tracing: Let tracepoints have data passed to tracepoint callbacks
[linux-2.6.git] / include / linux / tracepoint.h
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 <mathieu.desnoyers@polymtl.ca>
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_func {
24         void *func;
25         void *data;
26 };
27
28 struct tracepoint {
29         const char *name;               /* Tracepoint name */
30         int state;                      /* State. */
31         void (*regfunc)(void);
32         void (*unregfunc)(void);
33         struct tracepoint_func *funcs;
34 } __attribute__((aligned(32)));         /*
35                                          * Aligned on 32 bytes because it is
36                                          * globally visible and gcc happily
37                                          * align these on the structure size.
38                                          * Keep in sync with vmlinux.lds.h.
39                                          */
40
41 /*
42  * Connect a probe to a tracepoint.
43  * Internal API, should not be used directly.
44  */
45 extern int tracepoint_probe_register(const char *name, void *probe, void *data);
46
47 /*
48  * Disconnect a probe from a tracepoint.
49  * Internal API, should not be used directly.
50  */
51 extern int
52 tracepoint_probe_unregister(const char *name, void *probe, void *data);
53
54 extern int tracepoint_probe_register_noupdate(const char *name, void *probe,
55                                               void *data);
56 extern int tracepoint_probe_unregister_noupdate(const char *name, void *probe,
57                                                 void *data);
58 extern void tracepoint_probe_update_all(void);
59
60 struct tracepoint_iter {
61         struct module *module;
62         struct tracepoint *tracepoint;
63 };
64
65 extern void tracepoint_iter_start(struct tracepoint_iter *iter);
66 extern void tracepoint_iter_next(struct tracepoint_iter *iter);
67 extern void tracepoint_iter_stop(struct tracepoint_iter *iter);
68 extern void tracepoint_iter_reset(struct tracepoint_iter *iter);
69 extern int tracepoint_get_iter_range(struct tracepoint **tracepoint,
70         struct tracepoint *begin, struct tracepoint *end);
71
72 /*
73  * tracepoint_synchronize_unregister must be called between the last tracepoint
74  * probe unregistration and the end of module exit to make sure there is no
75  * caller executing a probe when it is freed.
76  */
77 static inline void tracepoint_synchronize_unregister(void)
78 {
79         synchronize_sched();
80 }
81
82 #define PARAMS(args...) args
83
84 #ifdef CONFIG_TRACEPOINTS
85 extern void tracepoint_update_probe_range(struct tracepoint *begin,
86         struct tracepoint *end);
87 #else
88 static inline void tracepoint_update_probe_range(struct tracepoint *begin,
89         struct tracepoint *end)
90 { }
91 #endif /* CONFIG_TRACEPOINTS */
92
93 #endif /* _LINUX_TRACEPOINT_H */
94
95 /*
96  * Note: we keep the TRACE_EVENT and DECLARE_TRACE outside the include
97  *  file ifdef protection.
98  *  This is due to the way trace events work. If a file includes two
99  *  trace event headers under one "CREATE_TRACE_POINTS" the first include
100  *  will override the TRACE_EVENT and break the second include.
101  */
102
103 #ifndef DECLARE_TRACE
104
105 #define TP_PROTO(args...)       args
106 #define TP_ARGS(args...)        args
107
108 #ifdef CONFIG_TRACEPOINTS
109
110 /*
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.
113  *
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".
119  */
120 #define __DO_TRACE(tp, proto, args)                                     \
121         do {                                                            \
122                 struct tracepoint_func *it_func_ptr;                    \
123                 void *it_func;                                          \
124                 void *__data;                                           \
125                                                                         \
126                 rcu_read_lock_sched_notrace();                          \
127                 it_func_ptr = rcu_dereference_sched((tp)->funcs);       \
128                 if (it_func_ptr) {                                      \
129                         do {                                            \
130                                 it_func = (it_func_ptr)->func;          \
131                                 __data = (it_func_ptr)->data;           \
132                                 ((void(*)(proto))(it_func))(args);      \
133                         } while ((++it_func_ptr)->func);                \
134                 }                                                       \
135                 rcu_read_unlock_sched_notrace();                        \
136         } while (0)
137
138 /*
139  * Make sure the alignment of the structure in the __tracepoints section will
140  * not add unwanted padding between the beginning of the section and the
141  * structure. Force alignment to the same alignment as the section start.
