1 /* SPDX-License-Identifier: GPL-2.0-only */
2 #ifndef _LINUX_TRACEPOINT_H
3 #define _LINUX_TRACEPOINT_H
4
5 /*
6 * Kernel Tracepoint API.
7 *
8 * See Documentation/trace/tracepoints.rst.
9 *
10 * Copyright (C) 2008-2014 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
11 *
12 * Heavily inspired from the Linux Kernel Markers.
13 */
14
15 #include <linux/smp.h>
16 #include <linux/srcu.h>
17 #include <linux/errno.h>
18 #include <linux/types.h>
19 #include <linux/cpumask.h>
20 #include <linux/rcupdate.h>
21 #include <linux/tracepoint-defs.h>
22 #include <linux/static_call.h>
23
24 struct module;
25 struct tracepoint;
26 struct notifier_block;
27
28 struct trace_eval_map {
29 const char *system;
30 const char *eval_string;
31 unsigned long eval_value;
32 };
33
34 #define TRACEPOINT_DEFAULT_PRIO 10
35
36 extern struct srcu_struct tracepoint_srcu;
37
38 extern int
39 tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data);
40 extern int
41 tracepoint_probe_register_prio(struct tracepoint *tp, void *probe, void *data,
42 int prio);
43 extern int
44 tracepoint_probe_register_prio_may_exist(struct tracepoint *tp, void *probe, void *data,
45 int prio);
46 extern int
47 tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data);
48 static inline int
tracepoint_probe_register_may_exist(struct tracepoint * tp,void * probe,void * data)49 tracepoint_probe_register_may_exist(struct tracepoint *tp, void *probe,
50 void *data)
51 {
52 return tracepoint_probe_register_prio_may_exist(tp, probe, data,
53 TRACEPOINT_DEFAULT_PRIO);
54 }
55 extern void
56 for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
57 void *priv);
58
59 #ifdef CONFIG_MODULES
60 struct tp_module {
61 struct list_head list;
62 struct module *mod;
63 };
64
65 bool trace_module_has_bad_taint(struct module *mod);
66 extern int register_tracepoint_module_notifier(struct notifier_block *nb);
67 extern int unregister_tracepoint_module_notifier(struct notifier_block *nb);
68 #else
trace_module_has_bad_taint(struct module * mod)69 static inline bool trace_module_has_bad_taint(struct module *mod)
70 {
71 return false;
72 }
73 static inline
register_tracepoint_module_notifier(struct notifier_block * nb)74 int register_tracepoint_module_notifier(struct notifier_block *nb)
75 {
76 return 0;
77 }
78 static inline
unregister_tracepoint_module_notifier(struct notifier_block * nb)79 int unregister_tracepoint_module_notifier(struct notifier_block *nb)
80 {
81 return 0;
82 }
83 #endif /* CONFIG_MODULES */
84
85 /*
86 * tracepoint_synchronize_unregister must be called between the last tracepoint
87 * probe unregistration and the end of module exit to make sure there is no
88 * caller executing a probe when it is freed.
89 */
90 #ifdef CONFIG_TRACEPOINTS
tracepoint_synchronize_unregister(void)91 static inline void tracepoint_synchronize_unregister(void)
92 {
93 synchronize_srcu(&tracepoint_srcu);
94 synchronize_rcu();
95 }
96 #else
tracepoint_synchronize_unregister(void)97 static inline void tracepoint_synchronize_unregister(void)
98 { }
99 #endif
100
101 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
102 extern int syscall_regfunc(void);
103 extern void syscall_unregfunc(void);
104 #endif /* CONFIG_HAVE_SYSCALL_TRACEPOINTS */
105
106 #ifndef PARAMS
107 #define PARAMS(args...) args
108 #endif
109
110 #define TRACE_DEFINE_ENUM(x)
111 #define TRACE_DEFINE_SIZEOF(x)
112
113 #ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS
tracepoint_ptr_deref(tracepoint_ptr_t * p)114 static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p)
115 {
116 return offset_to_ptr(p);
117 }
118
119 #define __TRACEPOINT_ENTRY(name) \
120 asm(" .section \"__tracepoints_ptrs\", \"a\" \n" \
121 " .balign 4 \n" \
122 " .long __tracepoint_" #name " - . \n" \
123 " .previous \n")
124 #else
tracepoint_ptr_deref(tracepoint_ptr_t * p)125 static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p)
126 {
127 return *p;
128 }
129
130 #define __TRACEPOINT_ENTRY(name) \
131 static tracepoint_ptr_t __tracepoint_ptr_##name __used \
132 __section("__tracepoints_ptrs") = &__tracepoint_##name
133 #endif
134
135 #endif /* _LINUX_TRACEPOINT_H */
136
137 /*
138 * Note: we keep the TRACE_EVENT and DECLARE_TRACE outside the include
139 * file ifdef protection.
