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/rcupdate.h>
20 #include <linux/tracepoint-defs.h>
21 #include <linux/static_call.h>
22
23 struct module;
24 struct tracepoint;
25 struct notifier_block;
26
27 struct trace_eval_map {
28 const char *system;
29 const char *eval_string;
30 unsigned long eval_value;
31 };
32
33 #define TRACEPOINT_DEFAULT_PRIO 10
34
35 extern struct srcu_struct tracepoint_srcu;
36
37 extern int
38 tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data);
39 extern int
40 tracepoint_probe_register_prio(struct tracepoint *tp, void *probe, void *data,
41 int prio);
42 extern int
43 tracepoint_probe_register_prio_may_exist(struct tracepoint *tp, void *probe, void *data,
44 int prio);
45 extern int
46 tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data);
47 static inline int
tracepoint_probe_register_may_exist(struct tracepoint * tp,void * probe,void * data)48 tracepoint_probe_register_may_exist(struct tracepoint *tp, void *probe,
49 void *data)
50 {
51 return tracepoint_probe_register_prio_may_exist(tp, probe, data,
52 TRACEPOINT_DEFAULT_PRIO);
53 }
54 extern void
55 for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
56 void *priv);
57
58 #ifdef CONFIG_MODULES
59 struct tp_module {
60 struct list_head list;
61 struct module *mod;
62 };
63
64 bool trace_module_has_bad_taint(struct module *mod);
65 extern int register_tracepoint_module_notifier(struct notifier_block *nb);
66 extern int unregister_tracepoint_module_notifier(struct notifier_block *nb);
67 void for_each_module_tracepoint(void (*fct)(struct tracepoint *,
68 struct module *, void *),
69 void *priv);
70 void for_each_tracepoint_in_module(struct module *,
71 void (*fct)(struct tracepoint *,
72 struct module *, void *),
73 void *priv);
74 #else
trace_module_has_bad_taint(struct module * mod)75 static inline bool trace_module_has_bad_taint(struct module *mod)
76 {
77 return false;
78 }
79 static inline
register_tracepoint_module_notifier(struct notifier_block * nb)80 int register_tracepoint_module_notifier(struct notifier_block *nb)
81 {
82 return 0;
83 }
84 static inline
unregister_tracepoint_module_notifier(struct notifier_block * nb)85 int unregister_tracepoint_module_notifier(struct notifier_block *nb)
86 {
87 return 0;
88 }
89 static inline
for_each_module_tracepoint(void (* fct)(struct tracepoint *,struct module *,void *),void * priv)90 void for_each_module_tracepoint(void (*fct)(struct tracepoint *,
91 struct module *, void *),
92 void *priv)
93 {
94 }
95 static inline
for_each_tracepoint_in_module(struct module * mod,void (* fct)(struct tracepoint *,struct module *,void *),void * priv)96 void for_each_tracepoint_in_module(struct module *mod,
97 void (*fct)(struct tracepoint *,
98 struct module *, void *),
99 void *priv)
100 {
101 }
102 #endif /* CONFIG_MODULES */
103
104 /*
105 * tracepoint_synchronize_unregister must be called between the last tracepoint
106 * probe unregistration and the end of module exit to make sure there is no
107 * caller executing a probe when it is freed.
108 */
109 #ifdef CONFIG_TRACEPOINTS
tracepoint_synchronize_unregister(void)110 static inline void tracepoint_synchronize_unregister(void)
111 {
112 synchronize_srcu(&tracepoint_srcu);
113 synchronize_rcu();
114 }
115 #else
tracepoint_synchronize_unregister(void)116 static inline void tracepoint_synchronize_unregister(void)
117 { }
118 #endif
119
120 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
121 extern int syscall_regfunc(void);
122 extern void syscall_unregfunc(void);
123 #endif /* CONFIG_HAVE_SYSCALL_TRACEPOINTS */
124
125 #ifndef PARAMS
126 #define PARAMS(args...) args
127 #endif
128
129 #define TRACE_DEFINE_ENUM(x)
130 #define TRACE_DEFINE_SIZEOF(x)
131
132 #ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS
tracepoint_ptr_deref(tracepoint_ptr_t * p)133 static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p)
134 {
135 return offset_to_ptr(p);
136 }
137
138 #define __TRACEPOINT_ENTRY(name) \
139 asm(" .section \"__tracepoints_ptrs\", \"a\" \n" \
140 " .balign 4 \n" \
141 " .long __tracepoint_" #name " - . \n" \
142 " .previous \n")
143 #else
tracepoint_ptr_deref(tracepoint_ptr_t * p)144 static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p)
145 {
146 return *p;
147 }
148
149 #define __TRACEPOINT_ENTRY(name) \
150 static tracepoint_ptr_t __tracepoint_ptr_##name __used \
151 __section("__tracepoints_ptrs") = &__tracepoint_##name
152 #endif
153
154 #endif /* _LINUX_TRACEPOINT_H */
155
156 /*
157 * Note: we keep the TRACE_EVENT and DECLARE_TRACE outside the include
158 * file ifdef protection.
