATE	= 2,
134 	TRACE_FILE_TIME_IN_NS	= 4,
135 };
136 
137 
138 typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
139 				      int flags, struct trace_event *event);
140 
141 struct trace_event_functions {
142 	trace_print_func	trace;
143 	trace_print_func	raw;
144 	trace_print_func	hex;
145 	trace_print_func	binary;
146 };
147 
148 struct trace_event {
149 	struct hlist_node		node;
150 	int				type;
151 	struct trace_event_functions	*funcs;
152 };
153 
154 extern int register_trace_event(struct trace_event *event);
155 extern int unregister_trace_event(struct trace_event *event);
156 
157 /* Return values for print_line callback */
158 enum print_line_t {
159 	TRACE_TYPE_PARTIAL_LINE	= 0,	/* Retry after flushing the seq */
160 	TRACE_TYPE_HANDLED	= 1,
161 	TRACE_TYPE_UNHANDLED	= 2,	/* Relay to other output functions */
162 	TRACE_TYPE_NO_CONSUME	= 3	/* Handled but ask to not consume */
163 };
164 
165 enum print_line_t trace_handle_return(struct trace_seq *s);
166 
tracing_generic_entry_update(struct trace_entry * entry,unsigned short type,unsigned int trace_ctx)167 static inline void tracing_generic_entry_update(struct trace_entry *entry,
168 						unsigned short type,
169 						unsigned int trace_ctx)
170 {
171 	entry->preempt_count		= trace_ctx & 0xff;
172 	entry->pid			= current->pid;
173 	entry->type			= type;
174 	entry->flags =			trace_ctx >> 16;
175 }
176 
177 unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status);
178 
179 enum trace_flag_type {
180 	TRACE_FLAG_IRQS_OFF		= 0x01,
181 	TRACE_FLAG_IRQS_NOSUPPORT	= 0x02,
182 	TRACE_FLAG_NEED_RESCHED		= 0x04,
183 	TRACE_FLAG_HARDIRQ		= 0x08,
184 	TRACE_FLAG_SOFTIRQ		= 0x10,
185 	TRACE_FLAG_PREEMPT_RESCHED	= 0x20,
186 	TRACE_FLAG_NMI			= 0x40,
187 	TRACE_FLAG_BH_OFF		= 0x80,
188 };
189 
190 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
tracing_gen_ctx_flags(unsigned long irqflags)191 static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
192 {
193 	unsigned int irq_status = irqs_disabled_flags(irqflags) ?
194 		TRACE_FLAG_IRQS_OFF : 0;
195 	return tracing_gen_ctx_irq_test(irq_status);
196 }
tracing_gen_ctx(void)197 static inline unsigned int tracing_gen_ctx(void)
198 {
199 	unsigned long irqflags;
200 
201 	local_save_flags(irqflags);
202 	return tracing_gen_ctx_flags(irqflags);
203 }
204 #else
205 
tracing_gen_ctx_flags(unsigned long irqflags)206 static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
207 {
208 	return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
209 }
tracing_gen_ctx(void)210 static inline unsigned int tracing_gen_ctx(void)
211 {
212 	return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
213 }
214 #endif
215 
tracing_gen_ctx_dec(void)216 static inline unsigned int tracing_gen_ctx_dec(void)
217 {
218 	unsigned int trace_ctx;
219 
220 	trace_ctx = tracing_gen_ctx();
221 	/*
222 	 * Subtract one from the preemption counter if preemption is enabled,
223 	 * see trace_event_buffer_reserve()for details.
224 	 */
225 	if (IS_ENABLED(CONFIG_PREEMPTION))
226 		trace_ctx--;
227 	return trace_ctx;
228 }
229 
230 struct trace_event_file;
231 
232 struct ring_buffer_event *
233 trace_event_buffer_lock_reserve(struct trace_buffer **current_buffer,
234 				struct trace_event_file *trace_file,
235 				int type, unsigned long len,
236 				unsigned int trace_ctx);
237 
238 #define TRACE_RECORD_CMDLINE	BIT(0)
239 #define TRACE_RECORD_TGID	BIT(1)
240 
241 void tracing_record_taskinfo(struct task_struct *task, int flags);
242 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
243 					  struct task_struct *next, int flags);
244 
245 void tracing_record_cmdline(struct task_struct *task);
246 void tracing_record_tgid(struct task_struct *task);
247 
248 int trace_output_call(struct trace_iterator *iter, char *name, char *fmt, ...)
