1 /* SPDX-License-Identifier: GPL-2.0-only */
2 #ifndef _LINUX_TRACEPOINT_H
3 #define _LINUX_TRACEPOINT_H
6 * Kernel Tracepoint API.
8 * See Documentation/trace/tracepoints.rst.
12 * Heavily inspired from the Linux Kernel Markers.
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>
26 struct notifier_block;
28 struct trace_eval_map {
30 const char *eval_string;
31 unsigned long eval_value;
34 #define TRACEPOINT_DEFAULT_PRIO 10
36 extern struct srcu_struct tracepoint_srcu;
39 tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data);
41 tracepoint_probe_register_prio(struct tracepoint *tp, void *probe, void *data,
44 tracepoint_probe_register_prio_may_exist(struct tracepoint *tp, void *probe, void *data,
47 tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data);
49 tracepoint_probe_register_may_exist(struct tracepoint *tp, void *probe,
52 return tracepoint_probe_register_prio_may_exist(tp, probe, data,
53 TRACEPOINT_DEFAULT_PRIO);
56 for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
61 struct list_head list;
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);
69 static inline bool trace_module_has_bad_taint(struct module *mod)
74 int register_tracepoint_module_notifier(struct notifier_block *nb)
79 int unregister_tracepoint_module_notifier(struct notifier_block *nb)
83 #endif /* CONFIG_MODULES */
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.
90 #ifdef CONFIG_TRACEPOINTS
91 static inline void tracepoint_synchronize_unregister(void)
93 synchronize_srcu(&tracepoint_srcu);
97 static inline void tracepoint_synchronize_unregister(void)
101 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
102 extern int syscall_regfunc(void);
103 extern void syscall_unregfunc(void);
104 #endif /* CONFIG_HAVE_SYSCALL_TRACEPOINTS */
107 #define PARAMS(args...) args
110 #define TRACE_DEFINE_ENUM(x)
111 #define TRACE_DEFINE_SIZEOF(x)
113 #ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS
114 static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p)
116 return offset_to_ptr(p);
119 #define __TRACEPOINT_ENTRY(name) \
120 asm(" .section \"__tracepoints_ptrs\", \"a\" \n" \
122 " .long __tracepoint_" #name " - . \n" \
125 static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p)
130 #define __TRACEPOINT_ENTRY(name) \
131 static tracepoint_ptr_t __tracepoint_ptr_##name __used \
132 __section("__tracepoints_ptrs") = &__tracepoint_##name
135 #endif /* _LINUX_TRACEPOINT_H */
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.
145 #ifndef DECLARE_TRACE
147 #define TP_PROTO(args...) args
148 #define TP_ARGS(args...) args
149 #define TP_CONDITION(args...) args
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.
158 #if defined(CONFIG_TRACEPOINTS) && !defined(NOTRACE)
159 #define TRACEPOINTS_ENABLED
162 #ifdef TRACEPOINTS_ENABLED
164 #ifdef CONFIG_HAVE_STATIC_CALL
165 #define __DO_TRACE_CALL(name, args) \
167 struct tracepoint_func *it_func_ptr; \
170 rcu_dereference_raw((&__tracepoint_##name)->funcs); \
172 __data = (it_func_ptr)->data; \
173 static_call(tp_func_##name)(__data, args); \
177 #define __DO_TRACE_CALL(name, args) __traceiter_##name(NULL, args)
178 #endif /* CONFIG_HAVE_STATIC_CALL */
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.
184 #ifdef CONFIG_ARCH_WANTS_NO_INSTR
185 #define RCUIDLE_COND(rcuidle) (rcuidle)
187 /* srcu can't be used from NMI */
188 #define RCUIDLE_COND(rcuidle) (rcuidle && in_nmi())
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.
195 #define __DO_TRACE(name, args, cond, rcuidle) \
197 int __maybe_unused __idx = 0; \
202 if (WARN_ON_ONCE(RCUIDLE_COND(rcuidle))) \
205 /* keep srcu and sched-rcu usage consistent */ \
206 preempt_disable_notrace(); \
209 * For rcuidle callers, use srcu since sched-rcu \
210 * doesn't work from the idle path. \
213 __idx = srcu_read_lock_notrace(&tracepoint_srcu);\
214 ct_irq_enter_irqson(); \
217 __DO_TRACE_CALL(name, TP_ARGS(args)); \
220 ct_irq_exit_irqson(); \
221 srcu_read_unlock_notrace(&tracepoint_srcu, __idx);\
224 preempt_enable_notrace(); \
228 #define __DECLARE_TRACE_RCU(name, proto, args, cond) \
229 static inline void trace_##name##_rcuidle(proto) \
231 if (static_key_false(&__tracepoint_##name.key)) \
234 TP_CONDITION(cond), 1); \
237 #define __DECLARE_TRACE_RCU(name, proto, args, cond)
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.
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.
