kvm tools, setup: Create private directory
[linux-2.6/next.git] / include / linux / tracepoint.h
blobd530a4460a0b96d4a73ebfb1a65e7fe32547f210
1 #ifndef _LINUX_TRACEPOINT_H
2 #define _LINUX_TRACEPOINT_H
4 /*
5 * Kernel Tracepoint API.
7 * See Documentation/trace/tracepoints.txt.
9 * (C) Copyright 2008 Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
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/jump_label.h>
22 struct module;
23 struct tracepoint;
25 struct tracepoint_func {
26 void *func;
27 void *data;
30 struct tracepoint {
31 const char *name; /* Tracepoint name */
32 struct jump_label_key key;
33 void (*regfunc)(void);
34 void (*unregfunc)(void);
35 struct tracepoint_func __rcu *funcs;
39 * Connect a probe to a tracepoint.
40 * Internal API, should not be used directly.
42 extern int tracepoint_probe_register(const char *name, void *probe, void *data);
45 * Disconnect a probe from a tracepoint.
46 * Internal API, should not be used directly.
48 extern int
49 tracepoint_probe_unregister(const char *name, void *probe, void *data);
51 extern int tracepoint_probe_register_noupdate(const char *name, void *probe,
52 void *data);
53 extern int tracepoint_probe_unregister_noupdate(const char *name, void *probe,
54 void *data);
55 extern void tracepoint_probe_update_all(void);
57 struct tracepoint_iter {
58 struct module *module;
59 struct tracepoint * const *tracepoint;
62 extern void tracepoint_iter_start(struct tracepoint_iter *iter);
63 extern void tracepoint_iter_next(struct tracepoint_iter *iter);
64 extern void tracepoint_iter_stop(struct tracepoint_iter *iter);
65 extern void tracepoint_iter_reset(struct tracepoint_iter *iter);
66 extern int tracepoint_get_iter_range(struct tracepoint * const **tracepoint,
67 struct tracepoint * const *begin, struct tracepoint * const *end);
70 * tracepoint_synchronize_unregister must be called between the last tracepoint
71 * probe unregistration and the end of module exit to make sure there is no
72 * caller executing a probe when it is freed.
74 static inline void tracepoint_synchronize_unregister(void)
76 synchronize_sched();
79 #define PARAMS(args...) args
81 #ifdef CONFIG_TRACEPOINTS
82 extern
83 void tracepoint_update_probe_range(struct tracepoint * const *begin,
84 struct tracepoint * const *end);
85 #else
86 static inline
87 void tracepoint_update_probe_range(struct tracepoint * const *begin,
88 struct tracepoint * const *end)
89 { }
90 #endif /* CONFIG_TRACEPOINTS */
92 #endif /* _LINUX_TRACEPOINT_H */
95 * Note: we keep the TRACE_EVENT and DECLARE_TRACE outside the include
96 * file ifdef protection.
97 * This is due to the way trace events work. If a file includes two
98 * trace event headers under one "CREATE_TRACE_POINTS" the first include
99 * will override the TRACE_EVENT and break the second include.
102 #ifndef DECLARE_TRACE
104 #define TP_PROTO(args...) args
105 #define TP_ARGS(args...) args
106 #define TP_CONDITION(args...) args
108 #ifdef CONFIG_TRACEPOINTS
111 * it_func[0] is never NULL because there is at least one element in the array
112 * when the array itself is non NULL.
114 * Note, the proto and args passed in includes "__data" as the first parameter.
115 * The reason for this is to handle the "void" prototype. If a tracepoint
116 * has a "void" prototype, then it is invalid to declare a function
117 * as "(void *, void)". The DECLARE_TRACE_NOARGS() will pass in just
118 * "void *data", where as the DECLARE_TRACE() will pass in "void *data, proto".