142  */
143 #define __DECLARE_TRACE(name, proto, args, data_proto, data_args)       \
144         extern struct tracepoint __tracepoint_##name;                   \
145         static inline void trace_##name(proto)                          \
146         {                                                               \
147                 if (unlikely(__tracepoint_##name.state))                \
148                         __DO_TRACE(&__tracepoint_##name,                \
149                                 TP_PROTO(data_proto),                   \
150                                 TP_ARGS(data_args));                    \
151         }                                                               \
152         static inline int                                               \
153         register_trace_##name(void (*probe)(data_proto), void *data)    \
154         {                                                               \
155                 return tracepoint_probe_register(#name, (void *)probe,  \
156                                                  data);                 \
157         }                                                               \
158         static inline int                                               \
159         unregister_trace_##name(void (*probe)(data_proto), void *data)  \
160         {                                                               \
161                 return tracepoint_probe_unregister(#name, (void *)probe, \
162                                                    data);               \
163         }                                                               \
164         static inline void                                              \
165         check_trace_callback_type_##name(void (*cb)(data_proto))        \
166         {                                                               \
167         }
168
169 #define DEFINE_TRACE_FN(name, reg, unreg)                               \
170         static const char __tpstrtab_##name[]                           \
171         __attribute__((section("__tracepoints_strings"))) = #name;      \
172         struct tracepoint __tracepoint_##name                           \
173         __attribute__((section("__tracepoints"), aligned(32))) =        \
174                 { __tpstrtab_##name, 0, reg, unreg, NULL }
175
176 #define DEFINE_TRACE(name)                                              \
177         DEFINE_TRACE_FN(name, NULL, NULL);
178
179 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)                              \
180         EXPORT_SYMBOL_GPL(__tracepoint_##name)
181 #define EXPORT_TRACEPOINT_SYMBOL(name)                                  \
182         EXPORT_SYMBOL(__tracepoint_##name)
183
184 #else /* !CONFIG_TRACEPOINTS */
185 #define __DECLARE_TRACE(name, proto, args, data_proto, data_args)       \
186         static inline void trace_##name(proto)                          \
187         { }                                                             \
188         static inline int                                               \
189         register_trace_##name(void (*probe)(data_proto),                \
190                               void *data)                               \
191         {                                                               \
192                 return -ENOSYS;                                         \
193         }                                                               \
194         static inline int                                               \
195         unregister_trace_##name(void (*probe)(data_proto),              \
196                                 void *data)                             \
197         {                                                               \
198                 return -ENOSYS;                                         \
199         }                                                               \
200         static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \
201         {                                                               \
202         }
203
204 #define DEFINE_TRACE_FN(name, reg, unreg)
205 #define DEFINE_TRACE(name)
206 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
207 #define EXPORT_TRACEPOINT_SYMBOL(name)
208
209 #endif /* CONFIG_TRACEPOINTS */
210
211 /*
212  * The need for the DECLARE_TRACE_NOARGS() is to handle the prototype
213  * (void). "void" is a special value in a function prototype and can
214  * not be combined with other arguments. Since the DECLARE_TRACE()
215  * macro adds a data element at the beginning of the prototype,
216  * we need a way to differentiate "(void *data, proto)" from
217  * "(void *data, void)". The second prototype is invalid.
218  *
219  * DECLARE_TRACE_NOARGS() passes "void" as the tracepoint prototype
220  * and "void *__data" as the callback prototype.
221  *
222  * DECLARE_TRACE() passes "proto" as the tracepoint protoype and
223  * "void *__data, proto" as the callback prototype.
224  */
225 #define DECLARE_TRACE_NOARGS(name)                                      \
226                 __DECLARE_TRACE(name, void, , void *__data, __data)
227
228 #define DECLARE_TRACE(name, proto, args)                                \
229                 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args),      \
230                                 PARAMS(void *__data, proto),            \
231                                 PARAMS(__data, args))
232
233 #endif /* DECLARE_TRACE */
234
235 #ifndef TRACE_EVENT
236 /*
237  * For use with the TRACE_EVENT macro:
238  *
239  * We define a tracepoint, its arguments, its printk format
240  * and its 'fast binay record' layout.