159 * This is due to the way trace events work. If a file includes two
160 * trace event headers under one "CREATE_TRACE_POINTS" the first include
161 * will override the TRACE_EVENT and break the second include.
162 */
163
164 #ifndef DECLARE_TRACE
165
166 #define TP_PROTO(args...) args
167 #define TP_ARGS(args...) args
168 #define TP_CONDITION(args...) args
169
170 /*
171 * Individual subsystem my have a separate configuration to
172 * enable their tracepoints. By default, this file will create
173 * the tracepoints if CONFIG_TRACEPOINTS is defined. If a subsystem
174 * wants to be able to disable its tracepoints from being created
175 * it can define NOTRACE before including the tracepoint headers.
176 */
177 #if defined(CONFIG_TRACEPOINTS) && !defined(NOTRACE)
178 #define TRACEPOINTS_ENABLED
179 #endif
180
181 #ifdef TRACEPOINTS_ENABLED
182
183 #ifdef CONFIG_HAVE_STATIC_CALL
184 #define __DO_TRACE_CALL(name, args) \
185 do { \
186 struct tracepoint_func *it_func_ptr; \
187 void *__data; \
188 it_func_ptr = \
189 rcu_dereference_raw((&__tracepoint_##name)->funcs); \
190 if (it_func_ptr) { \
191 __data = (it_func_ptr)->data; \
192 static_call(tp_func_##name)(__data, args); \
193 } \
194 } while (0)
195 #else
196 #define __DO_TRACE_CALL(name, args) __traceiter_##name(NULL, args)
197 #endif /* CONFIG_HAVE_STATIC_CALL */
198
199 /*
200 * ARCH_WANTS_NO_INSTR archs are expected to have sanitized entry and idle
201 * code that disallow any/all tracing/instrumentation when RCU isn't watching.
202 */
203 #ifdef CONFIG_ARCH_WANTS_NO_INSTR
204 #define RCUIDLE_COND(rcuidle) (rcuidle)
205 #else
206 /* srcu can't be used from NMI */
207 #define RCUIDLE_COND(rcuidle) (rcuidle && in_nmi())
208 #endif
209
210 /*
211 * it_func[0] is never NULL because there is at least one element in the array
212 * when the array itself is non NULL.
213 */
214 #define __DO_TRACE(name, args, cond, rcuidle) \
215 do { \
216 int __maybe_unused __idx = 0; \
217 \
218 if (!(cond)) \
219 return; \
220 \
221 if (WARN_ONCE(RCUIDLE_COND(rcuidle), \
222 "Bad RCU usage for tracepoint")) \
223 return; \
224 \
225 /* keep srcu and sched-rcu usage consistent */ \
226 preempt_disable_notrace(); \
227 \
228 /* \
229 * For rcuidle callers, use srcu since sched-rcu \
230 * doesn't work from the idle path. \
231 */ \
232 if (rcuidle) { \
233 __idx = srcu_read_lock_notrace(&tracepoint_srcu);\
234 ct_irq_enter_irqson(); \
235 } \
236 \
237 __DO_TRACE_CALL(name, TP_ARGS(args)); \
238 \
239 if (rcuidle) { \
240 ct_irq_exit_irqson(); \
241 srcu_read_unlock_notrace(&tracepoint_srcu, __idx);\
242 } \
243 \
244 preempt_enable_notrace(); \
245 } while (0)
246
247 #ifndef MODULE
248 #define __DECLARE_TRACE_RCU(name, proto, args, cond) \
249 static inline void trace_##name##_rcuidle(proto) \
250 { \
251 if (static_key_false(&__tracepoint_##name.key)) \
252 __DO_TRACE(name, \
253 TP_ARGS(args), \
254 TP_CONDITION(cond), 1); \
255 }
256 #else
257 #define __DECLARE_TRACE_RCU(name, proto, args, cond)
258 #endif
259
260 /*
261 * Make sure the alignment of the structure in the __tracepoints section will
262 * not add unwanted padding between the beginning of the section and the
263 * structure. Force alignment to the same alignment as the section start.