120 #define __DO_TRACE(tp, proto, args, cond) \
121 do { \
122 struct tracepoint_func *it_func_ptr; \
123 void *it_func; \
124 void *__data; \
126 if (!(cond)) \
127 return; \
128 rcu_read_lock_sched_notrace(); \
129 it_func_ptr = rcu_dereference_sched((tp)->funcs); \
130 if (it_func_ptr) { \
131 do { \
132 it_func = (it_func_ptr)->func; \
133 __data = (it_func_ptr)->data; \
134 ((void(*)(proto))(it_func))(args); \
135 } while ((++it_func_ptr)->func); \
137 rcu_read_unlock_sched_notrace(); \
138 } while (0)
141 * Make sure the alignment of the structure in the __tracepoints section will
142 * not add unwanted padding between the beginning of the section and the
143 * structure. Force alignment to the same alignment as the section start.
145 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
146 extern struct tracepoint __tracepoint_##name; \
147 static inline void trace_##name(proto) \
149 if (static_branch(&__tracepoint_##name.key)) \
150 __DO_TRACE(&__tracepoint_##name, \
151 TP_PROTO(data_proto), \
152 TP_ARGS(data_args), \
153 TP_CONDITION(cond)); \
155 static inline int \
156 register_trace_##name(void (*probe)(data_proto), void *data) \
158 return tracepoint_probe_register(#name, (void *)probe, \
159 data); \
161 static inline int \
162 unregister_trace_##name(void (*probe)(data_proto), void *data) \
164 return tracepoint_probe_unregister(#name, (void *)probe, \
165 data); \
167 static inline void \
168 check_trace_callback_type_##name(void (*cb)(data_proto)) \
173 * We have no guarantee that gcc and the linker won't up-align the tracepoint
174 * structures, so we create an array of pointers that will be used for iteration
175 * on the tracepoints.
177 #define DEFINE_TRACE_FN(name, reg, unreg) \
178 static const char __tpstrtab_##name[] \
179 __attribute__((section("__tracepoints_strings"))) = #name; \
180 struct tracepoint __tracepoint_##name \
181 __attribute__((section("__tracepoints"))) = \
182 { __tpstrtab_##name, JUMP_LABEL_INIT, reg, unreg, NULL };\
183 static struct tracepoint * const __tracepoint_ptr_##name __used \
184 __attribute__((section("__tracepoints_ptrs"))) = \
185 &__tracepoint_##name;
187 #define DEFINE_TRACE(name) \
188 DEFINE_TRACE_FN(name, NULL, NULL);
190 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \
191 EXPORT_SYMBOL_GPL(__tracepoint_##name)
192 #define EXPORT_TRACEPOINT_SYMBOL(name) \
193 EXPORT_SYMBOL(__tracepoint_##name)
195 #else /* !CONFIG_TRACEPOINTS */
196 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
197 static inline void trace_##name(proto) \
198 { } \
199 static inline int \
200 register_trace_##name(void (*probe)(data_proto), \
201 void *data) \
203 return -ENOSYS; \
205 static inline int \
206 unregister_trace_##name(void (*probe)(data_proto), \
207 void *data) \
209 return -ENOSYS; \
211 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \
215 #define DEFINE_TRACE_FN(name, reg, unreg)
216 #define DEFINE_TRACE(name)
217 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
218 #define EXPORT_TRACEPOINT_SYMBOL(name)
220 #endif /* CONFIG_TRACEPOINTS */
223 * The need for the DECLARE_TRACE_NOARGS() is to handle the prototype
224 * (void). "void" is a special value in a function prototype and can
225 * not be combined with other arguments. Since the DECLARE_TRACE()
226 * macro adds a data element at the beginning of the prototype,
227 * we need a way to differentiate "(void *data, proto)" from
228 * "(void *data, void)". The second prototype is invalid.
230 * DECLARE_TRACE_NOARGS() passes "void" as the tracepoint prototype
231 * and "void *__data" as the callback prototype.
233 * DECLARE_TRACE() passes "proto" as the tracepoint protoype and
234 * "void *__data, proto" as the callback prototype.
236 #define DECLARE_TRACE_NOARGS(name) \
237 __DECLARE_TRACE(name, void, , 1, void *__data, __data)
239 #define DECLARE_TRACE(name, proto, args) \
240 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), 1, \
241 PARAMS(void *__data, proto), \
242 PARAMS(__data, args))
244 #define DECLARE_TRACE_CONDITION(name, proto, args, cond) \
245 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), PARAMS(cond), \
246 PARAMS(void *__data, proto), \
247 PARAMS(__data, args))
249 #define TRACE_EVENT_FLAGS(event, flag)
251 #endif /* DECLARE_TRACE */
253 #ifndef TRACE_EVENT
255 * For use with the TRACE_EVENT macro:
257 * We define a tracepoint, its arguments, its printk format
258 * and its 'fast binay record' layout.
