]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - include/linux/tracepoint.h
Merge tag 'objtool-urgent-2020-04-25' of git://git.kernel.org/pub/scm/linux/kernel...
[mirror_ubuntu-jammy-kernel.git] / include / linux / tracepoint.h
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)".
160 */
161 #define __DO_TRACE(tp, proto, args, cond, rcuidle) \
162 do { \
163 struct tracepoint_func *it_func_ptr; \
164 void *it_func; \
165 void *__data; \
166 int __maybe_unused __idx = 0; \
167 \
168 if (!(cond)) \
169 return; \
170 \
171 /* srcu can't be used from NMI */ \
172 WARN_ON_ONCE(rcuidle && in_nmi()); \
173 \
174 /* keep srcu and sched-rcu usage consistent */ \
175 preempt_disable_notrace(); \
176 \
177 /* \
178 * For rcuidle callers, use srcu since sched-rcu \
179 * doesn't work from the idle path. \
180 */ \
181 if (rcuidle) { \
182 __idx = srcu_read_lock_notrace(&tracepoint_srcu);\
183 rcu_irq_enter_irqson(); \
184 } \
185 \
186 it_func_ptr = rcu_dereference_raw((tp)->funcs); \
187 \
188 if (it_func_ptr) { \
189 do { \
190 it_func = (it_func_ptr)->func; \
191 __data = (it_func_ptr)->data; \
192 ((void(*)(proto))(it_func))(args); \
193 } while ((++it_func_ptr)->func); \
194 } \
195 \
196 if (rcuidle) { \
197 rcu_irq_exit_irqson(); \
198 srcu_read_unlock_notrace(&tracepoint_srcu, __idx);\
199 } \
200 \
201 preempt_enable_notrace(); \
202 } while (0)
203
204 #ifndef MODULE
205 #define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args) \
206 static inline void trace_##name##_rcuidle(proto) \
207 { \
208 if (static_key_false(&__tracepoint_##name.key)) \
209 __DO_TRACE(&__tracepoint_##name, \
210 TP_PROTO(data_proto), \
211 TP_ARGS(data_args), \
212 TP_CONDITION(cond), 1); \
213 }
214 #else
215 #define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args)
216 #endif
217
218 /*
219 * Make sure the alignment of the structure in the __tracepoints section will
220 * not add unwanted padding between the beginning of the section and the
221 * structure. Force alignment to the same alignment as the section start.
222 *
223 * When lockdep is enabled, we make sure to always do the RCU portions of
224 * the tracepoint code, regardless of whether tracing is on. However,
225 * don't check if the condition is false, due to interaction with idle
226 * instrumentation. This lets us find RCU issues triggered with tracepoints
227 * even when this tracepoint is off. This code has no purpose other than
228 * poking RCU a bit.
229 */
230 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
231 extern struct tracepoint __tracepoint_##name; \
232 static inline void trace_##name(proto) \
233 { \
234 if (static_key_false(&__tracepoint_##name.key)) \
235 __DO_TRACE(&__tracepoint_##name, \
236 TP_PROTO(data_proto), \
237 TP_ARGS(data_args), \
238 TP_CONDITION(cond), 0); \
239 if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) { \
240 rcu_read_lock_sched_notrace(); \
241 rcu_dereference_sched(__tracepoint_##name.funcs);\
242 rcu_read_unlock_sched_notrace(); \
243 } \
244 } \
245 __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args), \
246 PARAMS(cond), PARAMS(data_proto), PARAMS(data_args)) \
247 static inline int \
248 register_trace_##name(void (*probe)(data_proto), void *data) \
249 { \
250 return tracepoint_probe_register(&__tracepoint_##name, \
251 (void *)probe, data); \
252 } \
253 static inline int \
254 register_trace_prio_##name(void (*probe)(data_proto), void *data,\
255 int prio) \
256 { \
257 return tracepoint_probe_register_prio(&__tracepoint_##name, \
258 (void *)probe, data, prio); \
259 } \
260 static inline int \
261 unregister_trace_##name(void (*probe)(data_proto), void *data) \
262 { \
263 return tracepoint_probe_unregister(&__tracepoint_##name,\
264 (void *)probe, data); \
265 } \
266 static inline void \
267 check_trace_callback_type_##name(void (*cb)(data_proto)) \
268 { \
269 } \
270 static inline bool \
271 trace_##name##_enabled(void) \
272 { \
273 return static_key_false(&__tracepoint_##name.key); \
274 }
275
276 /*
277 * We have no guarantee that gcc and the linker won't up-align the tracepoint
278 * structures, so we create an array of pointers that will be used for iteration
279 * on the tracepoints.
