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
24struct module;
25struct tracepoint;
26struct notifier_block;
27
28struct 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
36extern struct srcu_struct tracepoint_srcu;
37
38extern int
39tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data);
40extern int
41tracepoint_probe_register_prio(struct tracepoint *tp, void *probe, void *data,
42 int prio);
43extern int
44tracepoint_probe_register_prio_may_exist(struct tracepoint *tp, void *probe, void *data,
45 int prio);
46extern int
47tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data);
48static inline int
49tracepoint_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}
55extern void
56for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
57 void *priv);
58
59#ifdef CONFIG_MODULES
60struct tp_module {
61 struct list_head list;
62 struct module *mod;
63};
64
65bool trace_module_has_bad_taint(struct module *mod);
66extern int register_tracepoint_module_notifier(struct notifier_block *nb);
67extern int unregister_tracepoint_module_notifier(struct notifier_block *nb);
68#else
69static inline bool trace_module_has_bad_taint(struct module *mod)
70{
71 return false;
72}
73static inline
74int register_tracepoint_module_notifier(struct notifier_block *nb)
75{
76 return 0;
77}
78static inline
79int 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
91static inline void tracepoint_synchronize_unregister(void)
92{
93 synchronize_srcu(ssp: &tracepoint_srcu);
94 synchronize_rcu();
95}
96#else
97static inline void tracepoint_synchronize_unregister(void)
98{ }
99#endif
100
101#ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
102extern int syscall_regfunc(void);
103extern 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
114static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p)
115{
116 return offset_to_ptr(off: 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
125static 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_TRACEPOINTS 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, args) \
166 do { \
167 struct tracepoint_func *it_func_ptr; \
168 void *__data; \
169 it_func_ptr = \
170 rcu_dereference_raw((&__tracepoint_##name)->funcs); \
171 if (it_func_ptr) { \
172 __data = (it_func_ptr)->data; \
173 static_call(tp_func_##name)(__data, args); \
174 } \
175 } while (0)
176#else
177#define __DO_TRACE_CALL(name, args) __traceiter_##name(NULL, args)
178#endif /* CONFIG_HAVE_STATIC_CALL */
179
180/*
181 * ARCH_WANTS_NO_INSTR archs are expected to have sanitized entry and idle
182 * code that disallow any/all tracing/instrumentation when RCU isn't watching.
183 */
184#ifdef CONFIG_ARCH_WANTS_NO_INSTR
185#define RCUIDLE_COND(rcuidle) (rcuidle)
186#else
187/* srcu can't be used from NMI */
188#define RCUIDLE_COND(rcuidle) (rcuidle && in_nmi())
189#endif
190
191/*
192 * it_func[0] is never NULL because there is at least one element in the array
193 * when the array itself is non NULL.
194 */
195#define __DO_TRACE(name, args, cond, rcuidle) \
196 do { \
197 int __maybe_unused __idx = 0; \
198 \
199 if (!(cond)) \
200 return; \
201 \
202 if (WARN_ON_ONCE(RCUIDLE_COND(rcuidle))) \
203 return; \
204 \
205 /* keep srcu and sched-rcu usage consistent */ \
206 preempt_disable_notrace(); \
207 \
208 /* \
209 * For rcuidle callers, use srcu since sched-rcu \
210 * doesn't work from the idle path. \
211 */ \
212 if (rcuidle) { \
213 __idx = srcu_read_lock_notrace(&tracepoint_srcu);\
214 ct_irq_enter_irqson(); \
215 } \
216 \
217 __DO_TRACE_CALL(name, TP_ARGS(args)); \
218 \
219 if (rcuidle) { \
220 ct_irq_exit_irqson(); \
221 srcu_read_unlock_notrace(&tracepoint_srcu, __idx);\
222 } \
223 \
224 preempt_enable_notrace(); \
225 } while (0)
226
227#ifndef MODULE
228#define __DECLARE_TRACE_RCU(name, proto, args, cond) \
229 static inline void trace_##name##_rcuidle(proto) \
230 { \
231 if (static_key_false(&__tracepoint_##name.key)) \
232 __DO_TRACE(name, \
233 TP_ARGS(args), \
234 TP_CONDITION(cond), 1); \
235 }
236#else
237#define __DECLARE_TRACE_RCU(name, proto, args, cond)
238#endif
239
240/*
241 * Make sure the alignment of the structure in the __tracepoints section will
242 * not add unwanted padding between the beginning of the section and the
243 * structure. Force alignment to the same alignment as the section start.
