xref: /linux-6.15/include/linux/tracepoint.h (revision 37c8a5fa)
1 #ifndef _LINUX_TRACEPOINT_H
2 #define _LINUX_TRACEPOINT_H
3 
4 /*
5  * Kernel Tracepoint API.
6  *
7  * See Documentation/trace/tracepoints.rst.
8  *
9  * Copyright (C) 2008-2014 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/smp.h>
18 #include <linux/srcu.h>
19 #include <linux/errno.h>
20 #include <linux/types.h>
21 #include <linux/cpumask.h>
22 #include <linux/rcupdate.h>
23 #include <linux/tracepoint-defs.h>
24 
25 struct module;
26 struct tracepoint;
27 struct notifier_block;
28 
29 struct trace_eval_map {
30 	const char		*system;
31 	const char		*eval_string;
32 	unsigned long		eval_value;
33 };
34 
35 #define TRACEPOINT_DEFAULT_PRIO	10
36 
37 extern struct srcu_struct tracepoint_srcu;
38 
39 extern int
40 tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data);
41 extern int
42 tracepoint_probe_register_prio(struct tracepoint *tp, void *probe, void *data,
43 			       int prio);
44 extern int
45 tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data);
46 extern void
47 for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
48 		void *priv);
49 
50 #ifdef CONFIG_MODULES
51 struct tp_module {
52 	struct list_head list;
53 	struct module *mod;
54 };
55 
56 bool trace_module_has_bad_taint(struct module *mod);
57 extern int register_tracepoint_module_notifier(struct notifier_block *nb);
58 extern int unregister_tracepoint_module_notifier(struct notifier_block *nb);
59 #else
60 static inline bool trace_module_has_bad_taint(struct module *mod)
61 {
62 	return false;
63 }
64 static inline
65 int register_tracepoint_module_notifier(struct notifier_block *nb)
66 {
67 	return 0;
68 }
69 static inline
70 int unregister_tracepoint_module_notifier(struct notifier_block *nb)
71 {
72 	return 0;
73 }
74 #endif /* CONFIG_MODULES */
75 
76 /*
77  * tracepoint_synchronize_unregister must be called between the last tracepoint
78  * probe unregistration and the end of module exit to make sure there is no
79  * caller executing a probe when it is freed.
80  */
81 #ifdef CONFIG_TRACEPOINTS
82 static inline void tracepoint_synchronize_unregister(void)
83 {
84 	synchronize_srcu(&tracepoint_srcu);
85 	synchronize_sched();
86 }
87 #else
88 static inline void tracepoint_synchronize_unregister(void)
89 { }
90 #endif
91 
92 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
93 extern int syscall_regfunc(void);
94 extern void syscall_unregfunc(void);
95 #endif /* CONFIG_HAVE_SYSCALL_TRACEPOINTS */
96 
97 #define PARAMS(args...) args
98 
99 #define TRACE_DEFINE_ENUM(x)
100 #define TRACE_DEFINE_SIZEOF(x)
101 
102 #endif /* _LINUX_TRACEPOINT_H */
103 
104 /*
105  * Note: we keep the TRACE_EVENT and DECLARE_TRACE outside the include
106  *  file ifdef protection.
107  *  This is due to the way trace events work. If a file includes two
108  *  trace event headers under one "CREATE_TRACE_POINTS" the first include
109  *  will override the TRACE_EVENT and break the second include.
110  */
111 
112 #ifndef DECLARE_TRACE
113 
114 #define TP_PROTO(args...)	args
115 #define TP_ARGS(args...)	args
116 #define TP_CONDITION(args...)	args
117 
118 /*
119  * Individual subsystem my have a separate configuration to
120  * enable their tracepoints. By default, this file will create
121  * the tracepoints if CONFIG_TRACEPOINT is defined. If a subsystem
122  * wants to be able to disable its tracepoints from being created
123  * it can define NOTRACE before including the tracepoint headers.
124  */
125 #if defined(CONFIG_TRACEPOINTS) && !defined(NOTRACE)
126 #define TRACEPOINTS_ENABLED
127 #endif
128 
129 #ifdef TRACEPOINTS_ENABLED
130 
131 /*
132  * it_func[0] is never NULL because there is at least one element in the array
133  * when the array itself is non NULL.
