Merge branch 'sched/urgent' into sched/core
[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 * Copyright (C) 2008-2014 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
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/static_key.h>
21
22 struct module;
23 struct tracepoint;
24 struct notifier_block;
25
26 struct tracepoint_func {
27 void *func;
28 void *data;
29 };
30
31 struct tracepoint {
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;
37 };
38
39 struct trace_enum_map {
40 const char *system;
41 const char *enum_string;
42 unsigned long enum_value;
43 };
44
45 extern int
46 tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data);
47 extern int
48 tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data);
49 extern void
50 for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
51 void *priv);
52
53 #ifdef CONFIG_MODULES
54 struct tp_module {
55 struct list_head list;
56 struct module *mod;
57 };
58
59 bool trace_module_has_bad_taint(struct module *mod);
60 extern int register_tracepoint_module_notifier(struct notifier_block *nb);
61 extern int unregister_tracepoint_module_notifier(struct notifier_block *nb);
62 #else
63 static inline bool trace_module_has_bad_taint(struct module *mod)
64 {
65 return false;
66 }
67 static inline
68 int register_tracepoint_module_notifier(struct notifier_block *nb)
69 {
70 return 0;
71 }
72 static inline
73 int unregister_tracepoint_module_notifier(struct notifier_block *nb)
74 {
75 return 0;
76 }
77 #endif /* CONFIG_MODULES */
78
79 /*
80 * tracepoint_synchronize_unregister must be called between the last tracepoint
81 * probe unregistration and the end of module exit to make sure there is no
82 * caller executing a probe when it is freed.
83 */
84 static inline void tracepoint_synchronize_unregister(void)
85 {
86 synchronize_sched();
87 }
88
89 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
90 extern void syscall_regfunc(void);
91 extern void syscall_unregfunc(void);
92 #endif /* CONFIG_HAVE_SYSCALL_TRACEPOINTS */
93
94 #define PARAMS(args...) args
95
96 #define TRACE_DEFINE_ENUM(x)
97
98 #endif /* _LINUX_TRACEPOINT_H */
99
100 /*
101 * Note: we keep the TRACE_EVENT and DECLARE_TRACE outside the include
102 * file ifdef protection.
103 * This is due to the way trace events work. If a file includes two
104 * trace event headers under one "CREATE_TRACE_POINTS" the first include
105 * will override the TRACE_EVENT and break the second include.
106 */
107
108 #ifndef DECLARE_TRACE
109
110 #define TP_PROTO(args...) args
111 #define TP_ARGS(args...) args
112 #define TP_CONDITION(args...) args
113
114 #ifdef CONFIG_TRACEPOINTS
115
116 /*
117 * it_func[0] is never NULL because there is at least one element in the array
118 * when the array itself is non NULL.
119 *
120 * Note, the proto and args passed in includes "__data" as the first parameter.
121 * The reason for this is to handle the "void" prototype. If a tracepoint
122 * has a "void" prototype, then it is invalid to declare a function
123 * as "(void *, void)". The DECLARE_TRACE_NOARGS() will pass in just
124 * "void *data", where as the DECLARE_TRACE() will pass in "void *data, proto".
125 */
126 #define __DO_TRACE(tp, proto, args, cond, prercu, postrcu) \
127 do { \
128 struct tracepoint_func *it_func_ptr; \
129 void *it_func; \
130 void *__data; \
131 \
132 if (!(cond)) \
133 return; \
134 prercu; \
135 rcu_read_lock_sched_notrace(); \
136 it_func_ptr = rcu_dereference_sched((tp)->funcs); \
137 if (it_func_ptr) { \
138 do { \
139 it_func = (it_func_ptr)->func; \
140 __data = (it_func_ptr)->data; \
141 ((void(*)(proto))(it_func))(args); \
142 } while ((++it_func_ptr)->func); \
143 } \
144 rcu_read_unlock_sched_notrace(); \
145 postrcu; \
146 } while (0)
147
148 #ifndef MODULE
149 #define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args) \
150 static inline void trace_##name##_rcuidle(proto) \
151 { \
152 if (static_key_false(&__tracepoint_##name.key)) \
153 __DO_TRACE(&__tracepoint_##name, \
154 TP_PROTO(data_proto), \
155 TP_ARGS(data_args), \
156 TP_CONDITION(cond), \
157 rcu_irq_enter(), \
158 rcu_irq_exit()); \
159 }
160 #else
161 #define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args)
162 #endif
163
164 /*
165 * Make sure the alignment of the structure in the __tracepoints section will
166 * not add unwanted padding between the beginning of the section and the
167 * structure. Force alignment to the same alignment as the section start.