140 * This is due to the way trace events work. If a file includes two
141 * trace event headers under one "CREATE_TRACE_POINTS" the first include
142 * will override the TRACE_EVENT and break the second include.
143 */
144
145 #ifndef DECLARE_TRACE
146
147 #define TP_PROTO(args...) args
148 #define TP_ARGS(args...) args
149 #define TP_CONDITION(args...) args
150
151 /*
152 * Individual subsystem my have a separate configuration to
153 * enable their tracepoints. By default, this file will create
154 * the tracepoints if CONFIG_TRACEPOINT is defined. If a subsystem
155 * wants to be able to disable its tracepoints from being created
156 * it can define NOTRACE before including the tracepoint headers.
157 */
158 #if defined(CONFIG_TRACEPOINTS) && !defined(NOTRACE)
159 #define TRACEPOINTS_ENABLED
160 #endif
161
162 #ifdef TRACEPOINTS_ENABLED
163
164 #ifdef CONFIG_HAVE_STATIC_CALL
165 #define __DO_TRACE_CALL(name) static_call(tp_func_##name)
166 #else
167 #define __DO_TRACE_CALL(name) __traceiter_##name
168 #endif /* CONFIG_HAVE_STATIC_CALL */
169
170 /*
171 * it_func[0] is never NULL because there is at least one element in the array
172 * when the array itself is non NULL.
173 *
174 * Note, the proto and args passed in includes "__data" as the first parameter.
175 * The reason for this is to handle the "void" prototype. If a tracepoint
176 * has a "void" prototype, then it is invalid to declare a function
177 * as "(void *, void)".
178 */
179 #define __DO_TRACE(name, proto, args, cond, rcuidle) \
180 do { \
181 struct tracepoint_func *it_func_ptr; \
182 int __maybe_unused __idx = 0; \
183 void *__data; \
184 \
185 if (!(cond)) \
186 return; \
187 \
188 /* srcu can't be used from NMI */ \
189 WARN_ON_ONCE(rcuidle && in_nmi()); \
190 \
191 /* keep srcu and sched-rcu usage consistent */ \
192 preempt_disable_notrace(); \
193 \
194 /* \
195 * For rcuidle callers, use srcu since sched-rcu \
196 * doesn't work from the idle path. \
197 */ \
198 if (rcuidle) { \
199 __idx = srcu_read_lock_notrace(&tracepoint_srcu);\
200 rcu_irq_enter_irqson(); \
201 } \
202 \
203 it_func_ptr = \
204 rcu_dereference_raw((&__tracepoint_##name)->funcs); \
205 if (it_func_ptr) { \
206 __data = (it_func_ptr)->data; \
207 __DO_TRACE_CALL(name)(args); \
208 } \
209 \
210 if (rcuidle) { \
211 rcu_irq_exit_irqson(); \
212 srcu_read_unlock_notrace(&tracepoint_srcu, __idx);\
213 } \
214 \
215 preempt_enable_notrace(); \
216 } while (0)
217
218 #ifndef MODULE
219 #define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args) \
220 static inline void trace_##name##_rcuidle(proto) \
221 { \
222 if (static_key_false(&__tracepoint_##name.key)) \
223 __DO_TRACE(name, \
224 TP_PROTO(data_proto), \
225 TP_ARGS(data_args), \
226 TP_CONDITION(cond), 1); \
227 }
228 #else
229 #define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args)
230 #endif
231
232 /*
233 * Make sure the alignment of the structure in the __tracepoints section will
234 * not add unwanted padding between the beginning of the section and the
235 * structure. Force alignment to the same alignment as the section start.
236 *
237 * When lockdep is enabled, we make sure to always do the RCU portions of
238 * the tracepoint code, regardless of whether tracing is on. However,
239 * don't check if the condition is false, due to interaction with idle
240 * instrumentation. This lets us find RCU issues triggered with tracepoints
241 * even when this tracepoint is off. This code has no purpose other than
242 * poking RCU a bit.