260 * Firstly, name your tracepoint via TRACE_EVENT(name : the
261 * 'subsystem_event' notation is fine.
263 * Think about this whole construct as the
264 * 'trace_sched_switch() function' from now on.
267 * TRACE_EVENT(sched_switch,
270 * * A function has a regular function arguments
271 * * prototype, declare it via TP_PROTO():
274 * TP_PROTO(struct rq *rq, struct task_struct *prev,
275 * struct task_struct *next),
278 * * Define the call signature of the 'function'.
279 * * (Design sidenote: we use this instead of a
280 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
283 * TP_ARGS(rq, prev, next),
286 * * Fast binary tracing: define the trace record via
287 * * TP_STRUCT__entry(). You can think about it like a
288 * * regular C structure local variable definition.
290 * * This is how the trace record is structured and will
291 * * be saved into the ring buffer. These are the fields
292 * * that will be exposed to user-space in
293 * * /sys/kernel/debug/tracing/events/<*>/format.
295 * * The declared 'local variable' is called '__entry'
297 * * __field(pid_t, prev_prid) is equivalent to a standard declariton:
299 * * pid_t prev_pid;
301 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
303 * * char prev_comm[TASK_COMM_LEN];
306 * TP_STRUCT__entry(
307 * __array( char, prev_comm, TASK_COMM_LEN )
308 * __field( pid_t, prev_pid )
309 * __field( int, prev_prio )
310 * __array( char, next_comm, TASK_COMM_LEN )
311 * __field( pid_t, next_pid )
312 * __field( int, next_prio )
313 * ),
316 * * Assign the entry into the trace record, by embedding
317 * * a full C statement block into TP_fast_assign(). You
318 * * can refer to the trace record as '__entry' -
319 * * otherwise you can put arbitrary C code in here.
321 * * Note: this C code will execute every time a trace event
322 * * happens, on an active tracepoint.
325 * TP_fast_assign(
326 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
327 * __entry->prev_pid = prev->pid;
328 * __entry->prev_prio = prev->prio;
329 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
330 * __entry->next_pid = next->pid;
331 * __entry->next_prio = next->prio;
332 * ),
335 * * Formatted output of a trace record via TP_printk().
336 * * This is how the tracepoint will appear under ftrace
337 * * plugins that make use of this tracepoint.
339 * * (raw-binary tracing wont actually perform this step.)
342 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
343 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
344 * __entry->next_comm, __entry->next_pid, __entry->next_prio),
346 * );
348 * This macro construct is thus used for the regular printk format
349 * tracing setup, it is used to construct a function pointer based
350 * tracepoint callback (this is used by programmatic plugins and
351 * can also by used by generic instrumentation like SystemTap), and
352 * it is also used to expose a structured trace record in
353 * /sys/kernel/debug/tracing/events/.
355 * A set of (un)registration functions can be passed to the variant
356 * TRACE_EVENT_FN to perform any (un)registration work.
359 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
360 #define DEFINE_EVENT(template, name, proto, args) \
361 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
362 #define DEFINE_EVENT_PRINT(template, name, proto, args, print) \
363 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
364 #define DEFINE_EVENT_CONDITION(template, name, proto, \
365 args, cond) \
366 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
367 PARAMS(args), PARAMS(cond))
369 #define TRACE_EVENT(name, proto, args, struct, assign, print) \
370 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
371 #define TRACE_EVENT_FN(name, proto, args, struct, \
372 assign, print, reg, unreg) \
373 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
374 #define TRACE_EVENT_CONDITION(name, proto, args, cond, \
375 struct, assign, print) \
376 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
377 PARAMS(args), PARAMS(cond))
379 #define TRACE_EVENT_FLAGS(event, flag)
381 #endif /* ifdef TRACE_EVENT (see note above) */