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) \
257 if (static_key_false(&__tracepoint_##name.key)) \
260 TP_CONDITION(cond), 0); \
261 if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) { \
262 WARN_ON_ONCE(!rcu_is_watching()); \
265 __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args), \
268 register_trace_##name(void (*probe)(data_proto), void *data) \
270 return tracepoint_probe_register(&__tracepoint_##name, \
271 (void *)probe, data); \
274 register_trace_prio_##name(void (*probe)(data_proto), void *data,\
277 return tracepoint_probe_register_prio(&__tracepoint_##name, \
278 (void *)probe, data, prio); \
281 unregister_trace_##name(void (*probe)(data_proto), void *data) \
283 return tracepoint_probe_unregister(&__tracepoint_##name,\
284 (void *)probe, data); \
287 check_trace_callback_type_##name(void (*cb)(data_proto)) \
291 trace_##name##_enabled(void) \
293 return static_key_false(&__tracepoint_##name.key); \
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.
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 struct tracepoint __tracepoint_##_name __used \
307 __section("__tracepoints") = { \
308 .name = __tpstrtab_##_name, \
309 .key = STATIC_KEY_INIT_FALSE, \
310 .static_call_key = &STATIC_CALL_KEY(tp_func_##_name), \
311 .static_call_tramp = STATIC_CALL_TRAMP_ADDR(tp_func_##_name), \
312 .iterator = &__traceiter_##_name, \
314 .unregfunc = _unreg, \
316 __TRACEPOINT_ENTRY(_name); \
317 int __traceiter_##_name(void *__data, proto) \
319 struct tracepoint_func *it_func_ptr; \
323 rcu_dereference_raw((&__tracepoint_##_name)->funcs); \
326 it_func = READ_ONCE((it_func_ptr)->func); \
327 __data = (it_func_ptr)->data; \
328 ((void(*)(void *, proto))(it_func))(__data, args); \
329 } while ((++it_func_ptr)->func); \
333 DEFINE_STATIC_CALL(tp_func_##_name, __traceiter_##_name);
335 #define DEFINE_TRACE(name, proto, args) \
336 DEFINE_TRACE_FN(name, NULL, NULL, PARAMS(proto), PARAMS(args));
338 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \
339 EXPORT_SYMBOL_GPL(__tracepoint_##name); \
340 EXPORT_SYMBOL_GPL(__traceiter_##name); \
341 EXPORT_STATIC_CALL_GPL(tp_func_##name)
342 #define EXPORT_TRACEPOINT_SYMBOL(name) \
343 EXPORT_SYMBOL(__tracepoint_##name); \
344 EXPORT_SYMBOL(__traceiter_##name); \
345 EXPORT_STATIC_CALL(tp_func_##name)
348 #else /* !TRACEPOINTS_ENABLED */
349 #define __DECLARE_TRACE(name, proto, args, cond, data_proto) \
350 static inline void trace_##name(proto) \
352 static inline void trace_##name##_rcuidle(proto) \
355 register_trace_##name(void (*probe)(data_proto), \
361 unregister_trace_##name(void (*probe)(data_proto), \
366 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \
370 trace_##name##_enabled(void) \
375 #define DEFINE_TRACE_FN(name, reg, unreg, proto, args)
376 #define DEFINE_TRACE(name, proto, args)
377 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
378 #define EXPORT_TRACEPOINT_SYMBOL(name)
380 #endif /* TRACEPOINTS_ENABLED */
382 #ifdef CONFIG_TRACING
384 * tracepoint_string - register constant persistent string to trace system
385 * @str - a constant persistent string that will be referenced in tracepoints
387 * If constant strings are being used in tracepoints, it is faster and
388 * more efficient to just save the pointer to the string and reference
389 * that with a printf "%s" instead of saving the string in the ring buffer
390 * and wasting space and time.
392 * The problem with the above approach is that userspace tools that read
393 * the binary output of the trace buffers do not have access to the string.
394 * Instead they just show the address of the string which is not very
397 * With tracepoint_string(), the string will be registered to the tracing
398 * system and exported to userspace via the debugfs/tracing/printk_formats
399 * file that maps the string address to the string text. This way userspace
400 * tools that read the binary buffers have a way to map the pointers to
401 * the ASCII strings they represent.
403 * The @str used must be a constant string and persistent as it would not
404 * make sense to show a string that no longer exists. But it is still fine
405 * to be used with modules, because when modules are unloaded, if they
406 * had tracepoints, the ring buffers are cleared too. As long as the string
407 * does not change during the life of the module, it is fine to use
408 * tracepoint_string() within a module.
410 #define tracepoint_string(str) \
412 static const char *___tp_str __tracepoint_string = str; \
415 #define __tracepoint_string __used __section("__tracepoint_str")
418 * tracepoint_string() is used to save the string address for userspace
419 * tracing tools. When tracing isn't configured, there's no need to save
422 # define tracepoint_string(str) str
423 # define __tracepoint_string
426 #define DECLARE_TRACE(name, proto, args) \
427 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \
428 cpu_online(raw_smp_processor_id()), \
429 PARAMS(void *__data, proto))
431 #define DECLARE_TRACE_CONDITION(name, proto, args, cond) \
432 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \
433 cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \
434 PARAMS(void *__data, proto))
436 #define TRACE_EVENT_FLAGS(event, flag)
438 #define TRACE_EVENT_PERF_PERM(event, expr...)
