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/rcupdate.h>
20 #include <linux/rcupdate_trace.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
37 tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data);
39 tracepoint_probe_register_prio(struct tracepoint *tp, void *probe, void *data,
42 tracepoint_probe_register_prio_may_exist(struct tracepoint *tp, void *probe, void *data,
45 tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data);
47 tracepoint_probe_register_may_exist(struct tracepoint *tp, void *probe,
50 return tracepoint_probe_register_prio_may_exist(tp, probe, data,
51 TRACEPOINT_DEFAULT_PRIO);
54 for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
59 struct list_head list;
63 bool trace_module_has_bad_taint(struct module *mod);
64 extern int register_tracepoint_module_notifier(struct notifier_block *nb);
65 extern int unregister_tracepoint_module_notifier(struct notifier_block *nb);
66 void for_each_module_tracepoint(void (*fct)(struct tracepoint *,
67 struct module *, void *),
69 void for_each_tracepoint_in_module(struct module *,
70 void (*fct)(struct tracepoint *,
71 struct module *, void *),
74 static inline bool trace_module_has_bad_taint(struct module *mod)
79 int register_tracepoint_module_notifier(struct notifier_block *nb)
84 int unregister_tracepoint_module_notifier(struct notifier_block *nb)
89 void for_each_module_tracepoint(void (*fct)(struct tracepoint *,
90 struct module *, void *),
95 void for_each_tracepoint_in_module(struct module *mod,
96 void (*fct)(struct tracepoint *,
97 struct module *, void *),
101 #endif /* CONFIG_MODULES */
104 * tracepoint_synchronize_unregister must be called between the last tracepoint
105 * probe unregistration and the end of module exit to make sure there is no
106 * caller executing a probe when it is freed.
108 * An alternative is to use the following for batch reclaim associated
109 * with a given tracepoint:
111 * - tracepoint_is_faultable() == false: call_rcu()
112 * - tracepoint_is_faultable() == true: call_rcu_tasks_trace()
114 #ifdef CONFIG_TRACEPOINTS
115 static inline void tracepoint_synchronize_unregister(void)
117 synchronize_rcu_tasks_trace();
120 static inline bool tracepoint_is_faultable(struct tracepoint *tp)
122 return tp->ext && tp->ext->faultable;
125 static inline void tracepoint_synchronize_unregister(void)
127 static inline bool tracepoint_is_faultable(struct tracepoint *tp)
133 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
134 extern int syscall_regfunc(void);
135 extern void syscall_unregfunc(void);
136 #endif /* CONFIG_HAVE_SYSCALL_TRACEPOINTS */
139 #define PARAMS(args...) args
142 #define TRACE_DEFINE_ENUM(x)
143 #define TRACE_DEFINE_SIZEOF(x)
145 #ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS
146 static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p)
148 return offset_to_ptr(p);
151 #define __TRACEPOINT_ENTRY(name) \
152 asm(" .section \"__tracepoints_ptrs\", \"a\" \n" \
154 " .long __tracepoint_" #name " - . \n" \
157 static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p)
162 #define __TRACEPOINT_ENTRY(name) \
163 static tracepoint_ptr_t __tracepoint_ptr_##name __used \
164 __section("__tracepoints_ptrs") = &__tracepoint_##name
167 #endif /* _LINUX_TRACEPOINT_H */
170 * Note: we keep the TRACE_EVENT and DECLARE_TRACE outside the include
171 * file ifdef protection.
172 * This is due to the way trace events work. If a file includes two
173 * trace event headers under one "CREATE_TRACE_POINTS" the first include
174 * will override the TRACE_EVENT and break the second include.
177 #ifndef DECLARE_TRACE
179 #define TP_PROTO(args...) args
180 #define TP_ARGS(args...) args
181 #define TP_CONDITION(args...) args
184 * Individual subsystem my have a separate configuration to
185 * enable their tracepoints. By default, this file will create
186 * the tracepoints if CONFIG_TRACEPOINTS is defined. If a subsystem
187 * wants to be able to disable its tracepoints from being created
188 * it can define NOTRACE before including the tracepoint headers.