134  *
135  * Note, the proto and args passed in includes "__data" as the first parameter.
136  * The reason for this is to handle the "void" prototype. If a tracepoint
137  * has a "void" prototype, then it is invalid to declare a function
138  * as "(void *, void)". The DECLARE_TRACE_NOARGS() will pass in just
139  * "void *data", where as the DECLARE_TRACE() will pass in "void *data, proto".
140  */
141 #define __DO_TRACE(tp, proto, args, cond, rcuidle)			\
142 	do {								\
143 		struct tracepoint_func *it_func_ptr;			\
144 		void *it_func;						\
145 		void *__data;						\
146 		int __maybe_unused idx = 0;				\
147 									\
148 		if (!(cond))						\
149 			return;						\
150 									\
151 		/* srcu can't be used from NMI */			\
152 		WARN_ON_ONCE(rcuidle && in_nmi());			\
153 									\
154 		/* keep srcu and sched-rcu usage consistent */		\
155 		preempt_disable_notrace();				\
156 									\
157 		/*							\
158 		 * For rcuidle callers, use srcu since sched-rcu	\
159 		 * doesn't work from the idle path.			\
160 		 */							\
161 		if (rcuidle)						\
162 			idx = srcu_read_lock_notrace(&tracepoint_srcu);	\
163 									\
164 		it_func_ptr = rcu_dereference_raw((tp)->funcs);		\
165 									\
166 		if (it_func_ptr) {					\
167 			do {						\
168 				it_func = (it_func_ptr)->func;		\
169 				__data = (it_func_ptr)->data;		\
170 				((void(*)(proto))(it_func))(args);	\
171 			} while ((++it_func_ptr)->func);		\
172 		}							\
173 									\
174 		if (rcuidle)						\
175 			srcu_read_unlock_notrace(&tracepoint_srcu, idx);\
176 									\
177 		preempt_enable_notrace();				\
178 	} while (0)
179 
180 #ifndef MODULE
181 #define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args) \
182 	static inline void trace_##name##_rcuidle(proto)		\
183 	{								\
184 		if (static_key_false(&__tracepoint_##name.key))		\
185 			__DO_TRACE(&__tracepoint_##name,		\
186 				TP_PROTO(data_proto),			\
187 				TP_ARGS(data_args),			\
188 				TP_CONDITION(cond), 1);			\
189 	}
190 #else
191 #define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args)
192 #endif
193 
194 /*
195  * Make sure the alignment of the structure in the __tracepoints section will
196  * not add unwanted padding between the beginning of the section and the
197  * structure. Force alignment to the same alignment as the section start.
198  *
199  * When lockdep is enabled, we make sure to always do the RCU portions of
200  * the tracepoint code, regardless of whether tracing is on. However,
201  * don't check if the condition is false, due to interaction with idle
202  * instrumentation. This lets us find RCU issues triggered with tracepoints
203  * even when this tracepoint is off. This code has no purpose other than
204  * poking RCU a bit.