243 */
244 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
245 extern int __traceiter_##name(data_proto); \
246 DECLARE_STATIC_CALL(tp_func_##name, __traceiter_##name); \
247 extern struct tracepoint __tracepoint_##name; \
248 static inline void trace_##name(proto) \
249 { \
250 if (static_key_false(&__tracepoint_##name.key)) \
251 __DO_TRACE(name, \
252 TP_PROTO(data_proto), \
253 TP_ARGS(data_args), \
254 TP_CONDITION(cond), 0); \
255 if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) { \
256 rcu_read_lock_sched_notrace(); \
257 rcu_dereference_sched(__tracepoint_##name.funcs);\
258 rcu_read_unlock_sched_notrace(); \
259 } \
260 } \
261 __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args), \
262 PARAMS(cond), PARAMS(data_proto), PARAMS(data_args)) \
263 static inline int \
264 register_trace_##name(void (*probe)(data_proto), void *data) \
265 { \
266 return tracepoint_probe_register(&__tracepoint_##name, \
267 (void *)probe, data); \
268 } \
269 static inline int \
270 register_trace_prio_##name(void (*probe)(data_proto), void *data,\
271 int prio) \
272 { \
273 return tracepoint_probe_register_prio(&__tracepoint_##name, \
274 (void *)probe, data, prio); \
275 } \
276 static inline int \
277 unregister_trace_##name(void (*probe)(data_proto), void *data) \
278 { \
279 return tracepoint_probe_unregister(&__tracepoint_##name,\
280 (void *)probe, data); \
281 } \
282 static inline void \
283 check_trace_callback_type_##name(void (*cb)(data_proto)) \
284 { \
285 } \
286 static inline bool \
287 trace_##name##_enabled(void) \
288 { \
289 return static_key_false(&__tracepoint_##name.key); \
290 }
291
292 /*
293 * We have no guarantee that gcc and the linker won't up-align the tracepoint
294 * structures, so we create an array of pointers that will be used for iteration
295 * on the tracepoints.
296 */
297 #define DEFINE_TRACE_FN(_name, _reg, _unreg, proto, args) \
298 static const char __tpstrtab_##_name[] \
299 __section("__tracepoints_strings") = #_name; \
300 extern struct static_call_key STATIC_CALL_KEY(tp_func_##_name); \
301 int __traceiter_##_name(void *__data, proto); \
302 struct tracepoint __tracepoint_##_name __used \
303 __section("__tracepoints") = { \
304 .name = __tpstrtab_##_name, \
305 .key = STATIC_KEY_INIT_FALSE, \
306 .static_call_key = &STATIC_CALL_KEY(tp_func_##_name), \
307 .static_call_tramp = STATIC_CALL_TRAMP_ADDR(tp_func_##_name), \
308 .iterator = &__traceiter_##_name, \
309 .regfunc = _reg, \
310 .unregfunc = _unreg, \
311 .funcs = NULL }; \
312 __TRACEPOINT_ENTRY(_name); \
313 int __traceiter_##_name(void *__data, proto) \
314 { \
315 struct tracepoint_func *it_func_ptr; \
316 void *it_func; \
317 \
318 it_func_ptr = \
319 rcu_dereference_raw((&__tracepoint_##_name)->funcs); \
320 if (it_func_ptr) { \
321 do { \
322 it_func = (it_func_ptr)->func; \
323 __data = (it_func_ptr)->data; \
324 ((void(*)(void *, proto))(it_func))(__data, args); \
325 } while ((++it_func_ptr)->func); \
326 } \
327 return 0; \
328 } \
329 DEFINE_STATIC_CALL(tp_func_##_name, __traceiter_##_name);
330
331 #define DEFINE_TRACE(name, proto, args) \
332 DEFINE_TRACE_FN(name, NULL, NULL, PARAMS(proto), PARAMS(args));
333
334 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \
335 EXPORT_SYMBOL_GPL(__tracepoint_##name); \
336 EXPORT_SYMBOL_GPL(__traceiter_##name); \
337 EXPORT_STATIC_CALL_GPL(tp_func_##name)
338 #define EXPORT_TRACEPOINT_SYMBOL(name) \
339 EXPORT_SYMBOL(__tracepoint_##name); \
340 EXPORT_SYMBOL(__traceiter_##name); \
341 EXPORT_STATIC_CALL(tp_func_##name)
342
343
344 #else /* !TRACEPOINTS_ENABLED */
345 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
346 static inline void trace_##name(proto) \
347 { } \
348 static inline void trace_##name##_rcuidle(proto) \
349 { } \
350 static inline int \
351 register_trace_##name(void (*probe)(data_proto), \
352 void *data) \
353 { \
