1 #ifndef _LINUX_TRACEPOINT_H
2 #define _LINUX_TRACEPOINT_H
5 * Kernel Tracepoint API.
7 * See Documentation/trace/tracepoints.txt.
9 * Copyright (C) 2008-2014 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
11 * Heavily inspired from the Linux Kernel Markers.
13 * This file is released under the GPLv2.
14 * See the file COPYING for more details.
17 #include <linux/errno.h>
18 #include <linux/types.h>
19 #include <linux/rcupdate.h>
20 #include <linux/static_key.h>
24 struct notifier_block;
26 struct tracepoint_func {
32 const char *name; /* Tracepoint name */
33 struct static_key key;
34 void (*regfunc)(void);
35 void (*unregfunc)(void);
36 struct tracepoint_func __rcu *funcs;
40 tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data);
42 tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data);
44 for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
49 struct list_head list;
53 bool trace_module_has_bad_taint(struct module *mod);
54 extern int register_tracepoint_module_notifier(struct notifier_block *nb);
55 extern int unregister_tracepoint_module_notifier(struct notifier_block *nb);
57 static inline bool trace_module_has_bad_taint(struct module *mod)
62 int register_tracepoint_module_notifier(struct notifier_block *nb)
67 int unregister_tracepoint_module_notifier(struct notifier_block *nb)
71 #endif /* CONFIG_MODULES */
74 * tracepoint_synchronize_unregister must be called between the last tracepoint
75 * probe unregistration and the end of module exit to make sure there is no
76 * caller executing a probe when it is freed.
78 static inline void tracepoint_synchronize_unregister(void)
83 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
84 extern void syscall_regfunc(void);
85 extern void syscall_unregfunc(void);
86 #endif /* CONFIG_HAVE_SYSCALL_TRACEPOINTS */
88 #define PARAMS(args...) args
90 #endif /* _LINUX_TRACEPOINT_H */
93 * Note: we keep the TRACE_EVENT and DECLARE_TRACE outside the include
94 * file ifdef protection.
95 * This is due to the way trace events work. If a file includes two
96 * trace event headers under one "CREATE_TRACE_POINTS" the first include
97 * will override the TRACE_EVENT and break the second include.
100 #ifndef DECLARE_TRACE
102 #define TP_PROTO(args...) args
103 #define TP_ARGS(args...) args
104 #define TP_CONDITION(args...) args
106 #ifdef CONFIG_TRACEPOINTS
109 * it_func[0] is never NULL because there is at least one element in the array
110 * when the array itself is non NULL.
112 * Note, the proto and args passed in includes "__data" as the first parameter.
113 * The reason for this is to handle the "void" prototype. If a tracepoint
114 * has a "void" prototype, then it is invalid to declare a function
115 * as "(void *, void)". The DECLARE_TRACE_NOARGS() will pass in just
116 * "void *data", where as the DECLARE_TRACE() will pass in "void *data, proto".
118 #define __DO_TRACE(tp, proto, args, cond, prercu, postrcu) \
120 struct tracepoint_func *it_func_ptr; \
127 rcu_read_lock_sched_notrace(); \
128 it_func_ptr = rcu_dereference_sched((tp)->funcs); \
131 it_func = (it_func_ptr)->func; \
132 __data = (it_func_ptr)->data; \
133 ((void(*)(proto))(it_func))(args); \
134 } while ((++it_func_ptr)->func); \
136 rcu_read_unlock_sched_notrace(); \
141 #define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args) \
142 static inline void trace_##name##_rcuidle(proto) \
144 if (static_key_false(&__tracepoint_##name.key)) \
145 __DO_TRACE(&__tracepoint_##name, \
146 TP_PROTO(data_proto), \
147 TP_ARGS(data_args), \
148 TP_CONDITION(cond), \
153 #define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args)
157 * Make sure the alignment of the structure in the __tracepoints section will
158 * not add unwanted padding between the beginning of the section and the
159 * structure. Force alignment to the same alignment as the section start.
