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