280 */
281 #define DEFINE_TRACE_FN(name, reg, unreg) \
282 static const char __tpstrtab_##name[] \
283 __attribute__((section("__tracepoints_strings"))) = #name; \
284 struct tracepoint __tracepoint_##name \
285 __attribute__((section("__tracepoints"), used)) = \
286 { __tpstrtab_##name, STATIC_KEY_INIT_FALSE, reg, unreg, NULL };\
287 __TRACEPOINT_ENTRY(name);
288
289 #define DEFINE_TRACE(name) \
290 DEFINE_TRACE_FN(name, NULL, NULL);
291
292 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \
293 EXPORT_SYMBOL_GPL(__tracepoint_##name)
294 #define EXPORT_TRACEPOINT_SYMBOL(name) \
295 EXPORT_SYMBOL(__tracepoint_##name)
296
297 #else /* !TRACEPOINTS_ENABLED */
298 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
299 static inline void trace_##name(proto) \
300 { } \
301 static inline void trace_##name##_rcuidle(proto) \
302 { } \
303 static inline int \
304 register_trace_##name(void (*probe)(data_proto), \
305 void *data) \
306 { \
307 return -ENOSYS; \
308 } \
309 static inline int \
310 unregister_trace_##name(void (*probe)(data_proto), \
311 void *data) \
312 { \
313 return -ENOSYS; \
314 } \
315 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \
316 { \
317 } \
318 static inline bool \
319 trace_##name##_enabled(void) \
320 { \
321 return false; \
322 }
323
324 #define DEFINE_TRACE_FN(name, reg, unreg)
325 #define DEFINE_TRACE(name)
326 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
327 #define EXPORT_TRACEPOINT_SYMBOL(name)
328
329 #endif /* TRACEPOINTS_ENABLED */
330
331 #ifdef CONFIG_TRACING
332 /**
333 * tracepoint_string - register constant persistent string to trace system
334 * @str - a constant persistent string that will be referenced in tracepoints
335 *
336 * If constant strings are being used in tracepoints, it is faster and
337 * more efficient to just save the pointer to the string and reference
338 * that with a printf "%s" instead of saving the string in the ring buffer
339 * and wasting space and time.
340 *
341 * The problem with the above approach is that userspace tools that read
342 * the binary output of the trace buffers do not have access to the string.
343 * Instead they just show the address of the string which is not very
344 * useful to users.
345 *
346 * With tracepoint_string(), the string will be registered to the tracing
347 * system and exported to userspace via the debugfs/tracing/printk_formats
348 * file that maps the string address to the string text. This way userspace
349 * tools that read the binary buffers have a way to map the pointers to
350 * the ASCII strings they represent.
351 *
352 * The @str used must be a constant string and persistent as it would not
353 * make sense to show a string that no longer exists. But it is still fine
354 * to be used with modules, because when modules are unloaded, if they
355 * had tracepoints, the ring buffers are cleared too. As long as the string
356 * does not change during the life of the module, it is fine to use
357 * tracepoint_string() within a module.
358 */
359 #define tracepoint_string(str) \
360 ({ \
361 static const char *___tp_str __tracepoint_string = str; \
362 ___tp_str; \
363 })
364 #define __tracepoint_string __attribute__((section("__tracepoint_str")))
365 #else
366 /*
367 * tracepoint_string() is used to save the string address for userspace
368 * tracing tools. When tracing isn't configured, there's no need to save
369 * anything.
370 */
371 # define tracepoint_string(str) str
372 # define __tracepoint_string
373 #endif
374
375 #define DECLARE_TRACE(name, proto, args) \
376 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \
377 cpu_online(raw_smp_processor_id()), \
378 PARAMS(void *__data, proto), \
379 PARAMS(__data, args))
380
381 #define DECLARE_TRACE_CONDITION(name, proto, args, cond) \
382 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \
383 cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \
384 PARAMS(void *__data, proto), \
385 PARAMS(__data, args))
386
387 #define TRACE_EVENT_FLAGS(event, flag)
388
389 #define TRACE_EVENT_PERF_PERM(event, expr...)
390
391 #endif /* DECLARE_TRACE */
392
393 #ifndef TRACE_EVENT
394 /*
395 * For use with the TRACE_EVENT macro:
396 *
397 * We define a tracepoint, its arguments, its printk format
398 * and its 'fast binary record' layout.
399 *
400 * Firstly, name your tracepoint via TRACE_EVENT(name : the
401 * 'subsystem_event' notation is fine.
402 *
403 * Think about this whole construct as the
404 * 'trace_sched_switch() function' from now on.
