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