354 return -ENOSYS; \
355 } \
356 static inline int \
357 unregister_trace_##name(void (*probe)(data_proto), \
358 void *data) \
359 { \
360 return -ENOSYS; \
361 } \
362 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \
363 { \
364 } \
365 static inline bool \
366 trace_##name##_enabled(void) \
367 { \
368 return false; \
369 }
370
371 #define DEFINE_TRACE_FN(name, reg, unreg, proto, args)
372 #define DEFINE_TRACE(name, proto, args)
373 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
374 #define EXPORT_TRACEPOINT_SYMBOL(name)
375
376 #endif /* TRACEPOINTS_ENABLED */
377
378 #ifdef CONFIG_TRACING
379 /**
380 * tracepoint_string - register constant persistent string to trace system
381 * @str - a constant persistent string that will be referenced in tracepoints
382 *
383 * If constant strings are being used in tracepoints, it is faster and
384 * more efficient to just save the pointer to the string and reference
385 * that with a printf "%s" instead of saving the string in the ring buffer
386 * and wasting space and time.
387 *
388 * The problem with the above approach is that userspace tools that read
389 * the binary output of the trace buffers do not have access to the string.
390 * Instead they just show the address of the string which is not very
391 * useful to users.
392 *
393 * With tracepoint_string(), the string will be registered to the tracing
394 * system and exported to userspace via the debugfs/tracing/printk_formats
395 * file that maps the string address to the string text. This way userspace
396 * tools that read the binary buffers have a way to map the pointers to
397 * the ASCII strings they represent.
398 *
399 * The @str used must be a constant string and persistent as it would not
400 * make sense to show a string that no longer exists. But it is still fine
401 * to be used with modules, because when modules are unloaded, if they
402 * had tracepoints, the ring buffers are cleared too. As long as the string
403 * does not change during the life of the module, it is fine to use
404 * tracepoint_string() within a module.
405 */
406 #define tracepoint_string(str) \
407 ({ \
408 static const char *___tp_str __tracepoint_string = str; \
409 ___tp_str; \
410 })
411 #define __tracepoint_string __used __section("__tracepoint_str")
412 #else
413 /*
414 * tracepoint_string() is used to save the string address for userspace
415 * tracing tools. When tracing isn't configured, there's no need to save
416 * anything.
417 */
418 # define tracepoint_string(str) str
419 # define __tracepoint_string
420 #endif
421
422 #define DECLARE_TRACE(name, proto, args) \
423 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \
424 cpu_online(raw_smp_processor_id()), \
425 PARAMS(void *__data, proto), \
426 PARAMS(__data, args))
427
428 #define DECLARE_TRACE_CONDITION(name, proto, args, cond) \
429 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \
430 cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \
431 PARAMS(void *__data, proto), \
432 PARAMS(__data, args))
433
434 #define TRACE_EVENT_FLAGS(event, flag)
435
436 #define TRACE_EVENT_PERF_PERM(event, expr...)
437
438 #endif /* DECLARE_TRACE */
439
440 #ifndef TRACE_EVENT
441 /*
442 * For use with the TRACE_EVENT macro:
443 *
444 * We define a tracepoint, its arguments, its printk format
445 * and its 'fast binary record' layout.
446 *
447 * Firstly, name your tracepoint via TRACE_EVENT(name : the
448 * 'subsystem_event' notation is fine.
449 *
450 * Think about this whole construct as the
451 * 'trace_sched_switch() function' from now on.
452 *
453 *
454 * TRACE_EVENT(sched_switch,
455 *
456 * *
457 * * A function has a regular function arguments
458 * * prototype, declare it via TP_PROTO():
459 * *
460 *
461 * TP_PROTO(struct rq *rq, struct task_struct *prev,
462 * struct task_struct *next),
463 *
464 * *
465 * * Define the call signature of the 'function'.
