tracing: New flag to allow non privileged users to use a trace event
[deliverable/linux.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/errno.h>
18 #include <linux/types.h>
19 #include <linux/rcupdate.h>
20 #include <linux/jump_label.h>
21
22 struct module;
23 struct tracepoint;
24
25 struct tracepoint_func {
26 void *func;
27 void *data;
28 };
29
30 struct tracepoint {
31 const char *name; /* Tracepoint name */
32 int state; /* State. */
33 void (*regfunc)(void);
34 void (*unregfunc)(void);
35 struct tracepoint_func *funcs;
36 } __attribute__((aligned(32))); /*
37 * Aligned on 32 bytes because it is
38 * globally visible and gcc happily
39 * align these on the structure size.
40 * Keep in sync with vmlinux.lds.h.
41 */
42
43 /*
44 * Connect a probe to a tracepoint.
45 * Internal API, should not be used directly.
46 */
47 extern int tracepoint_probe_register(const char *name, void *probe, void *data);
48
49 /*
50 * Disconnect a probe from a tracepoint.
51 * Internal API, should not be used directly.
52 */
53 extern int
54 tracepoint_probe_unregister(const char *name, void *probe, void *data);
55
56 extern int tracepoint_probe_register_noupdate(const char *name, void *probe,
57 void *data);
58 extern int tracepoint_probe_unregister_noupdate(const char *name, void *probe,
59 void *data);
60 extern void tracepoint_probe_update_all(void);
61
62 struct tracepoint_iter {
63 struct module *module;
64 struct tracepoint *tracepoint;
65 };
66
67 extern void tracepoint_iter_start(struct tracepoint_iter *iter);
68 extern void tracepoint_iter_next(struct tracepoint_iter *iter);
69 extern void tracepoint_iter_stop(struct tracepoint_iter *iter);
70 extern void tracepoint_iter_reset(struct tracepoint_iter *iter);
71 extern int tracepoint_get_iter_range(struct tracepoint **tracepoint,
72 struct tracepoint *begin, struct tracepoint *end);
73
74 /*
75 * tracepoint_synchronize_unregister must be called between the last tracepoint
76 * probe unregistration and the end of module exit to make sure there is no
77 * caller executing a probe when it is freed.
78 */
79 static inline void tracepoint_synchronize_unregister(void)
80 {
81 synchronize_sched();
82 }
83
84 #define PARAMS(args...) args
85
86 #ifdef CONFIG_TRACEPOINTS
87 extern void tracepoint_update_probe_range(struct tracepoint *begin,
88 struct tracepoint *end);
89 #else
90 static inline void tracepoint_update_probe_range(struct tracepoint *begin,
91 struct tracepoint *end)
92 { }
93 #endif /* CONFIG_TRACEPOINTS */
94
95 #endif /* _LINUX_TRACEPOINT_H */
96
97 /*
98 * Note: we keep the TRACE_EVENT and DECLARE_TRACE outside the include
99 * file ifdef protection.
100 * This is due to the way trace events work. If a file includes two
101 * trace event headers under one "CREATE_TRACE_POINTS" the first include
102 * will override the TRACE_EVENT and break the second include.
103 */
104
105 #ifndef DECLARE_TRACE
106
107 #define TP_PROTO(args...) args
108 #define TP_ARGS(args...) args
109
110 #ifdef CONFIG_TRACEPOINTS
111
112 /*
113 * it_func[0] is never NULL because there is at least one element in the array
114 * when the array itself is non NULL.
115 *
116 * Note, the proto and args passed in includes "__data" as the first parameter.
117 * The reason for this is to handle the "void" prototype. If a tracepoint
118 * has a "void" prototype, then it is invalid to declare a function
119 * as "(void *, void)". The DECLARE_TRACE_NOARGS() will pass in just
120 * "void *data", where as the DECLARE_TRACE() will pass in "void *data, proto".
