xref: /linux-6.15/include/linux/tracepoint.h (revision 605bfaee)
1 #ifndef _LINUX_TRACEPOINT_H
2 #define _LINUX_TRACEPOINT_H
3 
4 /*
5  * Kernel Tracepoint API.
6  *
7  * See Documentation/trace/tracepoints.txt.
8  *
9  * (C) Copyright 2008 Mathieu Desnoyers <[email protected]>
10  *
11  * Heavily inspired from the Linux Kernel Markers.
12  *
13  * This file is released under the GPLv2.
14  * See the file COPYING for more details.
15  */
16 
17 #include <linux/types.h>
18 #include <linux/rcupdate.h>
19 
20 struct module;
21 struct tracepoint;
22 
23 struct tracepoint {
24 	const char *name;		/* Tracepoint name */
25 	int state;			/* State. */
26 	void (*regfunc)(void);
27 	void (*unregfunc)(void);
28 	void **funcs;
29 } __attribute__((aligned(32)));		/*
30 					 * Aligned on 32 bytes because it is
31 					 * globally visible and gcc happily
32 					 * align these on the structure size.
33 					 * Keep in sync with vmlinux.lds.h.
34 					 */
35 
36 #ifndef DECLARE_TRACE
37 
38 #define TP_PROTO(args...)	args
39 #define TP_ARGS(args...)	args
40 
41 #ifdef CONFIG_TRACEPOINTS
42 
43 /*
44  * it_func[0] is never NULL because there is at least one element in the array
45  * when the array itself is non NULL.
46  */
47 #define __DO_TRACE(tp, proto, args)					\
48 	do {								\
49 		void **it_func;						\
50 									\
51 		rcu_read_lock_sched_notrace();				\
52 		it_func = rcu_dereference((tp)->funcs);			\
53 		if (it_func) {						\
54 			do {						\
55 				((void(*)(proto))(*it_func))(args);	\
56 			} while (*(++it_func));				\
57 		}							\
58 		rcu_read_unlock_sched_notrace();			\
59 	} while (0)
60 
61 /*
62  * Make sure the alignment of the structure in the __tracepoints section will
63  * not add unwanted padding between the beginning of the section and the
64  * structure. Force alignment to the same alignment as the section start.
65  */
66 #define DECLARE_TRACE(name, proto, args)				\
67 	extern struct tracepoint __tracepoint_##name;			\
68 	static inline void trace_##name(proto)				\
69 	{								\
70 		if (unlikely(__tracepoint_##name.state))		\
71 			__DO_TRACE(&__tracepoint_##name,		\
72 				TP_PROTO(proto), TP_ARGS(args));	\
73 	}								\
74 	static inline int register_trace_##name(void (*probe)(proto))	\
75 	{								\
76 		return tracepoint_probe_register(#name, (void *)probe);	\
77 	}								\
78 	static inline int unregister_trace_##name(void (*probe)(proto))	\
79 	{								\
80 		return tracepoint_probe_unregister(#name, (void *)probe);\
81 	}
82 
83 
84 #define DEFINE_TRACE_FN(name, reg, unreg)				\
85 	static const char __tpstrtab_##name[]				\
86 	__attribute__((section("__tracepoints_strings"))) = #name;	\
87 	struct tracepoint __tracepoint_##name				\
88 	__attribute__((section("__tracepoints"), aligned(32))) =	\
89 		{ __tpstrtab_##name, 0, reg, unreg, NULL }
90 
91 #define DEFINE_TRACE(name)						\
92 	DEFINE_TRACE_FN(name, NULL, NULL);
93 
94 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)				\
95 	EXPORT_SYMBOL_GPL(__tracepoint_##name)
96 #define EXPORT_TRACEPOINT_SYMBOL(name)					\
97 	EXPORT_SYMBOL(__tracepoint_##name)
98 
99 extern void tracepoint_update_probe_range(struct tracepoint *begin,
100 	struct tracepoint *end);
101 
102 #else /* !CONFIG_TRACEPOINTS */
103 #define DECLARE_TRACE(name, proto, args)				\
104 	static inline void _do_trace_##name(struct tracepoint *tp, proto) \
105 	{ }								\
106 	static inline void trace_##name(proto)				\
107 	{ }								\
108 	static inline int register_trace_##name(void (*probe)(proto))	\
109 	{								\
110 		return -ENOSYS;						\
111 	}								\
112 	static inline int unregister_trace_##name(void (*probe)(proto))	\
113 	{								\
114 		return -ENOSYS;						\
115 	}
116 
117 #define DEFINE_TRACE_FN(name, reg, unreg)
118 #define DEFINE_TRACE(name)
119 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
120 #define EXPORT_TRACEPOINT_SYMBOL(name)
121 
122 static inline void tracepoint_update_probe_range(struct tracepoint *begin,
123 	struct tracepoint *end)
124 { }
125 #endif /* CONFIG_TRACEPOINTS */
126 #endif /* DECLARE_TRACE */
127 
128 /*
129  * Connect a probe to a tracepoint.
130  * Internal API, should not be used directly.
131  */
132 extern int tracepoint_probe_register(const char *name, void *probe);
133 
134 /*
135  * Disconnect a probe from a tracepoint.
136  * Internal API, should not be used directly.