190 #if defined(CONFIG_TRACEPOINTS) && !defined(NOTRACE)
191 #define TRACEPOINTS_ENABLED
194 #ifdef TRACEPOINTS_ENABLED
196 #ifdef CONFIG_HAVE_STATIC_CALL
197 #define __DO_TRACE_CALL(name, args) \
199 struct tracepoint_func *it_func_ptr; \
202 rcu_dereference_raw((&__tracepoint_##name)->funcs); \
204 __data = (it_func_ptr)->data; \
205 static_call(tp_func_##name)(__data, args); \
209 #define __DO_TRACE_CALL(name, args) __traceiter_##name(NULL, args)
210 #endif /* CONFIG_HAVE_STATIC_CALL */
213 * it_func[0] is never NULL because there is at least one element in the array
214 * when the array itself is non NULL.
216 * With @syscall=0, the tracepoint callback array dereference is
217 * protected by disabling preemption.
218 * With @syscall=1, the tracepoint callback array dereference is
219 * protected by Tasks Trace RCU, which allows probes to handle page
222 #define __DO_TRACE(name, args, cond, syscall) \
224 int __maybe_unused __idx = 0; \
230 rcu_read_lock_trace(); \
232 preempt_disable_notrace(); \
234 __DO_TRACE_CALL(name, TP_ARGS(args)); \
237 rcu_read_unlock_trace(); \
239 preempt_enable_notrace(); \
243 * Declare an exported function that Rust code can call to trigger this
244 * tracepoint. This function does not include the static branch; that is done
245 * in Rust to avoid a function call when the tracepoint is disabled.
247 #define DEFINE_RUST_DO_TRACE(name, proto, args)
248 #define __DEFINE_RUST_DO_TRACE(name, proto, args) \
249 notrace void rust_do_trace_##name(proto) \
251 __rust_do_trace_##name(args); \
255 * Make sure the alignment of the structure in the __tracepoints section will
256 * not add unwanted padding between the beginning of the section and the
257 * structure. Force alignment to the same alignment as the section start.
259 * When lockdep is enabled, we make sure to always test if RCU is
260 * "watching" regardless if the tracepoint is enabled or not. Tracepoints
261 * require RCU to be active, and it should always warn at the tracepoint
262 * site if it is not watching, as it will need to be active when the
263 * tracepoint is enabled.
265 #define __DECLARE_TRACE_COMMON(name, proto, args, cond, data_proto) \
266 extern int __traceiter_##name(data_proto); \
267 DECLARE_STATIC_CALL(tp_func_##name, __traceiter_##name); \
268 extern struct tracepoint __tracepoint_##name; \
269 extern void rust_do_trace_##name(proto); \
271 register_trace_##name(void (*probe)(data_proto), void *data) \
273 return tracepoint_probe_register(&__tracepoint_##name, \
274 (void *)probe, data); \
277 register_trace_prio_##name(void (*probe)(data_proto), void *data,\
280 return tracepoint_probe_register_prio(&__tracepoint_##name, \
281 (void *)probe, data, prio); \
284 unregister_trace_##name(void (*probe)(data_proto), void *data) \
286 return tracepoint_probe_unregister(&__tracepoint_##name,\
287 (void *)probe, data); \
290 check_trace_callback_type_##name(void (*cb)(data_proto)) \
294 trace_##name##_enabled(void) \
296 return static_branch_unlikely(&__tracepoint_##name.key);\
299 #define __DECLARE_TRACE(name, proto, args, cond, data_proto) \
300 __DECLARE_TRACE_COMMON(name, PARAMS(proto), PARAMS(args), cond, PARAMS(data_proto)) \
301 static inline void __rust_do_trace_##name(proto) \
305 TP_CONDITION(cond), 0); \
307 static inline void trace_##name(proto) \
309 if (static_branch_unlikely(&__tracepoint_##name.key)) \
312 TP_CONDITION(cond), 0); \
313 if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) { \
314 WARN_ONCE(!rcu_is_watching(), \
315 "RCU not watching for tracepoint"); \
319 #define __DECLARE_TRACE_SYSCALL(name, proto, args, cond, data_proto) \
320 __DECLARE_TRACE_COMMON(name, PARAMS(proto), PARAMS(args), cond, PARAMS(data_proto)) \
321 static inline void __rust_do_trace_##name(proto) \
325 TP_CONDITION(cond), 1); \
327 static inline void trace_##name(proto) \
330 if (static_branch_unlikely(&__tracepoint_##name.key)) \
333 TP_CONDITION(cond), 1); \
334 if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) { \
335 WARN_ONCE(!rcu_is_watching(), \
336 "RCU not watching for tracepoint"); \
341 * We have no guarantee that gcc and the linker won't up-align the tracepoint
342 * structures, so we create an array of pointers that will be used for iteration
343 * on the tracepoints.