161 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
162 extern struct tracepoint __tracepoint_##name; \
163 static inline void trace_##name(proto) \
165 if (static_key_false(&__tracepoint_##name.key)) \
166 __DO_TRACE(&__tracepoint_##name, \
167 TP_PROTO(data_proto), \
168 TP_ARGS(data_args), \
169 TP_CONDITION(cond),,); \
171 __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args), \
172 PARAMS(cond), PARAMS(data_proto), PARAMS(data_args)) \
174 register_trace_##name(void (*probe)(data_proto), void *data) \
176 return tracepoint_probe_register(&__tracepoint_##name, \
177 (void *)probe, data); \
180 unregister_trace_##name(void (*probe)(data_proto), void *data) \
182 return tracepoint_probe_unregister(&__tracepoint_##name,\
183 (void *)probe, data); \
186 check_trace_callback_type_##name(void (*cb)(data_proto)) \
190 trace_##name##_enabled(void) \
192 return static_key_false(&__tracepoint_##name.key); \
196 * We have no guarantee that gcc and the linker won't up-align the tracepoint
197 * structures, so we create an array of pointers that will be used for iteration
198 * on the tracepoints.
200 #define DEFINE_TRACE_FN(name, reg, unreg) \
201 static const char __tpstrtab_##name[] \
202 __attribute__((section("__tracepoints_strings"))) = #name; \
203 struct tracepoint __tracepoint_##name \
204 __attribute__((section("__tracepoints"))) = \
205 { __tpstrtab_##name, STATIC_KEY_INIT_FALSE, reg, unreg, NULL };\
206 static struct tracepoint * const __tracepoint_ptr_##name __used \
207 __attribute__((section("__tracepoints_ptrs"))) = \
208 &__tracepoint_##name;
210 #define DEFINE_TRACE(name) \
211 DEFINE_TRACE_FN(name, NULL, NULL);
213 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \
214 EXPORT_SYMBOL_GPL(__tracepoint_##name)
215 #define EXPORT_TRACEPOINT_SYMBOL(name) \
216 EXPORT_SYMBOL(__tracepoint_##name)
218 #else /* !CONFIG_TRACEPOINTS */
219 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
220 static inline void trace_##name(proto) \
222 static inline void trace_##name##_rcuidle(proto) \
225 register_trace_##name(void (*probe)(data_proto), \
231 unregister_trace_##name(void (*probe)(data_proto), \
236 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \
240 trace_##name##_enabled(void) \
245 #define DEFINE_TRACE_FN(name, reg, unreg)
246 #define DEFINE_TRACE(name)
247 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
248 #define EXPORT_TRACEPOINT_SYMBOL(name)
250 #endif /* CONFIG_TRACEPOINTS */
252 #ifdef CONFIG_TRACING
254 * tracepoint_string - register constant persistent string to trace system
255 * @str - a constant persistent string that will be referenced in tracepoints
257 * If constant strings are being used in tracepoints, it is faster and
258 * more efficient to just save the pointer to the string and reference
259 * that with a printf "%s" instead of saving the string in the ring buffer
260 * and wasting space and time.
262 * The problem with the above approach is that userspace tools that read
263 * the binary output of the trace buffers do not have access to the string.
264 * Instead they just show the address of the string which is not very
267 * With tracepoint_string(), the string will be registered to the tracing
268 * system and exported to userspace via the debugfs/tracing/printk_formats
269 * file that maps the string address to the string text. This way userspace
270 * tools that read the binary buffers have a way to map the pointers to
271 * the ASCII strings they represent.
273 * The @str used must be a constant string and persistent as it would not
274 * make sense to show a string that no longer exists. But it is still fine
275 * to be used with modules, because when modules are unloaded, if they
276 * had tracepoints, the ring buffers are cleared too. As long as the string
277 * does not change during the life of the module, it is fine to use
278 * tracepoint_string() within a module.
280 #define tracepoint_string(str) \
282 static const char *___tp_str __tracepoint_string = str; \
285 #define __tracepoint_string __attribute__((section("__tracepoint_str")))
288 * tracepoint_string() is used to save the string address for userspace
289 * tracing tools. When tracing isn't configured, there's no need to save
292 # define tracepoint_string(str) str
293 # define __tracepoint_string
297 * The need for the DECLARE_TRACE_NOARGS() is to handle the prototype
298 * (void). "void" is a special value in a function prototype and can
299 * not be combined with other arguments. Since the DECLARE_TRACE()
300 * macro adds a data element at the beginning of the prototype,
301 * we need a way to differentiate "(void *data, proto)" from
302 * "(void *data, void)". The second prototype is invalid.
304 * DECLARE_TRACE_NOARGS() passes "void" as the tracepoint prototype
305 * and "void *__data" as the callback prototype.
307 * DECLARE_TRACE() passes "proto" as the tracepoint protoype and
308 * "void *__data, proto" as the callback prototype.