241  *
242  * Firstly, name your tracepoint via TRACE_EVENT(name : the
243  * 'subsystem_event' notation is fine.
244  *
245  * Think about this whole construct as the
246  * 'trace_sched_switch() function' from now on.
247  *
248  *
249  *  TRACE_EVENT(sched_switch,
250  *
251  *      *
252  *      * A function has a regular function arguments
253  *      * prototype, declare it via TP_PROTO():
254  *      *
255  *
256  *      TP_PROTO(struct rq *rq, struct task_struct *prev,
257  *               struct task_struct *next),
258  *
259  *      *
260  *      * Define the call signature of the 'function'.
261  *      * (Design sidenote: we use this instead of a
262  *      *  TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
263  *      *
264  *
265  *      TP_ARGS(rq, prev, next),
266  *
267  *      *
268  *      * Fast binary tracing: define the trace record via
269  *      * TP_STRUCT__entry(). You can think about it like a
270  *      * regular C structure local variable definition.
271  *      *
272  *      * This is how the trace record is structured and will
273  *      * be saved into the ring buffer. These are the fields
274  *      * that will be exposed to user-space in
275  *      * /sys/kernel/debug/tracing/events/<*>/format.
276  *      *
277  *      * The declared 'local variable' is called '__entry'
278  *      *
279  *      * __field(pid_t, prev_prid) is equivalent to a standard declariton:
280  *      *
281  *      *       pid_t   prev_pid;
282  *      *
283  *      * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
284  *      *
285  *      *       char    prev_comm[TASK_COMM_LEN];
286  *      *
287  *
288  *      TP_STRUCT__entry(
289  *              __array(        char,   prev_comm,      TASK_COMM_LEN   )
290  *              __field(        pid_t,  prev_pid                        )
291  *              __field(        int,    prev_prio                       )
292  *              __array(        char,   next_comm,      TASK_COMM_LEN   )
293  *              __field(        pid_t,  next_pid                        )
294  *              __field(        int,    next_prio                       )
295  *      ),
296  *
297  *      *
298  *      * Assign the entry into the trace record, by embedding
299  *      * a full C statement block into TP_fast_assign(). You
300  *      * can refer to the trace record as '__entry' -
301  *      * otherwise you can put arbitrary C code in here.
302  *      *
303  *      * Note: this C code will execute every time a trace event
304  *      * happens, on an active tracepoint.
305  *      *
306  *
307  *      TP_fast_assign(
308  *              memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
309  *              __entry->prev_pid       = prev->pid;
310  *              __entry->prev_prio      = prev->prio;
311  *              memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
312  *              __entry->next_pid       = next->pid;
313  *              __entry->next_prio      = next->prio;
314  *      )
315  *
316  *      *
317  *      * Formatted output of a trace record via TP_printk().
318  *      * This is how the tracepoint will appear under ftrace
319  *      * plugins that make use of this tracepoint.
320  *      *
321  *      * (raw-binary tracing wont actually perform this step.)
322  *      *
323  *
324  *      TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
325  *              __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
326  *              __entry->next_comm, __entry->next_pid, __entry->next_prio),
327  *
328  * );
329  *
330  * This macro construct is thus used for the regular printk format
331  * tracing setup, it is used to construct a function pointer based
332  * tracepoint callback (this is used by programmatic plugins and
333  * can also by used by generic instrumentation like SystemTap), and
334  * it is also used to expose a structured trace record in
335  * /sys/kernel/debug/tracing/events/.
336  *
337  * A set of (un)registration functions can be passed to the variant
338  * TRACE_EVENT_FN to perform any (un)registration work.
339  */
340
341 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
342 #define DEFINE_EVENT(template, name, proto, args)               \
343         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
344 #define DEFINE_EVENT_PRINT(template, name, proto, args, print)  \
345         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
346
347 #define TRACE_EVENT(name, proto, args, struct, assign, print)   \
348         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
349 #define TRACE_EVENT_FN(name, proto, args, struct,               \
350                 assign, print, reg, unreg)                      \
351         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
352
353 #endif /* ifdef TRACE_EVENT (see note above) */