345 #define __DEFINE_TRACE_EXT(_name, _ext, proto, args) \
346 static const char __tpstrtab_##_name[] \
347 __section("__tracepoints_strings") = #_name; \
348 extern struct static_call_key STATIC_CALL_KEY(tp_func_##_name); \
349 int __traceiter_##_name(void *__data, proto); \
350 void __probestub_##_name(void *__data, proto); \
351 struct tracepoint __tracepoint_##_name __used \
352 __section("__tracepoints") = { \
353 .name = __tpstrtab_##_name, \
354 .key = STATIC_KEY_FALSE_INIT, \
355 .static_call_key = &STATIC_CALL_KEY(tp_func_##_name), \
356 .static_call_tramp = STATIC_CALL_TRAMP_ADDR(tp_func_##_name), \
357 .iterator = &__traceiter_##_name, \
358 .probestub = &__probestub_##_name, \
362 __TRACEPOINT_ENTRY(_name); \
363 int __traceiter_##_name(void *__data, proto) \
365 struct tracepoint_func *it_func_ptr; \
369 rcu_dereference_raw((&__tracepoint_##_name)->funcs); \
372 it_func = READ_ONCE((it_func_ptr)->func); \
373 __data = (it_func_ptr)->data; \
374 ((void(*)(void *, proto))(it_func))(__data, args); \
375 } while ((++it_func_ptr)->func); \
379 void __probestub_##_name(void *__data, proto) \
382 DEFINE_STATIC_CALL(tp_func_##_name, __traceiter_##_name); \
383 DEFINE_RUST_DO_TRACE(_name, TP_PROTO(proto), TP_ARGS(args))
385 #define DEFINE_TRACE_FN(_name, _reg, _unreg, _proto, _args) \
386 static struct tracepoint_ext __tracepoint_ext_##_name = { \
388 .unregfunc = _unreg, \
389 .faultable = false, \
391 __DEFINE_TRACE_EXT(_name, &__tracepoint_ext_##_name, PARAMS(_proto), PARAMS(_args));
393 #define DEFINE_TRACE_SYSCALL(_name, _reg, _unreg, _proto, _args) \
394 static struct tracepoint_ext __tracepoint_ext_##_name = { \
396 .unregfunc = _unreg, \
399 __DEFINE_TRACE_EXT(_name, &__tracepoint_ext_##_name, PARAMS(_proto), PARAMS(_args));
401 #define DEFINE_TRACE(_name, _proto, _args) \
402 __DEFINE_TRACE_EXT(_name, NULL, PARAMS(_proto), PARAMS(_args));
404 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \
405 EXPORT_SYMBOL_GPL(__tracepoint_##name); \
406 EXPORT_SYMBOL_GPL(__traceiter_##name); \
407 EXPORT_STATIC_CALL_GPL(tp_func_##name)
408 #define EXPORT_TRACEPOINT_SYMBOL(name) \
409 EXPORT_SYMBOL(__tracepoint_##name); \
410 EXPORT_SYMBOL(__traceiter_##name); \
411 EXPORT_STATIC_CALL(tp_func_##name)
414 #else /* !TRACEPOINTS_ENABLED */
415 #define __DECLARE_TRACE(name, proto, args, cond, data_proto) \
416 static inline void trace_##name(proto) \
419 register_trace_##name(void (*probe)(data_proto), \
425 unregister_trace_##name(void (*probe)(data_proto), \
430 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \
434 trace_##name##_enabled(void) \
439 #define __DECLARE_TRACE_SYSCALL __DECLARE_TRACE
441 #define DEFINE_TRACE_FN(name, reg, unreg, proto, args)
442 #define DEFINE_TRACE_SYSCALL(name, reg, unreg, proto, args)
443 #define DEFINE_TRACE(name, proto, args)
444 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
445 #define EXPORT_TRACEPOINT_SYMBOL(name)
447 #endif /* TRACEPOINTS_ENABLED */
449 #ifdef CONFIG_TRACING
451 * tracepoint_string - register constant persistent string to trace system