249 	 __printf(3, 4);
250 
251 struct event_filter;
252 
253 enum trace_reg {
254 	TRACE_REG_REGISTER,
255 	TRACE_REG_UNREGISTER,
256 #ifdef CONFIG_PERF_EVENTS
257 	TRACE_REG_PERF_REGISTER,
258 	TRACE_REG_PERF_UNREGISTER,
259 	TRACE_REG_PERF_OPEN,
260 	TRACE_REG_PERF_CLOSE,
261 	/*
262 	 * These (ADD/DEL) use a 'boolean' return value, where 1 (true) means a
263 	 * custom action was taken and the default action is not to be
264 	 * performed.
265 	 */
266 	TRACE_REG_PERF_ADD,
267 	TRACE_REG_PERF_DEL,
268 #endif
269 };
270 
271 struct trace_event_call;
272 
273 #define TRACE_FUNCTION_TYPE ((const char *)~0UL)
274 
275 struct trace_event_fields {
276 	const char *type;
277 	union {
278 		struct {
279 			const char *name;
280 			const int  size;
281 			const int  align;
282 			const int  is_signed;
283 			const int  filter_type;
284 			const int  len;
285 		};
286 		int (*define_fields)(struct trace_event_call *);
287 	};
288 };
289 
290 struct trace_event_class {
291 	const char		*system;
292 	void			*probe;
293 #ifdef CONFIG_PERF_EVENTS
294 	void			*perf_probe;
295 #endif
296 	int			(*reg)(struct trace_event_call *event,
297 				       enum trace_reg type, void *data);
298 	struct trace_event_fields *fields_array;
299 	struct list_head	*(*get_fields)(struct trace_event_call *);
300 	struct list_head	fields;
301 	int			(*raw_init)(struct trace_event_call *);
302 };
303 
304 extern int trace_event_reg(struct trace_event_call *event,
305 			    enum trace_reg type, void *data);
306 
307 struct trace_event_buffer {
308 	struct trace_buffer		*buffer;
309 	struct ring_buffer_event	*event;
310 	struct trace_event_file		*trace_file;
311 	void				*entry;
312 	unsigned int			trace_ctx;
313 	struct pt_regs			*regs;
314 };
315 
316 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
317 				  struct trace_event_file *trace_file,
318 				  unsigned long len);
319 
320 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer);
321 
322 enum {
323 	TRACE_EVENT_FL_FILTERED_BIT,
324 	TRACE_EVENT_FL_CAP_ANY_BIT,
325 	TRACE_EVENT_FL_NO_SET_FILTER_BIT,
326 	TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
327 	TRACE_EVENT_FL_TRACEPOINT_BIT,
328 	TRACE_EVENT_FL_DYNAMIC_BIT,
329 	TRACE_EVENT_FL_KPROBE_BIT,
330 	TRACE_EVENT_FL_UPROBE_BIT,
331 	TRACE_EVENT_FL_EPROBE_BIT,
332 	TRACE_EVENT_FL_FPROBE_BIT,
333 	TRACE_EVENT_FL_CUSTOM_BIT,
334 };
335 
336 /*
337  * Event flags:
338  *  FILTERED	  - The event has a filter attached
339  *  CAP_ANY	  - Any user can enable for perf
340  *  NO_SET_FILTER - Set when filter has error and is to be ignored
341  *  IGNORE_ENABLE - For trace internal events, do not enable with debugfs file
342  *  TRACEPOINT    - Event is a tracepoint
343  *  DYNAMIC       - Event is a dynamic event (created at run time)
344  *  KPROBE        - Event is a kprobe
345  *  UPROBE        - Event is a uprobe
346  *  EPROBE        - Event is an event probe
347  *  FPROBE        - Event is an function probe
348  *  CUSTOM        - Event is a custom event (to be attached to an exsiting tracepoint)
349  *                   This is set when the custom event has not been attached
350  *                   to a tracepoint yet, then it is cleared when it is.
351  */
352 enum {
353 	TRACE_EVENT_FL_FILTERED		= (1 << TRACE_EVENT_FL_FILTERED_BIT),
354 	TRACE_EVENT_FL_CAP_ANY		= (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
355 	TRACE_EVENT_FL_NO_SET_FILTER	= (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
356 	TRACE_EVENT_FL_IGNORE_ENABLE	= (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
357 	TRACE_EVENT_FL_TRACEPOINT	= (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
358 	TRACE_EVENT_FL_DYNAMIC		= (1 << TRACE_EVENT_FL_DYNAMIC_BIT),
359 	TRACE_EVENT_FL_KPROBE		= (1 << TRACE_EVENT_FL_KPROBE_BIT),
360 	TRACE_EVENT_FL_UPROBE		= (1 << TRACE_EVENT_FL_UPROBE_BIT),
361 	TRACE_EVENT_FL_EPROBE		= (1 << TRACE_EVENT_FL_EPROBE_BIT),
362 	TRACE_EVENT_FL_FPROBE		= (1 << TRACE_EVENT_FL_FPROBE_BIT),
363 	TRACE_EVENT_FL_CUSTOM		= (1 << TRACE_EVENT_FL_CUSTOM_BIT),
364 };
365 
366 #define TRACE_EVENT_FL_UKPROBE (TRACE_EVENT_FL_KPROBE | TRACE_EVENT_FL_UPROBE)
367 
368 struct trace_event_call {
369 	struct list_head	list;
370 	struct trace_event_class *class;
371 	union {
372 		char			*name;
373 		/* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
374 		struct tracepoint	*tp;
375 	};
376 	struct trace_event	event;
377 	char			*print_fmt;
378 	struct event_filter	*filter;
379 	/*
380 	 * Static events can disappear with modules,
381 	 * where as dynamic ones need their own ref count.
