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