310 #define DECLARE_TRACE_NOARGS(name) \
311 __DECLARE_TRACE(name, void, , 1, void *__data, __data)
313 #define DECLARE_TRACE(name, proto, args) \
314 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), 1, \
315 PARAMS(void *__data, proto), \
316 PARAMS(__data, args))
318 #define DECLARE_TRACE_CONDITION(name, proto, args, cond) \
319 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), PARAMS(cond), \
320 PARAMS(void *__data, proto), \
321 PARAMS(__data, args))
323 #define TRACE_EVENT_FLAGS(event, flag)
325 #define TRACE_EVENT_PERF_PERM(event, expr...)
327 #endif /* DECLARE_TRACE */
331 * For use with the TRACE_EVENT macro:
333 * We define a tracepoint, its arguments, its printk format
334 * and its 'fast binary record' layout.
336 * Firstly, name your tracepoint via TRACE_EVENT(name : the
337 * 'subsystem_event' notation is fine.
339 * Think about this whole construct as the
340 * 'trace_sched_switch() function' from now on.
343 * TRACE_EVENT(sched_switch,
346 * * A function has a regular function arguments
347 * * prototype, declare it via TP_PROTO():
350 * TP_PROTO(struct rq *rq, struct task_struct *prev,
351 * struct task_struct *next),
354 * * Define the call signature of the 'function'.
355 * * (Design sidenote: we use this instead of a
356 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
359 * TP_ARGS(rq, prev, next),
362 * * Fast binary tracing: define the trace record via
363 * * TP_STRUCT__entry(). You can think about it like a
364 * * regular C structure local variable definition.
366 * * This is how the trace record is structured and will
367 * * be saved into the ring buffer. These are the fields
368 * * that will be exposed to user-space in
369 * * /sys/kernel/debug/tracing/events/<*>/format.
371 * * The declared 'local variable' is called '__entry'
373 * * __field(pid_t, prev_prid) is equivalent to a standard declariton:
377 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
379 * * char prev_comm[TASK_COMM_LEN];
383 * __array( char, prev_comm, TASK_COMM_LEN )
384 * __field( pid_t, prev_pid )
385 * __field( int, prev_prio )
386 * __array( char, next_comm, TASK_COMM_LEN )
387 * __field( pid_t, next_pid )
388 * __field( int, next_prio )
392 * * Assign the entry into the trace record, by embedding
393 * * a full C statement block into TP_fast_assign(). You
394 * * can refer to the trace record as '__entry' -
395 * * otherwise you can put arbitrary C code in here.
397 * * Note: this C code will execute every time a trace event
398 * * happens, on an active tracepoint.
402 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
403 * __entry->prev_pid = prev->pid;
404 * __entry->prev_prio = prev->prio;
405 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
406 * __entry->next_pid = next->pid;
407 * __entry->next_prio = next->prio;
411 * * Formatted output of a trace record via TP_printk().
412 * * This is how the tracepoint will appear under ftrace
413 * * plugins that make use of this tracepoint.
415 * * (raw-binary tracing wont actually perform this step.)
418 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
419 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
420 * __entry->next_comm, __entry->next_pid, __entry->next_prio),
424 * This macro construct is thus used for the regular printk format
425 * tracing setup, it is used to construct a function pointer based
426 * tracepoint callback (this is used by programmatic plugins and
427 * can also by used by generic instrumentation like SystemTap), and
428 * it is also used to expose a structured trace record in
429 * /sys/kernel/debug/tracing/events/.
431 * A set of (un)registration functions can be passed to the variant
432 * TRACE_EVENT_FN to perform any (un)registration work.
435 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
436 #define DEFINE_EVENT(template, name, proto, args) \
437 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
438 #define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\
439 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
440 #define DEFINE_EVENT_PRINT(template, name, proto, args, print) \
441 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
442 #define DEFINE_EVENT_CONDITION(template, name, proto, \
444 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
445 PARAMS(args), PARAMS(cond))
447 #define TRACE_EVENT(name, proto, args, struct, assign, print) \
448 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
449 #define TRACE_EVENT_FN(name, proto, args, struct, \
450 assign, print, reg, unreg) \
451 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
452 #define TRACE_EVENT_CONDITION(name, proto, args, cond, \
453 struct, assign, print) \
454 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
455 PARAMS(args), PARAMS(cond))
457 #define TRACE_EVENT_FLAGS(event, flag)
459 #define TRACE_EVENT_PERF_PERM(event, expr...)
461 #endif /* ifdef TRACE_EVENT (see note above) */