440 #endif /* DECLARE_TRACE */
444 * For use with the TRACE_EVENT macro:
446 * We define a tracepoint, its arguments, its printk format
447 * and its 'fast binary record' layout.
449 * Firstly, name your tracepoint via TRACE_EVENT(name : the
450 * 'subsystem_event' notation is fine.
452 * Think about this whole construct as the
453 * 'trace_sched_switch() function' from now on.
456 * TRACE_EVENT(sched_switch,
459 * * A function has a regular function arguments
460 * * prototype, declare it via TP_PROTO():
463 * TP_PROTO(struct rq *rq, struct task_struct *prev,
464 * struct task_struct *next),
467 * * Define the call signature of the 'function'.
468 * * (Design sidenote: we use this instead of a
469 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
472 * TP_ARGS(rq, prev, next),
475 * * Fast binary tracing: define the trace record via
476 * * TP_STRUCT__entry(). You can think about it like a
477 * * regular C structure local variable definition.
479 * * This is how the trace record is structured and will
480 * * be saved into the ring buffer. These are the fields
481 * * that will be exposed to user-space in
482 * * /sys/kernel/tracing/events/<*>/format.
484 * * The declared 'local variable' is called '__entry'
486 * * __field(pid_t, prev_pid) is equivalent to a standard declaration:
490 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
492 * * char prev_comm[TASK_COMM_LEN];
496 * __array( char, prev_comm, TASK_COMM_LEN )
497 * __field( pid_t, prev_pid )
498 * __field( int, prev_prio )
499 * __array( char, next_comm, TASK_COMM_LEN )
500 * __field( pid_t, next_pid )
501 * __field( int, next_prio )
505 * * Assign the entry into the trace record, by embedding
506 * * a full C statement block into TP_fast_assign(). You
507 * * can refer to the trace record as '__entry' -
508 * * otherwise you can put arbitrary C code in here.
510 * * Note: this C code will execute every time a trace event
511 * * happens, on an active tracepoint.
515 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
516 * __entry->prev_pid = prev->pid;
517 * __entry->prev_prio = prev->prio;
518 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
519 * __entry->next_pid = next->pid;
520 * __entry->next_prio = next->prio;
524 * * Formatted output of a trace record via TP_printk().
525 * * This is how the tracepoint will appear under ftrace
526 * * plugins that make use of this tracepoint.
528 * * (raw-binary tracing wont actually perform this step.)
531 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
532 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
533 * __entry->next_comm, __entry->next_pid, __entry->next_prio),
537 * This macro construct is thus used for the regular printk format
538 * tracing setup, it is used to construct a function pointer based
539 * tracepoint callback (this is used by programmatic plugins and
540 * can also by used by generic instrumentation like SystemTap), and
541 * it is also used to expose a structured trace record in
542 * /sys/kernel/tracing/events/.
544 * A set of (un)registration functions can be passed to the variant
545 * TRACE_EVENT_FN to perform any (un)registration work.
548 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
549 #define DEFINE_EVENT(template, name, proto, args) \
550 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
551 #define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\
552 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
553 #define DEFINE_EVENT_PRINT(template, name, proto, args, print) \
554 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
555 #define DEFINE_EVENT_CONDITION(template, name, proto, \
557 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
558 PARAMS(args), PARAMS(cond))
560 #define TRACE_EVENT(name, proto, args, struct, assign, print) \
561 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
562 #define TRACE_EVENT_FN(name, proto, args, struct, \
563 assign, print, reg, unreg) \
564 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
565 #define TRACE_EVENT_FN_COND(name, proto, args, cond, struct, \
566 assign, print, reg, unreg) \
567 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
568 PARAMS(args), PARAMS(cond))
569 #define TRACE_EVENT_CONDITION(name, proto, args, cond, \
570 struct, assign, print) \
571 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
572 PARAMS(args), PARAMS(cond))
574 #define TRACE_EVENT_FLAGS(event, flag)
576 #define TRACE_EVENT_PERF_PERM(event, expr...)
578 #define DECLARE_EVENT_NOP(name, proto, args) \
579 static inline void trace_##name(proto) \
581 static inline bool trace_##name##_enabled(void) \
586 #define TRACE_EVENT_NOP(name, proto, args, struct, assign, print) \
587 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args))
589 #define DECLARE_EVENT_CLASS_NOP(name, proto, args, tstruct, assign, print)
590 #define DEFINE_EVENT_NOP(template, name, proto, args) \
591 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args))
593 #endif /* ifdef TRACE_EVENT (see note above) */