205  */
206 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
207 	extern struct tracepoint __tracepoint_##name;			\
208 	static inline void trace_##name(proto)				\
209 	{								\
210 		if (static_key_false(&__tracepoint_##name.key))		\
211 			__DO_TRACE(&__tracepoint_##name,		\
212 				TP_PROTO(data_proto),			\
213 				TP_ARGS(data_args),			\
214 				TP_CONDITION(cond), 0);			\
215 		if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) {		\
216 			rcu_read_lock_sched_notrace();			\
217 			rcu_dereference_sched(__tracepoint_##name.funcs);\
218 			rcu_read_unlock_sched_notrace();		\
219 		}							\
220 	}								\
221 	__DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args),		\
222 		PARAMS(cond), PARAMS(data_proto), PARAMS(data_args))	\
223 	static inline int						\
224 	register_trace_##name(void (*probe)(data_proto), void *data)	\
225 	{								\
226 		return tracepoint_probe_register(&__tracepoint_##name,	\
227 						(void *)probe, data);	\
228 	}								\
229 	static inline int						\
230 	register_trace_prio_##name(void (*probe)(data_proto), void *data,\
231 				   int prio)				\
232 	{								\
233 		return tracepoint_probe_register_prio(&__tracepoint_##name, \
234 					      (void *)probe, data, prio); \
235 	}								\
236 	static inline int						\
237 	unregister_trace_##name(void (*probe)(data_proto), void *data)	\
238 	{								\
239 		return tracepoint_probe_unregister(&__tracepoint_##name,\
240 						(void *)probe, data);	\
241 	}								\
242 	static inline void						\
243 	check_trace_callback_type_##name(void (*cb)(data_proto))	\
244 	{								\
245 	}								\
246 	static inline bool						\
247 	trace_##name##_enabled(void)					\
248 	{								\
249 		return static_key_false(&__tracepoint_##name.key);	\
250 	}
251 
252 #ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS
253 #define __TRACEPOINT_ENTRY(name)					\
254 	asm("	.section \"__tracepoints_ptrs\", \"a\"		\n"	\
255 	    "	.balign 4					\n"	\
256 	    "	.long 	__tracepoint_" #name " - .		\n"	\
257 	    "	.previous					\n")
258 #else
259 #define __TRACEPOINT_ENTRY(name)					 \
260 	static struct tracepoint * const __tracepoint_ptr_##name __used	 \
261 	__attribute__((section("__tracepoints_ptrs"))) =		 \
262 		&__tracepoint_##name
263 #endif
264 
265 /*
266  * We have no guarantee that gcc and the linker won't up-align the tracepoint
267  * structures, so we create an array of pointers that will be used for iteration
268  * on the tracepoints.
269  */
270 #define DEFINE_TRACE_FN(name, reg, unreg)				 \
271 	static const char __tpstrtab_##name[]				 \
272 	__attribute__((section("__tracepoints_strings"))) = #name;	 \
273 	struct tracepoint __tracepoint_##name				 \
274 	__attribute__((section("__tracepoints"), used)) =		 \
275 		{ __tpstrtab_##name, STATIC_KEY_INIT_FALSE, reg, unreg, NULL };\
276 	__TRACEPOINT_ENTRY(name);
277 
278 #define DEFINE_TRACE(name)						\
279 	DEFINE_TRACE_FN(name, NULL, NULL);
280 
281 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)				\
282 	EXPORT_SYMBOL_GPL(__tracepoint_##name)
283 #define EXPORT_TRACEPOINT_SYMBOL(name)					\
284 	EXPORT_SYMBOL(__tracepoint_##name)
285 
286 #else /* !TRACEPOINTS_ENABLED */
287 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
288 	static inline void trace_##name(proto)				\
289 	{ }								\
290 	static inline void trace_##name##_rcuidle(proto)		\
291 	{ }								\
292 	static inline int						\
293 	register_trace_##name(void (*probe)(data_proto),		\
294 			      void *data)				\
295 	{								\
296 		return -ENOSYS;						\
297 	}								\
298 	static inline int						\
299 	unregister_trace_##name(void (*probe)(data_proto),		\
300 				void *data)				\
301 	{								\
302 		return -ENOSYS;						\
303 	}								\
304 	static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \
305 	{								\
306 	}								\
307 	static inline bool						\
308 	trace_##name##_enabled(void)					\
309 	{								\
310 		return false;						\
311 	}
312 
313 #define DEFINE_TRACE_FN(name, reg, unreg)
314 #define DEFINE_TRACE(name)
315 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
316 #define EXPORT_TRACEPOINT_SYMBOL(name)
317 
318 #endif /* TRACEPOINTS_ENABLED */
319 
320 #ifdef CONFIG_TRACING
321 /**
322  * tracepoint_string - register constant persistent string to trace system
323  * @str - a constant persistent string that will be referenced in tracepoints
324  *
325  * If constant strings are being used in tracepoints, it is faster and
326  * more efficient to just save the pointer to the string and reference
327  * that with a printf "%s" instead of saving the string in the ring buffer
328  * and wasting space and time.
329  *
330  * The problem with the above approach is that userspace tools that read
331  * the binary output of the trace buffers do not have access to the string.