121 */
122 #define __DO_TRACE(tp, proto, args) \
123 do { \
124 struct tracepoint_func *it_func_ptr; \
125 void *it_func; \
126 void *__data; \
127 \
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); \
136 } \
137 rcu_read_unlock_sched_notrace(); \
138 } while (0)
139
140 /*
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.
144 */
145 #define __DECLARE_TRACE(name, proto, args, data_proto, data_args) \
146 extern struct tracepoint __tracepoint_##name; \
147 static inline void trace_##name(proto) \
148 { \
149 JUMP_LABEL(&__tracepoint_##name.state, do_trace); \
150 return; \
151 do_trace: \
152 __DO_TRACE(&__tracepoint_##name, \
153 TP_PROTO(data_proto), \
154 TP_ARGS(data_args)); \
155 } \
156 static inline int \
157 register_trace_##name(void (*probe)(data_proto), void *data) \
158 { \
159 return tracepoint_probe_register(#name, (void *)probe, \
160 data); \
161 } \
162 static inline int \
163 unregister_trace_##name(void (*probe)(data_proto), void *data) \
164 { \
165 return tracepoint_probe_unregister(#name, (void *)probe, \
166 data); \
167 } \
168 static inline void \
169 check_trace_callback_type_##name(void (*cb)(data_proto)) \
170 { \
171 }
172
173 #define DEFINE_TRACE_FN(name, reg, unreg) \
174 static const char __tpstrtab_##name[] \
175 __attribute__((section("__tracepoints_strings"))) = #name; \
176 struct tracepoint __tracepoint_##name \
177 __attribute__((section("__tracepoints"), aligned(32))) = \
178 { __tpstrtab_##name, 0, reg, unreg, NULL }
179
180 #define DEFINE_TRACE(name) \
181 DEFINE_TRACE_FN(name, NULL, NULL);
182
183 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \
184 EXPORT_SYMBOL_GPL(__tracepoint_##name)
185 #define EXPORT_TRACEPOINT_SYMBOL(name) \
186 EXPORT_SYMBOL(__tracepoint_##name)
187
188 #else /* !CONFIG_TRACEPOINTS */
189 #define __DECLARE_TRACE(name, proto, args, data_proto, data_args) \
190 static inline void trace_##name(proto) \
191 { } \
192 static inline int \
193 register_trace_##name(void (*probe)(data_proto), \
194 void *data) \
195 { \
196 return -ENOSYS; \
197 } \
198 static inline int \
199 unregister_trace_##name(void (*probe)(data_proto), \
200 void *data) \
201 { \
202 return -ENOSYS; \
203 } \
204 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \
205 { \
206 }
207
208 #define DEFINE_TRACE_FN(name, reg, unreg)
209 #define DEFINE_TRACE(name)
210 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
211 #define EXPORT_TRACEPOINT_SYMBOL(name)
212
213 #endif /* CONFIG_TRACEPOINTS */
214
215 /*
216 * The need for the DECLARE_TRACE_NOARGS() is to handle the prototype
217 * (void). "void" is a special value in a function prototype and can
218 * not be combined with other arguments. Since the DECLARE_TRACE()
219 * macro adds a data element at the beginning of the prototype,
220 * we need a way to differentiate "(void *data, proto)" from
221 * "(void *data, void)". The second prototype is invalid.
222 *
223 * DECLARE_TRACE_NOARGS() passes "void" as the tracepoint prototype
224 * and "void *__data" as the callback prototype.
225 *
226 * DECLARE_TRACE() passes "proto" as the tracepoint protoype and
227 * "void *__data, proto" as the callback prototype.
228 */
229 #define DECLARE_TRACE_NOARGS(name) \
230 __DECLARE_TRACE(name, void, , void *__data, __data)
231
232 #define DECLARE_TRACE(name, proto, args) \
233 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \
234 PARAMS(void *__data, proto), \
235 PARAMS(__data, args))
236
237 #endif /* DECLARE_TRACE */
238
239 #ifndef TRACE_EVENT
240 /*
241 * For use with the TRACE_EVENT macro:
242 *
243 * We define a tracepoint, its arguments, its printk format
244 * and its 'fast binay record' layout.