382 	 */
383 	union {
384 		void				*module;
385 		atomic_t			refcnt;
386 	};
387 	void			*data;
388 
389 	/* See the TRACE_EVENT_FL_* flags above */
390 	int			flags; /* static flags of different events */
391 
392 #ifdef CONFIG_PERF_EVENTS
393 	int				perf_refcount;
394 	struct hlist_head __percpu	*perf_events;
395 	struct bpf_prog_array __rcu	*prog_array;
396 
397 	int	(*perf_perm)(struct trace_event_call *,
398 			     struct perf_event *);
399 #endif
400 };
401 
402 #ifdef CONFIG_DYNAMIC_EVENTS
403 bool trace_event_dyn_try_get_ref(struct trace_event_call *call);
404 void trace_event_dyn_put_ref(struct trace_event_call *call);
405 bool trace_event_dyn_busy(struct trace_event_call *call);
406 #else
trace_event_dyn_try_get_ref(struct trace_event_call * call)407 static inline bool trace_event_dyn_try_get_ref(struct trace_event_call *call)
408 {
409 	/* Without DYNAMIC_EVENTS configured, nothing should be calling this */
410 	return false;
411 }
trace_event_dyn_put_ref(struct trace_event_call * call)412 static inline void trace_event_dyn_put_ref(struct trace_event_call *call)
413 {
414 }
trace_event_dyn_busy(struct trace_event_call * call)415 static inline bool trace_event_dyn_busy(struct trace_event_call *call)
416 {
417 	/* Nothing should call this without DYNAIMIC_EVENTS configured. */
418 	return true;
419 }
420 #endif
421 
trace_event_try_get_ref(struct trace_event_call * call)422 static inline bool trace_event_try_get_ref(struct trace_event_call *call)
423 {
424 	if (call->flags & TRACE_EVENT_FL_DYNAMIC)
425 		return trace_event_dyn_try_get_ref(call);
426 	else
427 		return try_module_get(call->module);
428 }
429 
trace_event_put_ref(struct trace_event_call * call)430 static inline void trace_event_put_ref(struct trace_event_call *call)
431 {
432 	if (call->flags & TRACE_EVENT_FL_DYNAMIC)
433 		trace_event_dyn_put_ref(call);
434 	else
435 		module_put(call->module);
436 }
437 
438 #ifdef CONFIG_PERF_EVENTS
bpf_prog_array_valid(struct trace_event_call * call)439 static inline bool bpf_prog_array_valid(struct trace_event_call *call)
440 {
441 	/*
442 	 * This inline function checks whether call->prog_array
443 	 * is valid or not. The function is called in various places,
444 	 * outside rcu_read_lock/unlock, as a heuristic to speed up execution.
445 	 *
446 	 * If this function returns true, and later call->prog_array
447 	 * becomes false inside rcu_read_lock/unlock region,
448 	 * we bail out then. If this function return false,
449 	 * there is a risk that we might miss a few events if the checking
450 	 * were delayed until inside rcu_read_lock/unlock region and
451 	 * call->prog_array happened to become non-NULL then.
452 	 *
453 	 * Here, READ_ONCE() is used instead of rcu_access_pointer().
454 	 * rcu_access_pointer() requires the actual definition of
455 	 * "struct bpf_prog_array" while READ_ONCE() only needs
456 	 * a declaration of the same type.