452 * @str - a constant persistent string that will be referenced in tracepoints
454 * If constant strings are being used in tracepoints, it is faster and
455 * more efficient to just save the pointer to the string and reference
456 * that with a printf "%s" instead of saving the string in the ring buffer
457 * and wasting space and time.
459 * The problem with the above approach is that userspace tools that read
460 * the binary output of the trace buffers do not have access to the string.
461 * Instead they just show the address of the string which is not very
464 * With tracepoint_string(), the string will be registered to the tracing
465 * system and exported to userspace via the debugfs/tracing/printk_formats
466 * file that maps the string address to the string text. This way userspace
467 * tools that read the binary buffers have a way to map the pointers to
468 * the ASCII strings they represent.
470 * The @str used must be a constant string and persistent as it would not
471 * make sense to show a string that no longer exists. But it is still fine
472 * to be used with modules, because when modules are unloaded, if they
473 * had tracepoints, the ring buffers are cleared too. As long as the string
474 * does not change during the life of the module, it is fine to use
475 * tracepoint_string() within a module.
477 #define tracepoint_string(str) \
479 static const char *___tp_str __tracepoint_string = str; \
482 #define __tracepoint_string __used __section("__tracepoint_str")
485 * tracepoint_string() is used to save the string address for userspace
486 * tracing tools. When tracing isn't configured, there's no need to save
489 # define tracepoint_string(str) str
490 # define __tracepoint_string
493 #define DECLARE_TRACE(name, proto, args) \
494 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \
495 cpu_online(raw_smp_processor_id()), \
496 PARAMS(void *__data, proto))
498 #define DECLARE_TRACE_CONDITION(name, proto, args, cond) \
499 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \
500 cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \
501 PARAMS(void *__data, proto))
503 #define DECLARE_TRACE_SYSCALL(name, proto, args) \
504 __DECLARE_TRACE_SYSCALL(name, PARAMS(proto), PARAMS(args), \
505 cpu_online(raw_smp_processor_id()), \
506 PARAMS(void *__data, proto))
508 #define TRACE_EVENT_FLAGS(event, flag)
510 #define TRACE_EVENT_PERF_PERM(event, expr...)
512 #endif /* DECLARE_TRACE */
516 * For use with the TRACE_EVENT macro:
518 * We define a tracepoint, its arguments, its printk format
519 * and its 'fast binary record' layout.
521 * Firstly, name your tracepoint via TRACE_EVENT(name : the
522 * 'subsystem_event' notation is fine.
524 * Think about this whole construct as the
525 * 'trace_sched_switch() function' from now on.
528 * TRACE_EVENT(sched_switch,
531 * * A function has a regular function arguments
532 * * prototype, declare it via TP_PROTO():
535 * TP_PROTO(struct rq *rq, struct task_struct *prev,
536 * struct task_struct *next),
539 * * Define the call signature of the 'function'.
