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/smp.h>
18 #include <linux/errno.h>
19 #include <linux/types.h>
20 #include <linux/cpumask.h>
21 #include <linux/rcupdate.h>
22 #include <linux/static_key.h>
23
24 struct module;
25 struct tracepoint;
26 struct notifier_block;
27
28 struct tracepoint_func {
29 void *func;
30 void *data;
31 int prio;
32 };
33
34 struct tracepoint {
35 const char *name; /* Tracepoint name */
36 struct static_key key;
37 void (*regfunc)(void);
38 void (*unregfunc)(void);
39 struct tracepoint_func __rcu *funcs;
40 };
41
42 struct trace_enum_map {
43 const char *system;
44 const char *enum_string;
45 unsigned long enum_value;
46 };
47
48 #define TRACEPOINT_DEFAULT_PRIO 10
49
50 extern int
51 tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data);
52 extern int
53 tracepoint_probe_register_prio(struct tracepoint *tp, void *probe, void *data,
54 int prio);
55 extern int
56 tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data);
57 extern void
58 for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
59 void *priv);
60
61 #ifdef CONFIG_MODULES
62 struct tp_module {
63 struct list_head list;
64 struct module *mod;
65 };
66
67 bool trace_module_has_bad_taint(struct module *mod);
68 extern int register_tracepoint_module_notifier(struct notifier_block *nb);
69 extern int unregister_tracepoint_module_notifier(struct notifier_block *nb);
70 #else
trace_module_has_bad_taint(struct module * mod)71 static inline bool trace_module_has_bad_taint(struct module *mod)
72 {
73 return false;
74 }
75 static inline
register_tracepoint_module_notifier(struct notifier_block * nb)76 int register_tracepoint_module_notifier(struct notifier_block *nb)
77 {
78 return 0;
79 }
80 static inline
unregister_tracepoint_module_notifier(struct notifier_block * nb)81 int unregister_tracepoint_module_notifier(struct notifier_block *nb)
82 {
83 return 0;
84 }
85 #endif /* CONFIG_MODULES */
86
87 /*
88 * tracepoint_synchronize_unregister must be called between the last tracepoint
89 * probe unregistration and the end of module exit to make sure there is no
90 * caller executing a probe when it is freed.
91 */
tracepoint_synchronize_unregister(void)92 static inline void tracepoint_synchronize_unregister(void)
93 {
94 synchronize_sched();
95 }
96
97 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
98 extern void syscall_regfunc(void);
99 extern void syscall_unregfunc(void);
100 #endif /* CONFIG_HAVE_SYSCALL_TRACEPOINTS */
101
102 #define PARAMS(args...) args
103
104 #define TRACE_DEFINE_ENUM(x)
105
106 #endif /* _LINUX_TRACEPOINT_H */
107
108 /*
109 * Note: we keep the TRACE_EVENT and DECLARE_TRACE outside the include
110 * file ifdef protection.
111 * This is due to the way trace events work. If a file includes two
112 * trace event headers under one "CREATE_TRACE_POINTS" the first include
113 * will override the TRACE_EVENT and break the second include.
114 */
115
116 #ifndef DECLARE_TRACE
117
118 #define TP_PROTO(args...) args
119 #define TP_ARGS(args...) args
120 #define TP_CONDITION(args...) args
121
122 /*
123 * Individual subsystem my have a separate configuration to
124 * enable their tracepoints. By default, this file will create
125 * the tracepoints if CONFIG_TRACEPOINT is defined. If a subsystem
126 * wants to be able to disable its tracepoints from being created
127 * it can define NOTRACE before including the tracepoint headers.
128 */
129 #if defined(CONFIG_TRACEPOINTS) && !defined(NOTRACE)
130 #define TRACEPOINTS_ENABLED
131 #endif
132
133 #ifdef TRACEPOINTS_ENABLED
134
135 /*
136 * it_func[0] is never NULL because there is at least one element in the array
137 * when the array itself is non NULL.
138 *
139 * Note, the proto and args passed in includes "__data" as the first parameter.
140 * The reason for this is to handle the "void" prototype. If a tracepoint
141 * has a "void" prototype, then it is invalid to declare a function
142 * as "(void *, void)". The DECLARE_TRACE_NOARGS() will pass in just
143 * "void *data", where as the DECLARE_TRACE() will pass in "void *data, proto".