457 	 */
458 	return !!READ_ONCE(call->prog_array);
459 }
460 #endif
461 
462 static inline const char *
trace_event_name(struct trace_event_call * call)463 trace_event_name(struct trace_event_call *call)
464 {
465 	if (call->flags & TRACE_EVENT_FL_CUSTOM)
466 		return call->name;
467 	else if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
468 		return call->tp ? call->tp->name : NULL;
469 	else
470 		return call->name;
471 }
472 
473 static inline struct list_head *
trace_get_fields(struct trace_event_call * event_call)474 trace_get_fields(struct trace_event_call *event_call)
475 {
476 	if (!event_call->class->get_fields)
477 		return &event_call->class->fields;
478 	return event_call->class->get_fields(event_call);
479 }
480 
481 struct trace_subsystem_dir;
482 
483 enum {
484 	EVENT_FILE_FL_ENABLED_BIT,
485 	EVENT_FILE_FL_RECORDED_CMD_BIT,
486 	EVENT_FILE_FL_RECORDED_TGID_BIT,
487 	EVENT_FILE_FL_FILTERED_BIT,
488 	EVENT_FILE_FL_NO_SET_FILTER_BIT,
489 	EVENT_FILE_FL_SOFT_MODE_BIT,
490 	EVENT_FILE_FL_SOFT_DISABLED_BIT,
491 	EVENT_FILE_FL_TRIGGER_MODE_BIT,
492 	EVENT_FILE_FL_TRIGGER_COND_BIT,
493 	EVENT_FILE_FL_PID_FILTER_BIT,
494 	EVENT_FILE_FL_WAS_ENABLED_BIT,
495 	EVENT_FILE_FL_FREED_BIT,
496 };
497 
498 extern struct trace_event_file *trace_get_event_file(const char *instance,
499 						     const char *system,
500 						     const char *event);
501 extern void trace_put_event_file(struct trace_event_file *file);
502 
503 #define MAX_DYNEVENT_CMD_LEN	(2048)
504 
505 enum dynevent_type {
506 	DYNEVENT_TYPE_SYNTH = 1,
507 	DYNEVENT_TYPE_KPROBE,
508 	DYNEVENT_TYPE_NONE,
509 };
510 
511 struct dynevent_cmd;
512 
513 typedef int (*dynevent_create_fn_t)(struct dynevent_cmd *cmd);
514 
515 struct dynevent_cmd {
516 	struct seq_buf		seq;
517 	const char		*event_name;
518 	unsigned int		n_fields;
519 	enum dynevent_type	type;
520 	dynevent_create_fn_t	run_command;
521 	void			*private_data;
522 };
523 
524 extern int dynevent_create(struct dynevent_cmd *cmd);
525 
526 extern int synth_event_delete(const char *name);
527 
528 extern void synth_event_cmd_init(struct dynevent_cmd *cmd,
529 				 char *buf, int maxlen);
530 
531 extern int __synth_event_gen_cmd_start(struct dynevent_cmd *cmd,
532 				       const char *name,
533 				       struct module *mod, ...);
534 
535 #define synth_event_gen_cmd_start(cmd, name, mod, ...)	\
536 	__synth_event_gen_cmd_start(cmd, name, mod, ## __VA_ARGS__, NULL)
537 
538 struct synth_field_desc {
539 	const char *type;
540 	const char *name;
541 };
542 
543 extern int synth_event_gen_cmd_array_start(struct dynevent_cmd *cmd,
544 					   const char *name,
545 					   struct module *mod,
546 					   struct synth_field_desc *fields,
547 					   unsigned int n_fields);
548 extern int synth_event_create(const char *name,
549 			      struct synth_field_desc *fields,
550 			      unsigned int n_fields, struct module *mod);
551 
552 extern int synth_event_add_field(struct dynevent_cmd *cmd,
553 				 const char *type,
554 				 const char *name);
555 extern int synth_event_add_field_str(struct dynevent_cmd *cmd,
556 				     const char *type_name);
557 extern int synth_event_add_fields(struct dynevent_cmd *cmd,
558 				  struct synth_field_desc *fields,
559 				  unsigned int n_fields);
560 
561 #define synth_event_gen_cmd_end(cmd)	\
562 	dynevent_create(cmd)
563 
564 struct synth_event;
565 
566 struct synth_event_trace_state {
567 	struct trace_event_buffer fbuffer;
568 	struct synth_trace_event *entry;
569 	struct trace_buffer *buffer;
570 	struct synth_event *event;
571 	unsigned int cur_field;
572 	unsigned int n_u64;
573 	bool disabled;
574 	bool add_next;
575 	bool add_name;
576 };
577 
578 extern int synth_event_trace(struct trace_event_file *file,
579 			     unsigned int n_vals, ...);
580 extern int synth_event_trace_array(struct trace_event_file *file, u64 *vals,
581 				   unsigned int n_vals);
582 extern int synth_event_trace_start(struct trace_event_file *file,
583 				   struct synth_event_trace_state *trace_state);
584 extern int synth_event_add_next_val(u64 val,
585 				    struct synth_event_trace_state *trace_state);
586 extern int synth_event_add_val(const char *field_name, u64 val,