245 *
246 * Firstly, name your tracepoint via TRACE_EVENT(name : the
247 * 'subsystem_event' notation is fine.
248 *
249 * Think about this whole construct as the
250 * 'trace_sched_switch() function' from now on.
251 *
252 *
253 * TRACE_EVENT(sched_switch,
254 *
255 * *
256 * * A function has a regular function arguments
257 * * prototype, declare it via TP_PROTO():
258 * *
259 *
260 * TP_PROTO(struct rq *rq, struct task_struct *prev,
261 * struct task_struct *next),
262 *
263 * *
264 * * Define the call signature of the 'function'.
265 * * (Design sidenote: we use this instead of a
266 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
267 * *
268 *
269 * TP_ARGS(rq, prev, next),
270 *
271 * *
272 * * Fast binary tracing: define the trace record via
273 * * TP_STRUCT__entry(). You can think about it like a
274 * * regular C structure local variable definition.
275 * *
276 * * This is how the trace record is structured and will
277 * * be saved into the ring buffer. These are the fields
278 * * that will be exposed to user-space in
279 * * /sys/kernel/debug/tracing/events/<*>/format.
280 * *
281 * * The declared 'local variable' is called '__entry'
282 * *
283 * * __field(pid_t, prev_prid) is equivalent to a standard declariton:
284 * *
285 * * pid_t prev_pid;
286 * *
287 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
288 * *
289 * * char prev_comm[TASK_COMM_LEN];
290 * *
291 *
292 * TP_STRUCT__entry(
293 * __array( char, prev_comm, TASK_COMM_LEN )
294 * __field( pid_t, prev_pid )
295 * __field( int, prev_prio )
296 * __array( char, next_comm, TASK_COMM_LEN )
297 * __field( pid_t, next_pid )
298 * __field( int, next_prio )
299 * ),
300 *
301 * *
302 * * Assign the entry into the trace record, by embedding
303 * * a full C statement block into TP_fast_assign(). You
304 * * can refer to the trace record as '__entry' -
305 * * otherwise you can put arbitrary C code in here.
306 * *
307 * * Note: this C code will execute every time a trace event
308 * * happens, on an active tracepoint.
309 * *
310 *
311 * TP_fast_assign(
312 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
313 * __entry->prev_pid = prev->pid;
314 * __entry->prev_prio = prev->prio;
315 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
316 * __entry->next_pid = next->pid;
317 * __entry->next_prio = next->prio;
318 * )
319 *
320 * *
321 * * Formatted output of a trace record via TP_printk().
322 * * This is how the tracepoint will appear under ftrace
323 * * plugins that make use of this tracepoint.
324 * *
325 * * (raw-binary tracing wont actually perform this step.)
326 * *
327 *
328 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
329 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
330 * __entry->next_comm, __entry->next_pid, __entry->next_prio),
331 *
332 * );
333 *
334 * This macro construct is thus used for the regular printk format
335 * tracing setup, it is used to construct a function pointer based
336 * tracepoint callback (this is used by programmatic plugins and
337 * can also by used by generic instrumentation like SystemTap), and
338 * it is also used to expose a structured trace record in
339 * /sys/kernel/debug/tracing/events/.
340 *
341 * A set of (un)registration functions can be passed to the variant
342 * TRACE_EVENT_FN to perform any (un)registration work.
343 */
344
345 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
346 #define DEFINE_EVENT(template, name, proto, args) \
347 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
348 #define DEFINE_EVENT_PRINT(template, name, proto, args, print) \
349 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
350
351 #define TRACE_EVENT(name, proto, args, struct, assign, print) \
352 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
353 #define TRACE_EVENT_FN(name, proto, args, struct, \
354 assign, print, reg, unreg) \
355 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
356
357 #endif /* ifdef TRACE_EVENT (see note above) */
This page took 0.047624 seconds and 6 git commands to generate.