144 */
145 #define __DO_TRACE(tp, proto, args, cond, prercu, postrcu) \
146 do { \
147 struct tracepoint_func *it_func_ptr; \
148 void *it_func; \
149 void *__data; \
150 \
151 if (!(cond)) \
152 return; \
153 prercu; \
154 rcu_read_lock_sched_notrace(); \
155 it_func_ptr = rcu_dereference_sched((tp)->funcs); \
156 if (it_func_ptr) { \
157 do { \
158 it_func = (it_func_ptr)->func; \
159 __data = (it_func_ptr)->data; \
160 ((void(*)(proto))(it_func))(args); \
161 } while ((++it_func_ptr)->func); \
162 } \
163 rcu_read_unlock_sched_notrace(); \
164 postrcu; \
165 } while (0)
166
167 #ifndef MODULE
168 #define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args) \
169 static inline void trace_##name##_rcuidle(proto) \
170 { \
171 if (static_key_false(&__tracepoint_##name.key)) \
172 __DO_TRACE(&__tracepoint_##name, \
173 TP_PROTO(data_proto), \
174 TP_ARGS(data_args), \
175 TP_CONDITION(cond), \
176 rcu_irq_enter(), \
177 rcu_irq_exit()); \
178 }
179 #else
180 #define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args)
181 #endif
182
183 /*
184 * Make sure the alignment of the structure in the __tracepoints section will
185 * not add unwanted padding between the beginning of the section and the
186 * structure. Force alignment to the same alignment as the section start.
187 *
188 * When lockdep is enabled, we make sure to always do the RCU portions of
189 * the tracepoint code, regardless of whether tracing is on. However,
190 * don't check if the condition is false, due to interaction with idle
191 * instrumentation. This lets us find RCU issues triggered with tracepoints
192 * even when this tracepoint is off. This code has no purpose other than
193 * poking RCU a bit.
194 */
195 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
196 extern struct tracepoint __tracepoint_##name; \
197 static inline void trace_##name(proto) \
198 { \
199 if (static_key_false(&__tracepoint_##name.key)) \
200 __DO_TRACE(&__tracepoint_##name, \
201 TP_PROTO(data_proto), \
202 TP_ARGS(data_args), \
203 TP_CONDITION(cond),,); \
204 if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) { \
205 rcu_read_lock_sched_notrace(); \
206 rcu_dereference_sched(__tracepoint_##name.funcs);\
207 rcu_read_unlock_sched_notrace(); \
208 } \
209 } \
210 __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args), \
211 PARAMS(cond), PARAMS(data_proto), PARAMS(data_args)) \
212 static inline int \
213 register_trace_##name(void (*probe)(data_proto), void *data) \
214 { \
215 return tracepoint_probe_register(&__tracepoint_##name, \
216 (void *)probe, data); \
217 } \
218 static inline int \
219 register_trace_prio_##name(void (*probe)(data_proto), void *data,\
220 int prio) \
221 { \
222 return tracepoint_probe_register_prio(&__tracepoint_##name, \
223 (void *)probe, data, prio); \
224 } \
225 static inline int \
226 unregister_trace_##name(void (*probe)(data_proto), void *data) \
227 { \
228 return tracepoint_probe_unregister(&__tracepoint_##name,\
229 (void *)probe, data); \
230 } \
231 static inline void \
232 check_trace_callback_type_##name(void (*cb)(data_proto)) \
233 { \
234 } \
235 static inline bool \
236 trace_##name##_enabled(void) \
237 { \
238 return static_key_false(&__tracepoint_##name.key); \
239 }
240
241 /*
242 * We have no guarantee that gcc and the linker won't up-align the tracepoint
243 * structures, so we create an array of pointers that will be used for iteration
244 * on the tracepoints.