137  */
138 extern int tracepoint_probe_unregister(const char *name, void *probe);
139 
140 extern int tracepoint_probe_register_noupdate(const char *name, void *probe);
141 extern int tracepoint_probe_unregister_noupdate(const char *name, void *probe);
142 extern void tracepoint_probe_update_all(void);
143 
144 struct tracepoint_iter {
145 	struct module *module;
146 	struct tracepoint *tracepoint;
147 };
148 
149 extern void tracepoint_iter_start(struct tracepoint_iter *iter);
150 extern void tracepoint_iter_next(struct tracepoint_iter *iter);
151 extern void tracepoint_iter_stop(struct tracepoint_iter *iter);
152 extern void tracepoint_iter_reset(struct tracepoint_iter *iter);
153 extern int tracepoint_get_iter_range(struct tracepoint **tracepoint,
154 	struct tracepoint *begin, struct tracepoint *end);
155 
156 /*
157  * tracepoint_synchronize_unregister must be called between the last tracepoint
158  * probe unregistration and the end of module exit to make sure there is no
159  * caller executing a probe when it is freed.
160  */
161 static inline void tracepoint_synchronize_unregister(void)
162 {
163 	synchronize_sched();
164 }
165 
166 #define PARAMS(args...) args
167 
168 #endif /* _LINUX_TRACEPOINT_H */
169 
170 /*
171  * Note: we keep the TRACE_EVENT outside the include file ifdef protection.
172  *  This is due to the way trace events work. If a file includes two
173  *  trace event headers under one "CREATE_TRACE_POINTS" the first include
174  *  will override the TRACE_EVENT and break the second include.
175  */
176 
177 #ifndef TRACE_EVENT
178 /*
179  * For use with the TRACE_EVENT macro:
180  *
181  * We define a tracepoint, its arguments, its printk format
182  * and its 'fast binay record' layout.
183  *
184  * Firstly, name your tracepoint via TRACE_EVENT(name : the
185  * 'subsystem_event' notation is fine.
186  *
187  * Think about this whole construct as the
188  * 'trace_sched_switch() function' from now on.
189  *
190  *
191  *  TRACE_EVENT(sched_switch,
192  *
193  *	*
194  *	* A function has a regular function arguments
195  *	* prototype, declare it via TP_PROTO():
196  *	*
197  *
198  *	TP_PROTO(struct rq *rq, struct task_struct *prev,
199  *		 struct task_struct *next),
200  *
201  *	*
202  *	* Define the call signature of the 'function'.
203  *	* (Design sidenote: we use this instead of a
204  *	*  TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
205  *	*
206  *
207  *	TP_ARGS(rq, prev, next),
208  *
209  *	*
210  *	* Fast binary tracing: define the trace record via
211  *	* TP_STRUCT__entry(). You can think about it like a
212  *	* regular C structure local variable definition.
213  *	*
214  *	* This is how the trace record is structured and will
215  *	* be saved into the ring buffer. These are the fields
216  *	* that will be exposed to user-space in
217  *	* /sys/kernel/debug/tracing/events/<*>/format.
218  *	*
219  *	* The declared 'local variable' is called '__entry'
220  *	*
221  *	* __field(pid_t, prev_prid) is equivalent to a standard declariton:
222  *	*
223  *	*	pid_t	prev_pid;
224  *	*
225  *	* __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
226  *	*
227  *	*	char	prev_comm[TASK_COMM_LEN];
228  *	*
229  *
230  *	TP_STRUCT__entry(
231  *		__array(	char,	prev_comm,	TASK_COMM_LEN	)
232  *		__field(	pid_t,	prev_pid			)
233  *		__field(	int,	prev_prio			)
234  *		__array(	char,	next_comm,	TASK_COMM_LEN	)
235  *		__field(	pid_t,	next_pid			)
236  *		__field(	int,	next_prio			)
237  *	),
238  *
239  *	*
240  *	* Assign the entry into the trace record, by embedding
241  *	* a full C statement block into TP_fast_assign(). You
242  *	* can refer to the trace record as '__entry' -
243  *	* otherwise you can put arbitrary C code in here.
244  *	*
245  *	* Note: this C code will execute every time a trace event
246  *	* happens, on an active tracepoint.
247  *	*
248  *
249  *	TP_fast_assign(
250  *		memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
251  *		__entry->prev_pid	= prev->pid;
252  *		__entry->prev_prio	= prev->prio;
253  *		memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
254  *		__entry->next_pid	= next->pid;
255  *		__entry->next_prio	= next->prio;
256  *	)
257  *
258  *	*
259  *	* Formatted output of a trace record via TP_printk().
260  *	* This is how the tracepoint will appear under ftrace
261  *	* plugins that make use of this tracepoint.
262  *	*
263  *	* (raw-binary tracing wont actually perform this step.)
264  *	*
265  *
266  *	TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
267  *		__entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
268  *		__entry->next_comm, __entry->next_pid, __entry->next_prio),
269  *
270  * );
271  *
272  * This macro construct is thus used for the regular printk format
273  * tracing setup, it is used to construct a function pointer based
274  * tracepoint callback (this is used by programmatic plugins and
275  * can also by used by generic instrumentation like SystemTap), and
276  * it is also used to expose a structured trace record in
277  * /sys/kernel/debug/tracing/events/.
278  *
279  * A set of (un)registration functions can be passed to the variant
280  * TRACE_EVENT_FN to perform any (un)registration work.
281  */
282 
283 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
284 #define DEFINE_EVENT(template, name, proto, args)		\
285 	DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
286 #define DEFINE_EVENT_PRINT(template, name, proto, args, print)	\
287 	DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
288 
289 #define TRACE_EVENT(name, proto, args, struct, assign, print)	\
290 	DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
291 #define TRACE_EVENT_FN(name, proto, args, struct,		\
292 		assign, print, reg, unreg)			\
293 	DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
294 
295 #endif /* ifdef TRACE_EVENT (see note above) */
296