405 *
406 *
407 * TRACE_EVENT(sched_switch,
408 *
409 * *
410 * * A function has a regular function arguments
411 * * prototype, declare it via TP_PROTO():
412 * *
413 *
414 * TP_PROTO(struct rq *rq, struct task_struct *prev,
415 * struct task_struct *next),
416 *
417 * *
418 * * Define the call signature of the 'function'.
419 * * (Design sidenote: we use this instead of a
420 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
421 * *
422 *
423 * TP_ARGS(rq, prev, next),
424 *
425 * *
426 * * Fast binary tracing: define the trace record via
427 * * TP_STRUCT__entry(). You can think about it like a
428 * * regular C structure local variable definition.
429 * *
430 * * This is how the trace record is structured and will
431 * * be saved into the ring buffer. These are the fields
432 * * that will be exposed to user-space in
433 * * /sys/kernel/debug/tracing/events/<*>/format.
434 * *
435 * * The declared 'local variable' is called '__entry'
436 * *
437 * * __field(pid_t, prev_prid) is equivalent to a standard declariton:
438 * *
439 * * pid_t prev_pid;
440 * *
441 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
442 * *
443 * * char prev_comm[TASK_COMM_LEN];
444 * *
445 *
446 * TP_STRUCT__entry(
447 * __array( char, prev_comm, TASK_COMM_LEN )
448 * __field( pid_t, prev_pid )
449 * __field( int, prev_prio )
450 * __array( char, next_comm, TASK_COMM_LEN )
451 * __field( pid_t, next_pid )
452 * __field( int, next_prio )
453 * ),
454 *
455 * *
456 * * Assign the entry into the trace record, by embedding
457 * * a full C statement block into TP_fast_assign(). You
458 * * can refer to the trace record as '__entry' -
459 * * otherwise you can put arbitrary C code in here.
460 * *
461 * * Note: this C code will execute every time a trace event
462 * * happens, on an active tracepoint.
463 * *
464 *
465 * TP_fast_assign(
466 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
467 * __entry->prev_pid = prev->pid;
468 * __entry->prev_prio = prev->prio;
469 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
470 * __entry->next_pid = next->pid;
471 * __entry->next_prio = next->prio;
472 * ),
473 *
474 * *
475 * * Formatted output of a trace record via TP_printk().
476 * * This is how the tracepoint will appear under ftrace
477 * * plugins that make use of this tracepoint.
478 * *
479 * * (raw-binary tracing wont actually perform this step.)
480 * *
481 *
482 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
483 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
484 * __entry->next_comm, __entry->next_pid, __entry->next_prio),
485 *
486 * );
487 *
488 * This macro construct is thus used for the regular printk format
489 * tracing setup, it is used to construct a function pointer based
490 * tracepoint callback (this is used by programmatic plugins and
491 * can also by used by generic instrumentation like SystemTap), and
492 * it is also used to expose a structured trace record in
493 * /sys/kernel/debug/tracing/events/.
494 *
495 * A set of (un)registration functions can be passed to the variant
496 * TRACE_EVENT_FN to perform any (un)registration work.
497 */
498
499 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
500 #define DEFINE_EVENT(template, name, proto, args) \
501 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
502 #define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\
503 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
504 #define DEFINE_EVENT_PRINT(template, name, proto, args, print) \
505 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
506 #define DEFINE_EVENT_CONDITION(template, name, proto, \
507 args, cond) \
508 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
509 PARAMS(args), PARAMS(cond))
510
511 #define TRACE_EVENT(name, proto, args, struct, assign, print) \
512 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
513 #define TRACE_EVENT_FN(name, proto, args, struct, \
514 assign, print, reg, unreg) \
515 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
516 #define TRACE_EVENT_FN_COND(name, proto, args, cond, struct, \
517 assign, print, reg, unreg) \
518 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
519 PARAMS(args), PARAMS(cond))
520 #define TRACE_EVENT_CONDITION(name, proto, args, cond, \
521 struct, assign, print) \
522 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
523 PARAMS(args), PARAMS(cond))
524
525 #define TRACE_EVENT_FLAGS(event, flag)
526
527 #define TRACE_EVENT_PERF_PERM(event, expr...)
528
529 #define DECLARE_EVENT_NOP(name, proto, args) \
530 static inline void trace_##name(proto) \
531 { } \
532 static inline bool trace_##name##_enabled(void) \
533 { \
534 return false; \
535 }
536
537 #define TRACE_EVENT_NOP(name, proto, args, struct, assign, print) \
538 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args))
539
540 #define DECLARE_EVENT_CLASS_NOP(name, proto, args, tstruct, assign, print)
541 #define DEFINE_EVENT_NOP(template, name, proto, args) \
542 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args))
543
544 #endif /* ifdef TRACE_EVENT (see note above) */