540 * * (Design sidenote: we use this instead of a
541 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
544 * TP_ARGS(rq, prev, next),
547 * * Fast binary tracing: define the trace record via
548 * * TP_STRUCT__entry(). You can think about it like a
549 * * regular C structure local variable definition.
551 * * This is how the trace record is structured and will
552 * * be saved into the ring buffer. These are the fields
553 * * that will be exposed to user-space in
554 * * /sys/kernel/tracing/events/<*>/format.
556 * * The declared 'local variable' is called '__entry'
558 * * __field(pid_t, prev_pid) is equivalent to a standard declaration:
562 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
564 * * char prev_comm[TASK_COMM_LEN];
568 * __array( char, prev_comm, TASK_COMM_LEN )
569 * __field( pid_t, prev_pid )
570 * __field( int, prev_prio )
571 * __array( char, next_comm, TASK_COMM_LEN )
572 * __field( pid_t, next_pid )
573 * __field( int, next_prio )
577 * * Assign the entry into the trace record, by embedding
578 * * a full C statement block into TP_fast_assign(). You
579 * * can refer to the trace record as '__entry' -
580 * * otherwise you can put arbitrary C code in here.
582 * * Note: this C code will execute every time a trace event
583 * * happens, on an active tracepoint.
587 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
588 * __entry->prev_pid = prev->pid;
589 * __entry->prev_prio = prev->prio;
590 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
591 * __entry->next_pid = next->pid;
592 * __entry->next_prio = next->prio;
596 * * Formatted output of a trace record via TP_printk().
597 * * This is how the tracepoint will appear under ftrace
598 * * plugins that make use of this tracepoint.
600 * * (raw-binary tracing wont actually perform this step.)
603 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
604 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
605 * __entry->next_comm, __entry->next_pid, __entry->next_prio),
609 * This macro construct is thus used for the regular printk format
610 * tracing setup, it is used to construct a function pointer based
611 * tracepoint callback (this is used by programmatic plugins and
612 * can also by used by generic instrumentation like SystemTap), and
613 * it is also used to expose a structured trace record in
614 * /sys/kernel/tracing/events/.
616 * A set of (un)registration functions can be passed to the variant
617 * TRACE_EVENT_FN to perform any (un)registration work.
620 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
621 #define DEFINE_EVENT(template, name, proto, args) \
622 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
623 #define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\
624 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
625 #define DEFINE_EVENT_PRINT(template, name, proto, args, print) \
626 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
627 #define DEFINE_EVENT_CONDITION(template, name, proto, \
629 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
630 PARAMS(args), PARAMS(cond))
632 #define TRACE_EVENT(name, proto, args, struct, assign, print) \
633 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
634 #define TRACE_EVENT_FN(name, proto, args, struct, \
635 assign, print, reg, unreg) \
636 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
637 #define TRACE_EVENT_FN_COND(name, proto, args, cond, struct, \
638 assign, print, reg, unreg) \
639 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
640 PARAMS(args), PARAMS(cond))
641 #define TRACE_EVENT_CONDITION(name, proto, args, cond, \
642 struct, assign, print) \
643 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \
644 PARAMS(args), PARAMS(cond))
645 #define TRACE_EVENT_SYSCALL(name, proto, args, struct, assign, \
647 DECLARE_TRACE_SYSCALL(name, PARAMS(proto), PARAMS(args))
649 #define TRACE_EVENT_FLAGS(event, flag)
651 #define TRACE_EVENT_PERF_PERM(event, expr...)
653 #define DECLARE_EVENT_NOP(name, proto, args) \
654 static inline void trace_##name(proto) \
656 static inline bool trace_##name##_enabled(void) \
661 #define TRACE_EVENT_NOP(name, proto, args, struct, assign, print) \
662 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args))
664 #define DECLARE_EVENT_CLASS_NOP(name, proto, args, tstruct, assign, print)
665 #define DEFINE_EVENT_NOP(template, name, proto, args) \
666 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args))
668 #endif /* ifdef TRACE_EVENT (see note above) */