264 *
265 * When lockdep is enabled, we make sure to always test if RCU is
266 * "watching" regardless if the tracepoint is enabled or not. Tracepoints
267 * require RCU to be active, and it should always warn at the tracepoint
268 * site if it is not watching, as it will need to be active when the
269 * tracepoint is enabled.
270 */
271 #define __DECLARE_TRACE(name, proto, args, cond, data_proto) \
272 extern int __traceiter_##name(data_proto); \
273 DECLARE_STATIC_CALL(tp_func_##name, __traceiter_##name); \
274 extern struct tracepoint __tracepoint_##name; \
275 static inline void trace_##name(proto) \
276 { \
277 if (static_key_false(&__tracepoint_##name.key)) \
278 __DO_TRACE(name, \
279 TP_ARGS(args), \
280 TP_CONDITION(cond), 0); \
281 if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) { \
282 WARN_ONCE(!rcu_is_watching(), \
283 "RCU not watching for tracepoint"); \
284 } \
285 } \
286 __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args), \
287 PARAMS(cond)) \
288 static inline int \
289 register_trace_##name(void (*probe)(data_proto), void *data) \
290 { \
291 return tracepoint_probe_register(&__tracepoint_##name, \
292 (void *)probe, data); \
293 } \
294 static inline int \
295 register_trace_prio_##name(void (*probe)(data_proto), void *data,\
296 int prio) \
297 { \
298 return tracepoint_probe_register_prio(&__tracepoint_##name, \
299 (void *)probe, data, prio); \
300 } \
301 static inline int \
302 unregister_trace_##name(void (*probe)(data_proto), void *data) \
303 { \
304 return tracepoint_probe_unregister(&__tracepoint_##name,\
305 (void *)probe, data); \
306 } \
307 static inline void \
308 check_trace_callback_type_##name(void (*cb)(data_proto)) \
309 { \
310 } \
311 static inline bool \
312 trace_##name##_enabled(void) \
313 { \
314 return static_key_false(&__tracepoint_##name.key); \
315 }
316
317 /*
318 * We have no guarantee that gcc and the linker won't up-align the tracepoint
319 * structures, so we create an array of pointers that will be used for iteration
320 * on the tracepoints.
321 */
322 #define DEFINE_TRACE_FN(_name, _reg, _unreg, proto, args) \
323 static const char __tpstrtab_##_name[] \
324 __section("__tracepoints_strings") = #_name; \
325 extern struct static_call_key STATIC_CALL_KEY(tp_func_##_name); \
326 int __traceiter_##_name(void *__data, proto); \
327 void __probestub_##_name(void *__data, proto); \
328 struct tracepoint __tracepoint_##_name __used \
329 __section("__tracepoints") = { \
330 .name = __tpstrtab_##_name, \
331 .key = STATIC_KEY_INIT_FALSE, \
332 .static_call_key = &STATIC_CALL_KEY(tp_func_##_name), \
333 .static_call_tramp = STATIC_CALL_TRAMP_ADDR(tp_func_##_name), \
334 .iterator = &__traceiter_##_name, \
335 .probestub = &__probestub_##_name, \
336 .regfunc = _reg, \
337 .unregfunc = _unreg, \
338 .funcs = NULL }; \
339 __TRACEPOINT_ENTRY(_name); \
340 int __traceiter_##_name(void *__data, proto) \
341 { \
342 struct tracepoint_func *it_func_ptr; \
343 void *it_func; \
344 \
345 it_func_ptr = \
346 rcu_dereference_raw((&__tracepoint_##_name)->funcs); \
347 if (it_func_ptr) { \
348 do { \
349 it_func = READ_ONCE((it_func_ptr)->func); \
350 __data = (it_func_ptr)->data; \
351 ((void(*)(void *, proto))(it_func))(__data, args); \
352 } while ((++it_func_ptr)->func); \
353 } \
354 return 0; \
355 } \
356 void __probestub_##_name(void *__data, proto) \
357 { \
358 } \
359 DEFINE_STATIC_CALL(tp_func_##_name, __traceiter_##_name);
360
361 #define DEFINE_TRACE(name, proto, args) \
362 DEFINE_TRACE_FN(name, NULL, NULL, PARAMS(proto), PARAMS(args));
363
364 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \
365 EXPORT_SYMBOL_GPL(__tracepoint_##name); \
366 EXPORT_SYMBOL_GPL(__traceiter_##name); \
367 EXPORT_STATIC_CALL_GPL(tp_func_##name)
368 #define EXPORT_TRACEPOINT_SYMBOL(name) \
369 EXPORT_SYMBOL(__tracepoint_##name); \
370 EXPORT_SYMBOL(__traceiter_##name); \
371 EXPORT_STATIC_CALL(tp_func_##name)
372
373
374 #else /* !TRACEPOINTS_ENABLED */
375 #define __DECLARE_TRACE(name, proto, args, cond, data_proto) \