244 *
245 * When lockdep is enabled, we make sure to always test if RCU is
246 * "watching" regardless if the tracepoint is enabled or not. Tracepoints
247 * require RCU to be active, and it should always warn at the tracepoint
248 * site if it is not watching, as it will need to be active when the
249 * tracepoint is enabled.
250 */
251#define __DECLARE_TRACE(name, proto, args, cond, data_proto) \
252 extern int __traceiter_##name(data_proto); \
253 DECLARE_STATIC_CALL(tp_func_##name, __traceiter_##name); \
254 extern struct tracepoint __tracepoint_##name; \
255 static inline void trace_##name(proto) \
256 { \
257 if (static_key_false(&__tracepoint_##name.key)) \
258 __DO_TRACE(name, \
259 TP_ARGS(args), \
260 TP_CONDITION(cond), 0); \
261 if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) { \
262 WARN_ON_ONCE(!rcu_is_watching()); \
263 } \
264 } \
265 __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args), \
266 PARAMS(cond)) \
267 static inline int \
268 register_trace_##name(void (*probe)(data_proto), void *data) \
269 { \
270 return tracepoint_probe_register(&__tracepoint_##name, \
271 (void *)probe, data); \
272 } \
273 static inline int \
274 register_trace_prio_##name(void (*probe)(data_proto), void *data,\
275 int prio) \
276 { \
277 return tracepoint_probe_register_prio(&__tracepoint_##name, \
278 (void *)probe, data, prio); \
279 } \
280 static inline int \
281 unregister_trace_##name(void (*probe)(data_proto), void *data) \
282 { \
283 return tracepoint_probe_unregister(&__tracepoint_##name,\
284 (void *)probe, data); \
285 } \
286 static inline void \
287 check_trace_callback_type_##name(void (*cb)(data_proto)) \
288 { \
289 } \
290 static inline bool \
291 trace_##name##_enabled(void) \
292 { \
293 return static_key_false(&__tracepoint_##name.key); \
294 }
295
296/*
297 * We have no guarantee that gcc and the linker won't up-align the tracepoint
298 * structures, so we create an array of pointers that will be used for iteration
299 * on the tracepoints.
300 */
301#define DEFINE_TRACE_FN(_name, _reg, _unreg, proto, args) \
302 static const char __tpstrtab_##_name[] \
303 __section("__tracepoints_strings") = #_name; \
304 extern struct static_call_key STATIC_CALL_KEY(tp_func_##_name); \
305 int __traceiter_##_name(void *__data, proto); \
306 void __probestub_##_name(void *__data, proto); \
307 struct tracepoint __tracepoint_##_name __used \
308 __section("__tracepoints") = { \
309 .name = __tpstrtab_##_name, \
310 .key = STATIC_KEY_INIT_FALSE, \
311 .static_call_key = &STATIC_CALL_KEY(tp_func_##_name), \
312 .static_call_tramp = STATIC_CALL_TRAMP_ADDR(tp_func_##_name), \
313 .iterator = &__traceiter_##_name, \
314 .probestub = &__probestub_##_name, \
315 .regfunc = _reg, \
316 .unregfunc = _unreg, \
317 .funcs = NULL }; \
318 __TRACEPOINT_ENTRY(_name); \
319 int __traceiter_##_name(void *__data, proto) \
320 { \
321 struct tracepoint_func *it_func_ptr; \
322 void *it_func; \
323 \
324 it_func_ptr = \
325 rcu_dereference_raw((&__tracepoint_##_name)->funcs); \
326 if (it_func_ptr) { \
327 do { \
328 it_func = READ_ONCE((it_func_ptr)->func); \
329 __data = (it_func_ptr)->data; \
330 ((void(*)(void *, proto))(it_func))(__data, args); \
331 } while ((++it_func_ptr)->func); \
332 } \
333 return 0; \
334 } \
335 void __probestub_##_name(void *__data, proto) \
336 { \
337 } \
338 DEFINE_STATIC_CALL(tp_func_##_name, __traceiter_##_name);
339
340#define DEFINE_TRACE(name, proto, args) \
341 DEFINE_TRACE_FN(name, NULL, NULL, PARAMS(proto), PARAMS(args));
342
343#define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \
344 EXPORT_SYMBOL_GPL(__tracepoint_##name); \
345 EXPORT_SYMBOL_GPL(__traceiter_##name); \
346 EXPORT_STATIC_CALL_GPL(tp_func_##name)
347#define EXPORT_TRACEPOINT_SYMBOL(name) \
348 EXPORT_SYMBOL(__tracepoint_##name); \
349 EXPORT_SYMBOL(__traceiter_##name); \
350 EXPORT_STATIC_CALL(tp_func_##name)
351
352
353#else /* !TRACEPOINTS_ENABLED */
354#define __DECLARE_TRACE(name, proto, args, cond, data_proto) \