587 			       struct synth_event_trace_state *trace_state);
588 extern int synth_event_trace_end(struct synth_event_trace_state *trace_state);
589 
590 extern int kprobe_event_delete(const char *name);
591 
592 extern void kprobe_event_cmd_init(struct dynevent_cmd *cmd,
593 				  char *buf, int maxlen);
594 
595 #define kprobe_event_gen_cmd_start(cmd, name, loc, ...)			\
596 	__kprobe_event_gen_cmd_start(cmd, false, name, loc, ## __VA_ARGS__, NULL)
597 
598 #define kretprobe_event_gen_cmd_start(cmd, name, loc, ...)		\
599 	__kprobe_event_gen_cmd_start(cmd, true, name, loc, ## __VA_ARGS__, NULL)
600 
601 extern int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd,
602 					bool kretprobe,
603 					const char *name,
604 					const char *loc, ...);
605 
606 #define kprobe_event_add_fields(cmd, ...)	\
607 	__kprobe_event_add_fields(cmd, ## __VA_ARGS__, NULL)
608 
609 #define kprobe_event_add_field(cmd, field)	\
610 	__kprobe_event_add_fields(cmd, field, NULL)
611 
612 extern int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...);
613 
614 #define kprobe_event_gen_cmd_end(cmd)		\
615 	dynevent_create(cmd)
616 
617 #define kretprobe_event_gen_cmd_end(cmd)	\
618 	dynevent_create(cmd)
619 
620 /*
621  * Event file flags:
622  *  ENABLED	  - The event is enabled
623  *  RECORDED_CMD  - The comms should be recorded at sched_switch
624  *  RECORDED_TGID - The tgids should be recorded at sched_switch
625  *  FILTERED	  - The event has a filter attached
626  *  NO_SET_FILTER - Set when filter has error and is to be ignored
627  *  SOFT_MODE     - The event is enabled/disabled by SOFT_DISABLED
628  *  SOFT_DISABLED - When set, do not trace the event (even though its
629  *                   tracepoint may be enabled)
630  *  TRIGGER_MODE  - When set, invoke the triggers associated with the event
631  *  TRIGGER_COND  - When set, one or more triggers has an associated filter
632  *  PID_FILTER    - When set, the event is filtered based on pid
633  *  WAS_ENABLED   - Set when enabled to know to clear trace on module removal
634  *  FREED         - File descriptor is freed, all fields should be considered invalid
635  */
636 enum {
637 	EVENT_FILE_FL_ENABLED		= (1 << EVENT_FILE_FL_ENABLED_BIT),
638 	EVENT_FILE_FL_RECORDED_CMD	= (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
639 	EVENT_FILE_FL_RECORDED_TGID	= (1 << EVENT_FILE_FL_RECORDED_TGID_BIT),
640 	EVENT_FILE_FL_FILTERED		= (1 << EVENT_FILE_FL_FILTERED_BIT),
641 	EVENT_FILE_FL_NO_SET_FILTER	= (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
642 	EVENT_FILE_FL_SOFT_MODE		= (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
643 	EVENT_FILE_FL_SOFT_DISABLED	= (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
644 	EVENT_FILE_FL_TRIGGER_MODE	= (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
645 	EVENT_FILE_FL_TRIGGER_COND	= (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
646 	EVENT_FILE_FL_PID_FILTER	= (1 << EVENT_FILE_FL_PID_FILTER_BIT),
647 	EVENT_FILE_FL_WAS_ENABLED	= (1 << EVENT_FILE_FL_WAS_ENABLED_BIT),
648 	EVENT_FILE_FL_FREED		= (1 << EVENT_FILE_FL_FREED_BIT),
649 };
650 
651 struct trace_event_file {
652 	struct list_head		list;
653 	struct trace_event_call		*event_call;
654 	struct event_filter __rcu	*filter;
655 	struct eventfs_inode		*ei;
656 	struct trace_array		*tr;
657 	struct trace_subsystem_dir	*system;
658 	struct list_head		triggers;
659 
660 	/*
661 	 * 32 bit flags:
662 	 *   bit 0:		enabled
663 	 *   bit 1:		enabled cmd record
664 	 *   bit 2:		enable/disable with the soft disable bit
665 	 *   bit 3:		soft disabled
666 	 *   bit 4:		trigger enabled
667 	 *
668 	 * Note: The bits must be set atomically to prevent races
669 	 * from other writers. Reads of flags do not need to be in
670 	 * sync as they occur in critical sections. But the way flags
671 	 * is currently used, these changes do not affect the code
672 	 * except that when a change is made, it may have a slight
673 	 * delay in propagating the changes to other CPUs due to
674 	 * caching and such. Which is mostly OK ;-)
675 	 */
676 	unsigned long		flags;
677 	atomic_t		ref;	/* ref count for opened files */