466 * * (Design sidenote: we use this instead of a
467 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
468 * *
469 *
470 * TP_ARGS(rq, prev, next),
471 *
472 * *
473 * * Fast binary tracing: define the trace record via
474 * * TP_STRUCT__entry(). You can think about it like a
475 * * regular C structure local variable definition.
476 * *
477 * * This is how the trace record is structured and will
478 * * be saved into the ring buffer. These are the fields
479 * * that will be exposed to user-space in
480 * * /sys/kernel/debug/tracing/events/<*>/format.
481 * *
482 * * The declared 'local variable' is called '__entry'
483 * *
484 * * __field(pid_t, prev_prid) is equivalent to a standard declariton:
485 * *
486 * * pid_t prev_pid;
487 * *
488 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
489 * *
490 * * char prev_comm[TASK_COMM_LEN];
491 * *
492 *
493 * TP_STRUCT__entry(
494 * __array( char, prev_comm, TASK_COMM_LEN )
495 * __field( pid_t, prev_pid )
496 * __field( int, prev_prio )
497 * __array( char, next_comm, TASK_COMM_LEN )
498 * __field( pid_t, next_pid )
499 * __field( int, next_prio )
500 * ),
501 *
502 * *
503 * * Assign the entry into the trace record, by embedding
504 * * a full C statement block into TP_fast_assign(). You
505 * * can refer to the trace record as '__entry' -
506 * * otherwise you can put arbitrary C code in here.
507 * *
508 * * Note: this C code will execute every time a trace event
509 * * happens, on an active tracepoint.
510 * *
511 *
512 * TP_fast_assign(
513 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
514 * __entry->prev_pid = prev->pid;
515 * __entry->prev_prio = prev->prio;
516 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
517 * __entry->next_pid = next->pid;
518 * __entry->next_prio = next->prio;
519 * ),
520 *
521 * *
522 * * Formatted output of a trace record via TP_printk().
523 * * This is how the tracepoint will appear under ftrace
524 * * plugins that make use of this tracepoint.
525 * *
526 * * (raw-binary tracing wont actually perform this step.)
527 * *
528 *
529 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
530 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
531 * __entry->next_comm, __entry->next_pid, __entry->next_prio),
532 *
533 * );
534 *
535 * This macro construct is thus used for the regular printk format
536 * tracing setup, it is used to construct a function pointer based
537 * tracepoint callback (this is used by programmatic plugins and
538 * can also by used by generic instrumentation like SystemTap), and
539 * it is also used to expose a structured trace record in
540 * /sys/kernel/debug/tracing/events/.
541 *
542 * A set of (un)registration functions can be passed to the variant
543 * TRACE_EVENT_FN to perform any (un)registration work.
544 */
545
546 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
547 #define DEFINE_EVENT(template, name, proto, args) \
548 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
549 #define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\
550 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
551 #define DEFINE_EVENT_PRINT(template, name, proto, args, print) \
552 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
553 #define DEFINE_EVENT_CONDITION(template, name, proto, \
554 args, cond) \
555 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
556 PARAMS(args), PARAMS(cond))
557
558 #define TRACE_EVENT(name, proto, args, struct, assign, print) \
559 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
560 #define TRACE_EVENT_FN(name, proto, args, struct, \
561 assign, print, reg, unreg) \
562 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
563 #define TRACE_EVENT_FN_COND(name, proto, args, cond, struct, \
564 assign, print, reg, unreg) \
565 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
566 PARAMS(args), PARAMS(cond))
567 #define TRACE_EVENT_CONDITION(name, proto, args, cond, \
568 struct, assign, print) \
569 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
570 PARAMS(args), PARAMS(cond))
571
572 #define TRACE_EVENT_FLAGS(event, flag)
573
574 #define TRACE_EVENT_PERF_PERM(event, expr...)
575
576 #define DECLARE_EVENT_NOP(name, proto, args) \
577 static inline void trace_##name(proto) \
578 { } \
579 static inline bool trace_##name##_enabled(void) \
580 { \
581 return false; \
582 }
583
584 #define TRACE_EVENT_NOP(name, proto, args, struct, assign, print) \
585 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args))
586
587 #define DECLARE_EVENT_CLASS_NOP(name, proto, args, tstruct, assign, print)
588 #define DEFINE_EVENT_NOP(template, name, proto, args) \
589 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args))
590
591 #endif /* ifdef TRACE_EVENT (see note above) */
592