355 static inline void trace_##name(proto) \
356 { } \
357 static inline void trace_##name##_rcuidle(proto) \
358 { } \
359 static inline int \
360 register_trace_##name(void (*probe)(data_proto), \
361 void *data) \
362 { \
363 return -ENOSYS; \
364 } \
365 static inline int \
366 unregister_trace_##name(void (*probe)(data_proto), \
367 void *data) \
368 { \
369 return -ENOSYS; \
370 } \
371 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \
372 { \
373 } \
374 static inline bool \
375 trace_##name##_enabled(void) \
376 { \
377 return false; \
378 }
379
380#define DEFINE_TRACE_FN(name, reg, unreg, proto, args)
381#define DEFINE_TRACE(name, proto, args)
382#define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
383#define EXPORT_TRACEPOINT_SYMBOL(name)
384
385#endif /* TRACEPOINTS_ENABLED */
386
387#ifdef CONFIG_TRACING
388/**
389 * tracepoint_string - register constant persistent string to trace system
390 * @str - a constant persistent string that will be referenced in tracepoints
391 *
392 * If constant strings are being used in tracepoints, it is faster and
393 * more efficient to just save the pointer to the string and reference
394 * that with a printf "%s" instead of saving the string in the ring buffer
395 * and wasting space and time.
396 *
397 * The problem with the above approach is that userspace tools that read
398 * the binary output of the trace buffers do not have access to the string.
399 * Instead they just show the address of the string which is not very
400 * useful to users.
401 *
402 * With tracepoint_string(), the string will be registered to the tracing
403 * system and exported to userspace via the debugfs/tracing/printk_formats
404 * file that maps the string address to the string text. This way userspace
405 * tools that read the binary buffers have a way to map the pointers to
406 * the ASCII strings they represent.
407 *
408 * The @str used must be a constant string and persistent as it would not
409 * make sense to show a string that no longer exists. But it is still fine
410 * to be used with modules, because when modules are unloaded, if they
411 * had tracepoints, the ring buffers are cleared too. As long as the string
412 * does not change during the life of the module, it is fine to use
413 * tracepoint_string() within a module.
414 */
415#define tracepoint_string(str) \
416 ({ \
417 static const char *___tp_str __tracepoint_string = str; \
418 ___tp_str; \
419 })
420#define __tracepoint_string __used __section("__tracepoint_str")
421#else
422/*
423 * tracepoint_string() is used to save the string address for userspace
424 * tracing tools. When tracing isn't configured, there's no need to save
425 * anything.
426 */
427# define tracepoint_string(str) str
428# define __tracepoint_string
429#endif
430
431#define DECLARE_TRACE(name, proto, args) \
432 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \
433 cpu_online(raw_smp_processor_id()), \
434 PARAMS(void *__data, proto))
435
436#define DECLARE_TRACE_CONDITION(name, proto, args, cond) \
437 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \
438 cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \
439 PARAMS(void *__data, proto))
440
441#define TRACE_EVENT_FLAGS(event, flag)
442
443#define TRACE_EVENT_PERF_PERM(event, expr...)
444
445#endif /* DECLARE_TRACE */
446
447#ifndef TRACE_EVENT
448/*
449 * For use with the TRACE_EVENT macro:
450 *
451 * We define a tracepoint, its arguments, its printk format
452 * and its 'fast binary record' layout.
453 *
454 * Firstly, name your tracepoint via TRACE_EVENT(name : the
455 * 'subsystem_event' notation is fine.
456 *
457 * Think about this whole construct as the
458 * 'trace_sched_switch() function' from now on.
459 *
460 *
461 * TRACE_EVENT(sched_switch,
462 *
463 * *
464 * * A function has a regular function arguments
465 * * prototype, declare it via TP_PROTO():
466 * *
467 *
468 * TP_PROTO(struct rq *rq, struct task_struct *prev,
469 * struct task_struct *next),
470 *
471 * *
472 * * Define the call signature of the 'function'.