168 *
169 * When lockdep is enabled, we make sure to always do the RCU portions of
170 * the tracepoint code, regardless of whether tracing is on or we match the
171 * condition. This lets us find RCU issues triggered with tracepoints even
172 * when this tracepoint is off. This code has no purpose other than poking
173 * RCU a bit.
174 */
175 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
176 extern struct tracepoint __tracepoint_##name; \
177 static inline void trace_##name(proto) \
178 { \
179 if (static_key_false(&__tracepoint_##name.key)) \
180 __DO_TRACE(&__tracepoint_##name, \
181 TP_PROTO(data_proto), \
182 TP_ARGS(data_args), \
183 TP_CONDITION(cond),,); \
184 if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) { \
185 rcu_read_lock_sched_notrace(); \
186 rcu_dereference_sched(__tracepoint_##name.funcs);\
187 rcu_read_unlock_sched_notrace(); \
188 } \
189 } \
190 __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args), \
191 PARAMS(cond), PARAMS(data_proto), PARAMS(data_args)) \
192 static inline int \
193 register_trace_##name(void (*probe)(data_proto), void *data) \
194 { \
195 return tracepoint_probe_register(&__tracepoint_##name, \
196 (void *)probe, data); \
197 } \
198 static inline int \
199 unregister_trace_##name(void (*probe)(data_proto), void *data) \
200 { \
201 return tracepoint_probe_unregister(&__tracepoint_##name,\
202 (void *)probe, data); \
203 } \
204 static inline void \
205 check_trace_callback_type_##name(void (*cb)(data_proto)) \
206 { \
207 } \
208 static inline bool \
209 trace_##name##_enabled(void) \
210 { \
211 return static_key_false(&__tracepoint_##name.key); \
212 }
213
214 /*
215 * We have no guarantee that gcc and the linker won't up-align the tracepoint
216 * structures, so we create an array of pointers that will be used for iteration
217 * on the tracepoints.
218 */
219 #define DEFINE_TRACE_FN(name, reg, unreg) \
220 static const char __tpstrtab_##name[] \
221 __attribute__((section("__tracepoints_strings"))) = #name; \
222 struct tracepoint __tracepoint_##name \
223 __attribute__((section("__tracepoints"))) = \
224 { __tpstrtab_##name, STATIC_KEY_INIT_FALSE, reg, unreg, NULL };\
225 static struct tracepoint * const __tracepoint_ptr_##name __used \
226 __attribute__((section("__tracepoints_ptrs"))) = \
227 &__tracepoint_##name;
228
229 #define DEFINE_TRACE(name) \
230 DEFINE_TRACE_FN(name, NULL, NULL);
231
232 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \
233 EXPORT_SYMBOL_GPL(__tracepoint_##name)
234 #define EXPORT_TRACEPOINT_SYMBOL(name) \
235 EXPORT_SYMBOL(__tracepoint_##name)
236
237 #else /* !CONFIG_TRACEPOINTS */
238 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
239 static inline void trace_##name(proto) \
240 { } \
241 static inline void trace_##name##_rcuidle(proto) \
242 { } \
243 static inline int \
244 register_trace_##name(void (*probe)(data_proto), \
245 void *data) \
246 { \
247 return -ENOSYS; \
248 } \
249 static inline int \
250 unregister_trace_##name(void (*probe)(data_proto), \
251 void *data) \
252 { \
253 return -ENOSYS; \
254 } \
255 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \
256 { \
257 } \
258 static inline bool \
259 trace_##name##_enabled(void) \
260 { \
261 return false; \
262 }
263
264 #define DEFINE_TRACE_FN(name, reg, unreg)
265 #define DEFINE_TRACE(name)
266 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
267 #define EXPORT_TRACEPOINT_SYMBOL(name)
268
269 #endif /* CONFIG_TRACEPOINTS */
270
271 #ifdef CONFIG_TRACING
272 /**
273 * tracepoint_string - register constant persistent string to trace system
274 * @str - a constant persistent string that will be referenced in tracepoints
275 *
276 * If constant strings are being used in tracepoints, it is faster and
277 * more efficient to just save the pointer to the string and reference
278 * that with a printf "%s" instead of saving the string in the ring buffer
279 * and wasting space and time.