332  * Instead they just show the address of the string which is not very
333  * useful to users.
334  *
335  * With tracepoint_string(), the string will be registered to the tracing
336  * system and exported to userspace via the debugfs/tracing/printk_formats
337  * file that maps the string address to the string text. This way userspace
338  * tools that read the binary buffers have a way to map the pointers to
339  * the ASCII strings they represent.
340  *
341  * The @str used must be a constant string and persistent as it would not
342  * make sense to show a string that no longer exists. But it is still fine
343  * to be used with modules, because when modules are unloaded, if they
344  * had tracepoints, the ring buffers are cleared too. As long as the string
345  * does not change during the life of the module, it is fine to use
346  * tracepoint_string() within a module.
347  */
348 #define tracepoint_string(str)						\
349 	({								\
350 		static const char *___tp_str __tracepoint_string = str; \
351 		___tp_str;						\
352 	})
353 #define __tracepoint_string	__attribute__((section("__tracepoint_str")))
354 #else
355 /*
356  * tracepoint_string() is used to save the string address for userspace
357  * tracing tools. When tracing isn't configured, there's no need to save
358  * anything.
359  */
360 # define tracepoint_string(str) str
361 # define __tracepoint_string
362 #endif
363 
364 /*
365  * The need for the DECLARE_TRACE_NOARGS() is to handle the prototype
366  * (void). "void" is a special value in a function prototype and can
367  * not be combined with other arguments. Since the DECLARE_TRACE()
368  * macro adds a data element at the beginning of the prototype,
369  * we need a way to differentiate "(void *data, proto)" from
370  * "(void *data, void)". The second prototype is invalid.
371  *
372  * DECLARE_TRACE_NOARGS() passes "void" as the tracepoint prototype
373  * and "void *__data" as the callback prototype.
374  *
375  * DECLARE_TRACE() passes "proto" as the tracepoint protoype and
376  * "void *__data, proto" as the callback prototype.
377  */
378 #define DECLARE_TRACE_NOARGS(name)					\
379 	__DECLARE_TRACE(name, void, ,					\
380 			cpu_online(raw_smp_processor_id()),		\
381 			void *__data, __data)
382 
383 #define DECLARE_TRACE(name, proto, args)				\
384 	__DECLARE_TRACE(name, PARAMS(proto), PARAMS(args),		\
385 			cpu_online(raw_smp_processor_id()),		\
386 			PARAMS(void *__data, proto),			\
387 			PARAMS(__data, args))
388 
389 #define DECLARE_TRACE_CONDITION(name, proto, args, cond)		\
390 	__DECLARE_TRACE(name, PARAMS(proto), PARAMS(args),		\
391 			cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \
392 			PARAMS(void *__data, proto),			\
393 			PARAMS(__data, args))
394 
395 #define TRACE_EVENT_FLAGS(event, flag)
396 
397 #define TRACE_EVENT_PERF_PERM(event, expr...)
398 
399 #endif /* DECLARE_TRACE */
400 
401 #ifndef TRACE_EVENT
402 /*
403  * For use with the TRACE_EVENT macro:
404  *
405  * We define a tracepoint, its arguments, its printk format
406  * and its 'fast binary record' layout.
407  *
408  * Firstly, name your tracepoint via TRACE_EVENT(name : the
409  * 'subsystem_event' notation is fine.
410  *
411  * Think about this whole construct as the
412  * 'trace_sched_switch() function' from now on.
413  *
414  *
415  *  TRACE_EVENT(sched_switch,
416  *
417  *	*
418  *	* A function has a regular function arguments
419  *	* prototype, declare it via TP_PROTO():
420  *	*
421  *
422  *	TP_PROTO(struct rq *rq, struct task_struct *prev,
423  *		 struct task_struct *next),
424  *
425  *	*
426  *	* Define the call signature of the 'function'.
427  *	* (Design sidenote: we use this instead of a
428  *	*  TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
429  *	*
430  *
431  *	TP_ARGS(rq, prev, next),
432  *
433  *	*
434  *	* Fast binary tracing: define the trace record via
435  *	* TP_STRUCT__entry(). You can think about it like a
436  *	* regular C structure local variable definition.