245 */
246 #define DEFINE_TRACE_FN(name, reg, unreg) \
247 static const char __tpstrtab_##name[] \
248 __attribute__((section("__tracepoints_strings"))) = #name; \
249 struct tracepoint __tracepoint_##name \
250 __attribute__((section("__tracepoints"))) = \
251 { __tpstrtab_##name, STATIC_KEY_INIT_FALSE, reg, unreg, NULL };\
252 static struct tracepoint * const __tracepoint_ptr_##name __used \
253 __attribute__((section("__tracepoints_ptrs"))) = \
254 &__tracepoint_##name;
255
256 #define DEFINE_TRACE(name) \
257 DEFINE_TRACE_FN(name, NULL, NULL);
258
259 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \
260 EXPORT_SYMBOL_GPL(__tracepoint_##name)
261 #define EXPORT_TRACEPOINT_SYMBOL(name) \
262 EXPORT_SYMBOL(__tracepoint_##name)
263
264 #else /* !TRACEPOINTS_ENABLED */
265 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
266 static inline void trace_##name(proto) \
267 { } \
268 static inline void trace_##name##_rcuidle(proto) \
269 { } \
270 static inline int \
271 register_trace_##name(void (*probe)(data_proto), \
272 void *data) \
273 { \
274 return -ENOSYS; \
275 } \
276 static inline int \
277 unregister_trace_##name(void (*probe)(data_proto), \
278 void *data) \
279 { \
280 return -ENOSYS; \
281 } \
282 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \
283 { \
284 } \
285 static inline bool \
286 trace_##name##_enabled(void) \
287 { \
288 return false; \
289 }
290
291 #define DEFINE_TRACE_FN(name, reg, unreg)
292 #define DEFINE_TRACE(name)
293 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
294 #define EXPORT_TRACEPOINT_SYMBOL(name)
295
296 #endif /* TRACEPOINTS_ENABLED */
297
298 #ifdef CONFIG_TRACING
299 /**
300 * tracepoint_string - register constant persistent string to trace system
301 * @str - a constant persistent string that will be referenced in tracepoints
302 *
303 * If constant strings are being used in tracepoints, it is faster and
304 * more efficient to just save the pointer to the string and reference
305 * that with a printf "%s" instead of saving the string in the ring buffer
306 * and wasting space and time.
307 *
308 * The problem with the above approach is that userspace tools that read
309 * the binary output of the trace buffers do not have access to the string.
310 * Instead they just show the address of the string which is not very
311 * useful to users.
312 *
313 * With tracepoint_string(), the string will be registered to the tracing
314 * system and exported to userspace via the debugfs/tracing/printk_formats
315 * file that maps the string address to the string text. This way userspace
316 * tools that read the binary buffers have a way to map the pointers to
317 * the ASCII strings they represent.
318 *
319 * The @str used must be a constant string and persistent as it would not
320 * make sense to show a string that no longer exists. But it is still fine
321 * to be used with modules, because when modules are unloaded, if they
322 * had tracepoints, the ring buffers are cleared too. As long as the string
323 * does not change during the life of the module, it is fine to use
324 * tracepoint_string() within a module.
325 */
326 #define tracepoint_string(str) \
327 ({ \
328 static const char *___tp_str __tracepoint_string = str; \
329 ___tp_str; \
330 })
331 #define __tracepoint_string __attribute__((section("__tracepoint_str"), used))
332 #else
333 /*
334 * tracepoint_string() is used to save the string address for userspace
335 * tracing tools. When tracing isn't configured, there's no need to save
336 * anything.
337 */
338 # define tracepoint_string(str) str
339 # define __tracepoint_string
340 #endif
341
342 /*
343 * The need for the DECLARE_TRACE_NOARGS() is to handle the prototype
344 * (void). "void" is a special value in a function prototype and can
345 * not be combined with other arguments. Since the DECLARE_TRACE()
346 * macro adds a data element at the beginning of the prototype,
347 * we need a way to differentiate "(void *data, proto)" from
348 * "(void *data, void)". The second prototype is invalid.
349 *
350 * DECLARE_TRACE_NOARGS() passes "void" as the tracepoint prototype
351 * and "void *__data" as the callback prototype.
352 *
353 * DECLARE_TRACE() passes "proto" as the tracepoint protoype and
354 * "void *__data, proto" as the callback prototype.
355 */
356 #define DECLARE_TRACE_NOARGS(name) \
357 __DECLARE_TRACE(name, void, , \
358 cpu_online(raw_smp_processor_id()), \
359 void *__data, __data)
360
361 #define DECLARE_TRACE(name, proto, args) \
362 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \
363 cpu_online(raw_smp_processor_id()), \
364 PARAMS(void *__data, proto), \
365 PARAMS(__data, args))
366
367 #define DECLARE_TRACE_CONDITION(name, proto, args, cond) \
368 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \
369 cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \
370 PARAMS(void *__data, proto), \
371 PARAMS(__data, args))
372
373 #define TRACE_EVENT_FLAGS(event, flag)
374
375 #define TRACE_EVENT_PERF_PERM(event, expr...)
376
377 #endif /* DECLARE_TRACE */
378
379 #ifndef TRACE_EVENT
380 /*
381 * For use with the TRACE_EVENT macro:
382 *
383 * We define a tracepoint, its arguments, its printk format
384 * and its 'fast binary record' layout.