678 	atomic_t		sm_ref;	/* soft-mode reference counter */
679 	atomic_t		tm_ref;	/* trigger-mode reference counter */
680 };
681 
682 #define __TRACE_EVENT_FLAGS(name, value)				\
683 	static int __init trace_init_flags_##name(void)			\
684 	{								\
685 		event_##name.flags |= value;				\
686 		return 0;						\
687 	}								\
688 	early_initcall(trace_init_flags_##name);
689 
690 #define __TRACE_EVENT_PERF_PERM(name, expr...)				\
691 	static int perf_perm_##name(struct trace_event_call *tp_event, \
692 				    struct perf_event *p_event)		\
693 	{								\
694 		return ({ expr; });					\
695 	}								\
696 	static int __init trace_init_perf_perm_##name(void)		\
697 	{								\
698 		event_##name.perf_perm = &perf_perm_##name;		\
699 		return 0;						\
700 	}								\
701 	early_initcall(trace_init_perf_perm_##name);
702 
703 #define PERF_MAX_TRACE_SIZE	8192
704 
705 #define MAX_FILTER_STR_VAL	256U	/* Should handle KSYM_SYMBOL_LEN */
706 
707 enum event_trigger_type {
708 	ETT_NONE		= (0),
709 	ETT_TRACE_ONOFF		= (1 << 0),
710 	ETT_SNAPSHOT		= (1 << 1),
711 	ETT_STACKTRACE		= (1 << 2),
712 	ETT_EVENT_ENABLE	= (1 << 3),
713 	ETT_EVENT_HIST		= (1 << 4),
714 	ETT_HIST_ENABLE		= (1 << 5),
715 	ETT_EVENT_EPROBE	= (1 << 6),
716 };
717 
718 extern int filter_match_preds(struct event_filter *filter, void *rec);
719 
720 extern enum event_trigger_type
721 event_triggers_call(struct trace_event_file *file,
722 		    struct trace_buffer *buffer, void *rec,
723 		    struct ring_buffer_event *event);
724 extern void
725 event_triggers_post_call(struct trace_event_file *file,
726 			 enum event_trigger_type tt);
727 
728 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file);
729 
730 bool __trace_trigger_soft_disabled(struct trace_event_file *file);
731 
732 /**
733  * trace_trigger_soft_disabled - do triggers and test if soft disabled
734  * @file: The file pointer of the event to test
735  *
736  * If any triggers without filters are attached to this event, they
737  * will be called here. If the event is soft disabled and has no
738  * triggers that require testing the fields, it will return true,
739  * otherwise false.
740  */
741 static __always_inline bool
trace_trigger_soft_disabled(struct trace_event_file * file)742 trace_trigger_soft_disabled(struct trace_event_file *file)
743 {
744 	unsigned long eflags = file->flags;
745 
746 	if (likely(!(eflags & (EVENT_FILE_FL_TRIGGER_MODE |
747 			       EVENT_FILE_FL_SOFT_DISABLED |
748 			       EVENT_FILE_FL_PID_FILTER))))
749 		return false;
750 
751 	if (likely(eflags & EVENT_FILE_FL_TRIGGER_COND))
752 		return false;
753 
754 	return __trace_trigger_soft_disabled(file);
755 }
756 
757 #ifdef CONFIG_BPF_EVENTS
758 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx);
759 int perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie);
760 void perf_event_detach_bpf_prog(struct perf_event *event);
761 int perf_event_query_prog_array(struct perf_event *event, void __user *info);
762 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
763 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
764 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name);
765 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp);
766 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
767 			    u32 *fd_type, const char **buf,
768 			    u64 *probe_offset, u64 *probe_addr,
769 			    unsigned long *missed);
770 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog);
771 int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog);
772 #else
trace_call_bpf(struct trace_event_call * call,void * ctx)773 static inline unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
774 {
775 	return 1;
776 }
777 
778 static inline int
perf_event_attach_bpf_prog(struct perf_event * event,struct bpf_prog * prog,u64 bpf_cookie)779 perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie)
780 {
781 	return -EOPNOTSUPP;
782 }
783 
perf_event_detach_bpf_prog(struct perf_event * event)784 static inline void perf_event_detach_bpf_prog(struct perf_event *event) { }
785 
786 static inline int
perf_event_query_prog_array(struct perf_event * event,void __user * info)787 perf_event_query_prog_array(struct perf_event *event, void __user *info)