437  *	*
438  *	* This is how the trace record is structured and will
439  *	* be saved into the ring buffer. These are the fields
440  *	* that will be exposed to user-space in
441  *	* /sys/kernel/debug/tracing/events/<*>/format.
442  *	*
443  *	* The declared 'local variable' is called '__entry'
444  *	*
445  *	* __field(pid_t, prev_prid) is equivalent to a standard declariton:
446  *	*
447  *	*	pid_t	prev_pid;
448  *	*
449  *	* __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
450  *	*
451  *	*	char	prev_comm[TASK_COMM_LEN];
452  *	*
453  *
454  *	TP_STRUCT__entry(
455  *		__array(	char,	prev_comm,	TASK_COMM_LEN	)
456  *		__field(	pid_t,	prev_pid			)
457  *		__field(	int,	prev_prio			)
458  *		__array(	char,	next_comm,	TASK_COMM_LEN	)
459  *		__field(	pid_t,	next_pid			)
460  *		__field(	int,	next_prio			)
461  *	),
462  *
463  *	*
464  *	* Assign the entry into the trace record, by embedding
465  *	* a full C statement block into TP_fast_assign(). You
466  *	* can refer to the trace record as '__entry' -
467  *	* otherwise you can put arbitrary C code in here.
468  *	*
469  *	* Note: this C code will execute every time a trace event
470  *	* happens, on an active tracepoint.
471  *	*
472  *
473  *	TP_fast_assign(
474  *		memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
475  *		__entry->prev_pid	= prev->pid;
476  *		__entry->prev_prio	= prev->prio;
477  *		memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
478  *		__entry->next_pid	= next->pid;
479  *		__entry->next_prio	= next->prio;
480  *	),
481  *
482  *	*
483  *	* Formatted output of a trace record via TP_printk().
484  *	* This is how the tracepoint will appear under ftrace
485  *	* plugins that make use of this tracepoint.
486  *	*
487  *	* (raw-binary tracing wont actually perform this step.)
488  *	*
489  *
490  *	TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
491  *		__entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
492  *		__entry->next_comm, __entry->next_pid, __entry->next_prio),
493  *
494  * );
495  *
496  * This macro construct is thus used for the regular printk format
497  * tracing setup, it is used to construct a function pointer based
498  * tracepoint callback (this is used by programmatic plugins and
499  * can also by used by generic instrumentation like SystemTap), and
500  * it is also used to expose a structured trace record in
501  * /sys/kernel/debug/tracing/events/.
502  *
503  * A set of (un)registration functions can be passed to the variant
504  * TRACE_EVENT_FN to perform any (un)registration work.
505  */
506 
507 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
508 #define DEFINE_EVENT(template, name, proto, args)		\
509 	DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
510 #define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\
511 	DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
512 #define DEFINE_EVENT_PRINT(template, name, proto, args, print)	\
513 	DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
514 #define DEFINE_EVENT_CONDITION(template, name, proto,		\
515 			       args, cond)			\
516 	DECLARE_TRACE_CONDITION(name, PARAMS(proto),		\
517 				PARAMS(args), PARAMS(cond))
518 
519 #define TRACE_EVENT(name, proto, args, struct, assign, print)	\
520 	DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
521 #define TRACE_EVENT_FN(name, proto, args, struct,		\
522 		assign, print, reg, unreg)			\
523 	DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
524 #define TRACE_EVENT_FN_COND(name, proto, args, cond, struct,		\
525 		assign, print, reg, unreg)			\
526 	DECLARE_TRACE_CONDITION(name, PARAMS(proto),	\
527 			PARAMS(args), PARAMS(cond))
528 #define TRACE_EVENT_CONDITION(name, proto, args, cond,		\
529 			      struct, assign, print)		\
530 	DECLARE_TRACE_CONDITION(name, PARAMS(proto),		\
531 				PARAMS(args), PARAMS(cond))
532 
533 #define TRACE_EVENT_FLAGS(event, flag)
534 
535 #define TRACE_EVENT_PERF_PERM(event, expr...)
536 
537 #endif /* ifdef TRACE_EVENT (see note above) */
538