376 static inline void trace_##name(proto) \
377 { } \
378 static inline void trace_##name##_rcuidle(proto) \
379 { } \
380 static inline int \
381 register_trace_##name(void (*probe)(data_proto), \
382 void *data) \
383 { \
384 return -ENOSYS; \
385 } \
386 static inline int \
387 unregister_trace_##name(void (*probe)(data_proto), \
388 void *data) \
389 { \
390 return -ENOSYS; \
391 } \
392 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \
393 { \
394 } \
395 static inline bool \
396 trace_##name##_enabled(void) \
397 { \
398 return false; \
399 }
400
401 #define DEFINE_TRACE_FN(name, reg, unreg, proto, args)
402 #define DEFINE_TRACE(name, proto, args)
403 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
404 #define EXPORT_TRACEPOINT_SYMBOL(name)
405
406 #endif /* TRACEPOINTS_ENABLED */
407
408 #ifdef CONFIG_TRACING
409 /**
410 * tracepoint_string - register constant persistent string to trace system
411 * @str - a constant persistent string that will be referenced in tracepoints
412 *
413 * If constant strings are being used in tracepoints, it is faster and
414 * more efficient to just save the pointer to the string and reference
415 * that with a printf "%s" instead of saving the string in the ring buffer
416 * and wasting space and time.
417 *
418 * The problem with the above approach is that userspace tools that read
419 * the binary output of the trace buffers do not have access to the string.
420 * Instead they just show the address of the string which is not very
421 * useful to users.
422 *
423 * With tracepoint_string(), the string will be registered to the tracing
424 * system and exported to userspace via the debugfs/tracing/printk_formats
425 * file that maps the string address to the string text. This way userspace
426 * tools that read the binary buffers have a way to map the pointers to
427 * the ASCII strings they represent.
428 *
429 * The @str used must be a constant string and persistent as it would not
430 * make sense to show a string that no longer exists. But it is still fine
431 * to be used with modules, because when modules are unloaded, if they
432 * had tracepoints, the ring buffers are cleared too. As long as the string
433 * does not change during the life of the module, it is fine to use
434 * tracepoint_string() within a module.
435 */
436 #define tracepoint_string(str) \
437 ({ \
438 static const char *___tp_str __tracepoint_string = str; \
439 ___tp_str; \
440 })
441 #define __tracepoint_string __used __section("__tracepoint_str")
442 #else
443 /*
444 * tracepoint_string() is used to save the string address for userspace
445 * tracing tools. When tracing isn't configured, there's no need to save
446 * anything.
447 */
448 # define tracepoint_string(str) str
449 # define __tracepoint_string
450 #endif
451
452 #define DECLARE_TRACE(name, proto, args) \
453 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \
454 cpu_online(raw_smp_processor_id()), \
455 PARAMS(void *__data, proto))
456
457 #define DECLARE_TRACE_CONDITION(name, proto, args, cond) \
458 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \
459 cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \
460 PARAMS(void *__data, proto))
461
462 #define TRACE_EVENT_FLAGS(event, flag)
463
464 #define TRACE_EVENT_PERF_PERM(event, expr...)
465
466 #endif /* DECLARE_TRACE */
467
468 #ifndef TRACE_EVENT
469 /*
470 * For use with the TRACE_EVENT macro:
471 *
472 * We define a tracepoint, its arguments, its printk format
473 * and its 'fast binary record' layout.
474 *
475 * Firstly, name your tracepoint via TRACE_EVENT(name : the
476 * 'subsystem_event' notation is fine.
477 *
478 * Think about this whole construct as the
479 * 'trace_sched_switch() function' from now on.
480 *
481 *
482 * TRACE_EVENT(sched_switch,
483 *
484 * *
485 * * A function has a regular function arguments
486 * * prototype, declare it via TP_PROTO():
487 * *
488 *
489 * TP_PROTO(struct rq *rq, struct task_struct *prev,
490 * struct task_struct *next),
491 *
492 * *
493 * * Define the call signature of the 'function'.