280 *
281 * The problem with the above approach is that userspace tools that read
282 * the binary output of the trace buffers do not have access to the string.
283 * Instead they just show the address of the string which is not very
284 * useful to users.
285 *
286 * With tracepoint_string(), the string will be registered to the tracing
287 * system and exported to userspace via the debugfs/tracing/printk_formats
288 * file that maps the string address to the string text. This way userspace
289 * tools that read the binary buffers have a way to map the pointers to
290 * the ASCII strings they represent.
291 *
292 * The @str used must be a constant string and persistent as it would not
293 * make sense to show a string that no longer exists. But it is still fine
294 * to be used with modules, because when modules are unloaded, if they
295 * had tracepoints, the ring buffers are cleared too. As long as the string
296 * does not change during the life of the module, it is fine to use
297 * tracepoint_string() within a module.
298 */
299 #define tracepoint_string(str) \
300 ({ \
301 static const char *___tp_str __tracepoint_string = str; \
302 ___tp_str; \
303 })
304 #define __tracepoint_string __attribute__((section("__tracepoint_str")))
305 #else
306 /*
307 * tracepoint_string() is used to save the string address for userspace
308 * tracing tools. When tracing isn't configured, there's no need to save
309 * anything.
310 */
311 # define tracepoint_string(str) str
312 # define __tracepoint_string
313 #endif
314
315 /*
316 * The need for the DECLARE_TRACE_NOARGS() is to handle the prototype
317 * (void). "void" is a special value in a function prototype and can
318 * not be combined with other arguments. Since the DECLARE_TRACE()
319 * macro adds a data element at the beginning of the prototype,
320 * we need a way to differentiate "(void *data, proto)" from
321 * "(void *data, void)". The second prototype is invalid.
322 *
323 * DECLARE_TRACE_NOARGS() passes "void" as the tracepoint prototype
324 * and "void *__data" as the callback prototype.
325 *
326 * DECLARE_TRACE() passes "proto" as the tracepoint protoype and
327 * "void *__data, proto" as the callback prototype.
328 */
329 #define DECLARE_TRACE_NOARGS(name) \
330 __DECLARE_TRACE(name, void, , 1, void *__data, __data)
331
332 #define DECLARE_TRACE(name, proto, args) \
333 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), 1, \
334 PARAMS(void *__data, proto), \
335 PARAMS(__data, args))
336
337 #define DECLARE_TRACE_CONDITION(name, proto, args, cond) \
338 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), PARAMS(cond), \
339 PARAMS(void *__data, proto), \
340 PARAMS(__data, args))
341
342 #define TRACE_EVENT_FLAGS(event, flag)
343
344 #define TRACE_EVENT_PERF_PERM(event, expr...)
345
346 #endif /* DECLARE_TRACE */
347
348 #ifndef TRACE_EVENT
349 /*
350 * For use with the TRACE_EVENT macro:
351 *
352 * We define a tracepoint, its arguments, its printk format
353 * and its 'fast binary record' layout.
354 *
355 * Firstly, name your tracepoint via TRACE_EVENT(name : the
356 * 'subsystem_event' notation is fine.
357 *
358 * Think about this whole construct as the
359 * 'trace_sched_switch() function' from now on.