788 {
789 	return -EOPNOTSUPP;
790 }
bpf_probe_register(struct bpf_raw_event_map * btp,struct bpf_prog * p)791 static inline int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *p)
792 {
793 	return -EOPNOTSUPP;
794 }
bpf_probe_unregister(struct bpf_raw_event_map * btp,struct bpf_prog * p)795 static inline int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *p)
796 {
797 	return -EOPNOTSUPP;
798 }
bpf_get_raw_tracepoint(const char * name)799 static inline struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
800 {
801 	return NULL;
802 }
bpf_put_raw_tracepoint(struct bpf_raw_event_map * btp)803 static inline void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
804 {
805 }
bpf_get_perf_event_info(const struct perf_event * event,u32 * prog_id,u32 * fd_type,const char ** buf,u64 * probe_offset,u64 * probe_addr,unsigned long * missed)806 static inline int bpf_get_perf_event_info(const struct perf_event *event,
807 					  u32 *prog_id, u32 *fd_type,
808 					  const char **buf, u64 *probe_offset,
809 					  u64 *probe_addr, unsigned long *missed)
810 {
811 	return -EOPNOTSUPP;
812 }
813 static inline int
bpf_kprobe_multi_link_attach(const union bpf_attr * attr,struct bpf_prog * prog)814 bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
815 {
816 	return -EOPNOTSUPP;
817 }
818 static inline int
bpf_uprobe_multi_link_attach(const union bpf_attr * attr,struct bpf_prog * prog)819 bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
820 {
821 	return -EOPNOTSUPP;
822 }
823 #endif
824 
825 enum {
826 	FILTER_OTHER = 0,
827 	FILTER_STATIC_STRING,
828 	FILTER_DYN_STRING,
829 	FILTER_RDYN_STRING,
830 	FILTER_PTR_STRING,
831 	FILTER_TRACE_FN,
832 	FILTER_CPUMASK,
833 	FILTER_COMM,
834 	FILTER_CPU,
835 	FILTER_STACKTRACE,
836 };
837 
838 extern int trace_event_raw_init(struct trace_event_call *call);
839 extern int trace_define_field(struct trace_event_call *call, const char *type,
840 			      const char *name, int offset, int size,
841 			      int is_signed, int filter_type);
842 extern int trace_add_event_call(struct trace_event_call *call);
843 extern int trace_remove_event_call(struct trace_event_call *call);
844 extern int trace_event_get_offsets(struct trace_event_call *call);
845 
846 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set);
847 int trace_set_clr_event(const char *system, const char *event, int set);
848 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
849 		const char *event, bool enable);
850 /*
851  * The double __builtin_constant_p is because gcc will give us an error
852  * if we try to allocate the static variable to fmt if it is not a
853  * constant. Even with the outer if statement optimizing out.
854  */
855 #define event_trace_printk(ip, fmt, args...)				\
856 do {									\
857 	__trace_printk_check_format(fmt, ##args);			\
858 	tracing_record_cmdline(current);				\
859 	if (__builtin_constant_p(fmt)) {				\
860 		static const char *trace_printk_fmt			\
861 		  __section("__trace_printk_fmt") =			\
862 			__builtin_constant_p(fmt) ? fmt : NULL;		\
863 									\
864 		__trace_bprintk(ip, trace_printk_fmt, ##args);		\
865 	} else								\
866 		__trace_printk(ip, fmt, ##args);			\
867 } while (0)
868 
869 #ifdef CONFIG_PERF_EVENTS
870 struct perf_event;
871 
872 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
873 
874 extern int  perf_trace_init(struct perf_event *event);
875 extern void perf_trace_destroy(struct perf_event *event);
876 extern int  perf_trace_add(struct perf_event *event, int flags);
877 extern void perf_trace_del(struct perf_event *event, int flags);
878 #ifdef CONFIG_KPROBE_EVENTS
879 extern int  perf_kprobe_init(struct perf_event *event, bool is_retprobe);
880 extern void perf_kprobe_destroy(struct perf_event *event);
881 extern int bpf_get_kprobe_info(const struct perf_event *event,
882 			       u32 *fd_type, const char **symbol,
883 			       u64 *probe_offset, u64 *probe_addr,
884 			       unsigned long *missed,
885 			       bool perf_type_tracepoint);
886 #endif
887 #ifdef CONFIG_UPROBE_EVENTS
888 extern int  perf_uprobe_init(struct perf_event *event,