494 * * (Design sidenote: we use this instead of a
495 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
496 * *
497 *
498 * TP_ARGS(rq, prev, next),
499 *
500 * *
501 * * Fast binary tracing: define the trace record via
502 * * TP_STRUCT__entry(). You can think about it like a
503 * * regular C structure local variable definition.
504 * *
505 * * This is how the trace record is structured and will
506 * * be saved into the ring buffer. These are the fields
507 * * that will be exposed to user-space in
508 * * /sys/kernel/tracing/events/<*>/format.
509 * *
510 * * The declared 'local variable' is called '__entry'
511 * *
512 * * __field(pid_t, prev_pid) is equivalent to a standard declaration:
513 * *
514 * * pid_t prev_pid;
515 * *
516 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
517 * *
518 * * char prev_comm[TASK_COMM_LEN];
519 * *
520 *
521 * TP_STRUCT__entry(
522 * __array( char, prev_comm, TASK_COMM_LEN )
523 * __field( pid_t, prev_pid )
524 * __field( int, prev_prio )
525 * __array( char, next_comm, TASK_COMM_LEN )
526 * __field( pid_t, next_pid )
527 * __field( int, next_prio )
528 * ),
529 *
530 * *
531 * * Assign the entry into the trace record, by embedding
532 * * a full C statement block into TP_fast_assign(). You
533 * * can refer to the trace record as '__entry' -
534 * * otherwise you can put arbitrary C code in here.
535 * *
536 * * Note: this C code will execute every time a trace event
537 * * happens, on an active tracepoint.
538 * *
539 *
540 * TP_fast_assign(
541 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
542 * __entry->prev_pid = prev->pid;
543 * __entry->prev_prio = prev->prio;
544 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
545 * __entry->next_pid = next->pid;
546 * __entry->next_prio = next->prio;
547 * ),
548 *
549 * *
550 * * Formatted output of a trace record via TP_printk().
551 * * This is how the tracepoint will appear under ftrace
552 * * plugins that make use of this tracepoint.
553 * *
554 * * (raw-binary tracing wont actually perform this step.)
555 * *
556 *
557 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
558 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
559 * __entry->next_comm, __entry->next_pid, __entry->next_prio),
560 *
561 * );
562 *
563 * This macro construct is thus used for the regular printk format
564 * tracing setup, it is used to construct a function pointer based
565 * tracepoint callback (this is used by programmatic plugins and
566 * can also by used by generic instrumentation like SystemTap), and
567 * it is also used to expose a structured trace record in
568 * /sys/kernel/tracing/events/.
569 *
570 * A set of (un)registration functions can be passed to the variant
571 * TRACE_EVENT_FN to perform any (un)registration work.
572 */
573
574 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
575 #define DEFINE_EVENT(template, name, proto, args) \
576 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
577 #define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\
578 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
579 #define DEFINE_EVENT_PRINT(template, name, proto, args, print) \
580 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
581 #define DEFINE_EVENT_CONDITION(template, name, proto, \
582 args, cond) \
583 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
584 PARAMS(args), PARAMS(cond))
585
586 #define TRACE_EVENT(name, proto, args, struct, assign, print) \
587 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
588 #define TRACE_EVENT_FN(name, proto, args, struct, \
589 assign, print, reg, unreg) \
590 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
591 #define TRACE_EVENT_FN_COND(name, proto, args, cond, struct, \
592 assign, print, reg, unreg) \
593 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
594 PARAMS(args), PARAMS(cond))
595 #define TRACE_EVENT_CONDITION(name, proto, args, cond, \
596 struct, assign, print) \
597 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
598 PARAMS(args), PARAMS(cond))
599
600 #define TRACE_EVENT_FLAGS(event, flag)
601
602 #define TRACE_EVENT_PERF_PERM(event, expr...)
603
604 #define DECLARE_EVENT_NOP(name, proto, args) \
605 static inline void trace_##name(proto) \
606 { } \
607 static inline bool trace_##name##_enabled(void) \
608 { \
609 return false; \
610 }
611
612 #define TRACE_EVENT_NOP(name, proto, args, struct, assign, print) \
613 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args))
614
615 #define DECLARE_EVENT_CLASS_NOP(name, proto, args, tstruct, assign, print)
616 #define DEFINE_EVENT_NOP(template, name, proto, args) \
617 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args))
618
619 #endif /* ifdef TRACE_EVENT (see note above) */
620