360 *
361 *
362 * TRACE_EVENT(sched_switch,
363 *
364 * *
365 * * A function has a regular function arguments
366 * * prototype, declare it via TP_PROTO():
367 * *
368 *
369 * TP_PROTO(struct rq *rq, struct task_struct *prev,
370 * struct task_struct *next),
371 *
372 * *
373 * * Define the call signature of the 'function'.
374 * * (Design sidenote: we use this instead of a
375 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
376 * *
377 *
378 * TP_ARGS(rq, prev, next),
379 *
380 * *
381 * * Fast binary tracing: define the trace record via
382 * * TP_STRUCT__entry(). You can think about it like a
383 * * regular C structure local variable definition.
384 * *
385 * * This is how the trace record is structured and will
386 * * be saved into the ring buffer. These are the fields
387 * * that will be exposed to user-space in
388 * * /sys/kernel/debug/tracing/events/<*>/format.
389 * *
390 * * The declared 'local variable' is called '__entry'
391 * *
392 * * __field(pid_t, prev_prid) is equivalent to a standard declariton:
393 * *
394 * * pid_t prev_pid;
395 * *
396 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
397 * *
398 * * char prev_comm[TASK_COMM_LEN];
399 * *
400 *
401 * TP_STRUCT__entry(
402 * __array( char, prev_comm, TASK_COMM_LEN )
403 * __field( pid_t, prev_pid )
404 * __field( int, prev_prio )
405 * __array( char, next_comm, TASK_COMM_LEN )
406 * __field( pid_t, next_pid )
407 * __field( int, next_prio )
408 * ),
409 *
410 * *
411 * * Assign the entry into the trace record, by embedding
412 * * a full C statement block into TP_fast_assign(). You
413 * * can refer to the trace record as '__entry' -
414 * * otherwise you can put arbitrary C code in here.
415 * *
416 * * Note: this C code will execute every time a trace event
417 * * happens, on an active tracepoint.
418 * *
419 *
420 * TP_fast_assign(
421 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
422 * __entry->prev_pid = prev->pid;
423 * __entry->prev_prio = prev->prio;
424 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
425 * __entry->next_pid = next->pid;
426 * __entry->next_prio = next->prio;
427 * ),
428 *
429 * *
430 * * Formatted output of a trace record via TP_printk().
431 * * This is how the tracepoint will appear under ftrace
432 * * plugins that make use of this tracepoint.
433 * *
434 * * (raw-binary tracing wont actually perform this step.)
435 * *
436 *
437 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
438 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
439 * __entry->next_comm, __entry->next_pid, __entry->next_prio),
440 *
441 * );
442 *
443 * This macro construct is thus used for the regular printk format
444 * tracing setup, it is used to construct a function pointer based
445 * tracepoint callback (this is used by programmatic plugins and
446 * can also by used by generic instrumentation like SystemTap), and
447 * it is also used to expose a structured trace record in
448 * /sys/kernel/debug/tracing/events/.
449 *
450 * A set of (un)registration functions can be passed to the variant
451 * TRACE_EVENT_FN to perform any (un)registration work.
452 */
453
454 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
455 #define DEFINE_EVENT(template, name, proto, args) \
456 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
457 #define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\
458 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
459 #define DEFINE_EVENT_PRINT(template, name, proto, args, print) \
460 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
461 #define DEFINE_EVENT_CONDITION(template, name, proto, \
462 args, cond) \
463 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
464 PARAMS(args), PARAMS(cond))
465
466 #define TRACE_EVENT(name, proto, args, struct, assign, print) \
467 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
468 #define TRACE_EVENT_FN(name, proto, args, struct, \
469 assign, print, reg, unreg) \
470 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
471 #define TRACE_EVENT_CONDITION(name, proto, args, cond, \
472 struct, assign, print) \
473 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
474 PARAMS(args), PARAMS(cond))
475
476 #define TRACE_EVENT_FLAGS(event, flag)
477
478 #define TRACE_EVENT_PERF_PERM(event, expr...)
479
480 #endif /* ifdef TRACE_EVENT (see note above) */
This page took 0.064688 seconds and 6 git commands to generate.