473 * * (Design sidenote: we use this instead of a
474 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
475 * *
476 *
477 * TP_ARGS(rq, prev, next),
478 *
479 * *
480 * * Fast binary tracing: define the trace record via
481 * * TP_STRUCT__entry(). You can think about it like a
482 * * regular C structure local variable definition.
483 * *
484 * * This is how the trace record is structured and will
485 * * be saved into the ring buffer. These are the fields
486 * * that will be exposed to user-space in
487 * * /sys/kernel/tracing/events/<*>/format.
488 * *
489 * * The declared 'local variable' is called '__entry'
490 * *
491 * * __field(pid_t, prev_pid) is equivalent to a standard declaration:
492 * *
493 * * pid_t prev_pid;
494 * *
495 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
496 * *
497 * * char prev_comm[TASK_COMM_LEN];
498 * *
499 *
500 * TP_STRUCT__entry(
501 * __array( char, prev_comm, TASK_COMM_LEN )
502 * __field( pid_t, prev_pid )
503 * __field( int, prev_prio )
504 * __array( char, next_comm, TASK_COMM_LEN )
505 * __field( pid_t, next_pid )
506 * __field( int, next_prio )
507 * ),
508 *
509 * *
510 * * Assign the entry into the trace record, by embedding
511 * * a full C statement block into TP_fast_assign(). You
512 * * can refer to the trace record as '__entry' -
513 * * otherwise you can put arbitrary C code in here.
514 * *
515 * * Note: this C code will execute every time a trace event
516 * * happens, on an active tracepoint.
517 * *
518 *
519 * TP_fast_assign(
520 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
521 * __entry->prev_pid = prev->pid;
522 * __entry->prev_prio = prev->prio;
523 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
524 * __entry->next_pid = next->pid;
525 * __entry->next_prio = next->prio;
526 * ),
527 *
528 * *
529 * * Formatted output of a trace record via TP_printk().
530 * * This is how the tracepoint will appear under ftrace
531 * * plugins that make use of this tracepoint.
532 * *
533 * * (raw-binary tracing wont actually perform this step.)
534 * *
535 *
536 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
537 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
538 * __entry->next_comm, __entry->next_pid, __entry->next_prio),
539 *
540 * );
541 *
542 * This macro construct is thus used for the regular printk format
543 * tracing setup, it is used to construct a function pointer based
544 * tracepoint callback (this is used by programmatic plugins and
545 * can also by used by generic instrumentation like SystemTap), and
546 * it is also used to expose a structured trace record in
547 * /sys/kernel/tracing/events/.
548 *
549 * A set of (un)registration functions can be passed to the variant
550 * TRACE_EVENT_FN to perform any (un)registration work.
551 */
552
553#define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
554#define DEFINE_EVENT(template, name, proto, args) \
555 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
556#define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\
557 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
558#define DEFINE_EVENT_PRINT(template, name, proto, args, print) \
559 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
560#define DEFINE_EVENT_CONDITION(template, name, proto, \
561 args, cond) \
562 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
563 PARAMS(args), PARAMS(cond))
564
565#define TRACE_EVENT(name, proto, args, struct, assign, print) \
566 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
567#define TRACE_EVENT_FN(name, proto, args, struct, \
568 assign, print, reg, unreg) \
569 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
570#define TRACE_EVENT_FN_COND(name, proto, args, cond, struct, \
571 assign, print, reg, unreg) \
572 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
573 PARAMS(args), PARAMS(cond))
574#define TRACE_EVENT_CONDITION(name, proto, args, cond, \
575 struct, assign, print) \
576 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
577 PARAMS(args), PARAMS(cond))
578
579#define TRACE_EVENT_FLAGS(event, flag)
580
581#define TRACE_EVENT_PERF_PERM(event, expr...)
582
583#define DECLARE_EVENT_NOP(name, proto, args) \
584 static inline void trace_##name(proto) \
585 { } \
586 static inline bool trace_##name##_enabled(void) \
587 { \
588 return false; \
589 }
590
591#define TRACE_EVENT_NOP(name, proto, args, struct, assign, print) \
592 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args))
593
594#define DECLARE_EVENT_CLASS_NOP(name, proto, args, tstruct, assign, print)
595#define DEFINE_EVENT_NOP(template, name, proto, args) \
596 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args))
597
598#endif /* ifdef TRACE_EVENT (see note above) */
599

source code of linux/include/linux/tracepoint.h