385 *
386 * Firstly, name your tracepoint via TRACE_EVENT(name : the
387 * 'subsystem_event' notation is fine.
388 *
389 * Think about this whole construct as the
390 * 'trace_sched_switch() function' from now on.
391 *
392 *
393 * TRACE_EVENT(sched_switch,
394 *
395 * *
396 * * A function has a regular function arguments
397 * * prototype, declare it via TP_PROTO():
398 * *
399 *
400 * TP_PROTO(struct rq *rq, struct task_struct *prev,
401 * struct task_struct *next),
402 *
403 * *
404 * * Define the call signature of the 'function'.
405 * * (Design sidenote: we use this instead of a
406 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
407 * *
408 *
409 * TP_ARGS(rq, prev, next),
410 *
411 * *
412 * * Fast binary tracing: define the trace record via
413 * * TP_STRUCT__entry(). You can think about it like a
414 * * regular C structure local variable definition.
415 * *
416 * * This is how the trace record is structured and will
417 * * be saved into the ring buffer. These are the fields
418 * * that will be exposed to user-space in
419 * * /sys/kernel/debug/tracing/events/<*>/format.
420 * *
421 * * The declared 'local variable' is called '__entry'
422 * *
423 * * __field(pid_t, prev_prid) is equivalent to a standard declariton:
424 * *
425 * * pid_t prev_pid;
426 * *
427 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
428 * *
429 * * char prev_comm[TASK_COMM_LEN];
430 * *
431 *
432 * TP_STRUCT__entry(
433 * __array( char, prev_comm, TASK_COMM_LEN )
434 * __field( pid_t, prev_pid )
435 * __field( int, prev_prio )
436 * __array( char, next_comm, TASK_COMM_LEN )
437 * __field( pid_t, next_pid )
438 * __field( int, next_prio )
439 * ),
440 *
441 * *
442 * * Assign the entry into the trace record, by embedding
443 * * a full C statement block into TP_fast_assign(). You
444 * * can refer to the trace record as '__entry' -
445 * * otherwise you can put arbitrary C code in here.
446 * *
447 * * Note: this C code will execute every time a trace event
448 * * happens, on an active tracepoint.
449 * *
450 *
451 * TP_fast_assign(
452 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
453 * __entry->prev_pid = prev->pid;
454 * __entry->prev_prio = prev->prio;
455 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
456 * __entry->next_pid = next->pid;
457 * __entry->next_prio = next->prio;
458 * ),
459 *
460 * *
461 * * Formatted output of a trace record via TP_printk().
462 * * This is how the tracepoint will appear under ftrace
463 * * plugins that make use of this tracepoint.
464 * *
465 * * (raw-binary tracing wont actually perform this step.)
466 * *
467 *
468 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
469 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
470 * __entry->next_comm, __entry->next_pid, __entry->next_prio),
471 *
472 * );
473 *
474 * This macro construct is thus used for the regular printk format
475 * tracing setup, it is used to construct a function pointer based
476 * tracepoint callback (this is used by programmatic plugins and
477 * can also by used by generic instrumentation like SystemTap), and
478 * it is also used to expose a structured trace record in
479 * /sys/kernel/debug/tracing/events/.
480 *
481 * A set of (un)registration functions can be passed to the variant
482 * TRACE_EVENT_FN to perform any (un)registration work.
483 */
484
485 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
486 #define DEFINE_EVENT(template, name, proto, args) \
487 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
488 #define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\
489 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
490 #define DEFINE_EVENT_PRINT(template, name, proto, args, print) \
491 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
492 #define DEFINE_EVENT_CONDITION(template, name, proto, \
493 args, cond) \
494 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
495 PARAMS(args), PARAMS(cond))
496
497 #define TRACE_EVENT(name, proto, args, struct, assign, print) \
498 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
499 #define TRACE_EVENT_FN(name, proto, args, struct, \
500 assign, print, reg, unreg) \
501 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
502 #define TRACE_EVENT_CONDITION(name, proto, args, cond, \
503 struct, assign, print) \
504 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
505 PARAMS(args), PARAMS(cond))
506
507 #define TRACE_EVENT_FLAGS(event, flag)
508
509 #define TRACE_EVENT_PERF_PERM(event, expr...)
510
511 #endif /* ifdef TRACE_EVENT (see note above) */
512