889 			     unsigned long ref_ctr_offset, bool is_retprobe);
890 extern void perf_uprobe_destroy(struct perf_event *event);
891 extern int bpf_get_uprobe_info(const struct perf_event *event,
892 			       u32 *fd_type, const char **filename,
893 			       u64 *probe_offset, u64 *probe_addr,
894 			       bool perf_type_tracepoint);
895 #endif
896 extern int  ftrace_profile_set_filter(struct perf_event *event, int event_id,
897 				     char *filter_str);
898 extern void ftrace_profile_free_filter(struct perf_event *event);
899 void perf_trace_buf_update(void *record, u16 type);
900 void *perf_trace_buf_alloc(int size, struct pt_regs **regs, int *rctxp);
901 
902 int perf_event_set_bpf_prog(struct perf_event *event, struct bpf_prog *prog, u64 bpf_cookie);
903 void perf_event_free_bpf_prog(struct perf_event *event);
904 
905 void bpf_trace_run1(struct bpf_prog *prog, u64 arg1);
906 void bpf_trace_run2(struct bpf_prog *prog, u64 arg1, u64 arg2);
907 void bpf_trace_run3(struct bpf_prog *prog, u64 arg1, u64 arg2,
908 		    u64 arg3);
909 void bpf_trace_run4(struct bpf_prog *prog, u64 arg1, u64 arg2,
910 		    u64 arg3, u64 arg4);
911 void bpf_trace_run5(struct bpf_prog *prog, u64 arg1, u64 arg2,
912 		    u64 arg3, u64 arg4, u64 arg5);
913 void bpf_trace_run6(struct bpf_prog *prog, u64 arg1, u64 arg2,
914 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6);
915 void bpf_trace_run7(struct bpf_prog *prog, u64 arg1, u64 arg2,
916 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7);
917 void bpf_trace_run8(struct bpf_prog *prog, u64 arg1, u64 arg2,
918 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
919 		    u64 arg8);
920 void bpf_trace_run9(struct bpf_prog *prog, u64 arg1, u64 arg2,
921 		    u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
922 		    u64 arg8, u64 arg9);
923 void bpf_trace_run10(struct bpf_prog *prog, u64 arg1, u64 arg2,
924 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
925 		     u64 arg8, u64 arg9, u64 arg10);
926 void bpf_trace_run11(struct bpf_prog *prog, u64 arg1, u64 arg2,
927 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
928 		     u64 arg8, u64 arg9, u64 arg10, u64 arg11);
929 void bpf_trace_run12(struct bpf_prog *prog, u64 arg1, u64 arg2,
930 		     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
931 		     u64 arg8, u64 arg9, u64 arg10, u64 arg11, u64 arg12);
932 void perf_trace_run_bpf_submit(void *raw_data, int size, int rctx,
933 			       struct trace_event_call *call, u64 count,
934 			       struct pt_regs *regs, struct hlist_head *head,
935 			       struct task_struct *task);
936 
937 static inline void
perf_trace_buf_submit(void * raw_data,int size,int rctx,u16 type,u64 count,struct pt_regs * regs,void * head,struct task_struct * task)938 perf_trace_buf_submit(void *raw_data, int size, int rctx, u16 type,
939 		       u64 count, struct pt_regs *regs, void *head,
940 		       struct task_struct *task)
941 {
942 	perf_tp_event(type, count, raw_data, size, regs, head, rctx, task);
943 }
944 
945 #endif
946 
947 #define TRACE_EVENT_STR_MAX	512
948 
949 /*
950  * gcc warns that you can not use a va_list in an inlined
951  * function. But lets me make it into a macro :-/
952  */
953 #define __trace_event_vstr_len(fmt, va)			\
954 ({							\
955 	va_list __ap;					\
956 	int __ret;					\
957 							\
958 	va_copy(__ap, *(va));				\
959 	__ret = vsnprintf(NULL, 0, fmt, __ap) + 1;	\
960 	va_end(__ap);					\
961 							\
962 	min(__ret, TRACE_EVENT_STR_MAX);		\
963 })
964 
965 #endif /* _LINUX_TRACE_EVENT_H */
966 
967 /*
968  * Note: we keep the TRACE_CUSTOM_EVENT outside the include file ifdef protection.
969  *  This is due to the way trace custom events work. If a file includes two
970  *  trace event headers under one "CREATE_CUSTOM_TRACE_EVENTS" the first include
971  *  will override the TRACE_CUSTOM_EVENT and break the second include.
972  */
973 
974 #ifndef TRACE_CUSTOM_EVENT
975 
976 #define DECLARE_CUSTOM_EVENT_CLASS(name, proto, args, tstruct, assign, print)
977 #define DEFINE_CUSTOM_EVENT(template, name, proto, args)
978 #define TRACE_CUSTOM_EVENT(name, proto, args, struct, assign, print)
979 
980 #endif /* ifdef TRACE_CUSTOM_EVENT (see note above) */
981