GNU Linux-libre 5.10.217-gnu1
[releases.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 #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
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
69 static inline bool trace_module_has_bad_taint(struct module *mod)
70 {
71         return false;
72 }
73 static inline
74 int register_tracepoint_module_notifier(struct notifier_block *nb)
75 {
76         return 0;
77 }
78 static inline
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
91 static inline void tracepoint_synchronize_unregister(void)
92 {
93         synchronize_srcu(&tracepoint_srcu);
94         synchronize_rcu();
95 }
96 #else
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
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
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)   static_call(tp_func_##name)
166 #else
167 #define __DO_TRACE_CALL(name)   __traceiter_##name
168 #endif /* CONFIG_HAVE_STATIC_CALL */
169
170 /*
171  * it_func[0] is never NULL because there is at least one element in the array
172  * when the array itself is non NULL.
173  *
174  * Note, the proto and args passed in includes "__data" as the first parameter.
175  * The reason for this is to handle the "void" prototype. If a tracepoint
176  * has a "void" prototype, then it is invalid to declare a function
177  * as "(void *, void)".
178  */
179 #define __DO_TRACE(name, proto, args, cond, rcuidle)                    \
180         do {                                                            \
181                 struct tracepoint_func *it_func_ptr;                    \
182                 int __maybe_unused __idx = 0;                           \
183                 void *__data;                                           \
184                                                                         \
185                 if (!(cond))                                            \
186                         return;                                         \
187                                                                         \
188                 /* srcu can't be used from NMI */                       \
189                 WARN_ON_ONCE(rcuidle && in_nmi());                      \
190                                                                         \
191                 /* keep srcu and sched-rcu usage consistent */          \
192                 preempt_disable_notrace();                              \
193                                                                         \
194                 /*                                                      \
195                  * For rcuidle callers, use srcu since sched-rcu        \
196                  * doesn't work from the idle path.                     \
197                  */                                                     \
198                 if (rcuidle) {                                          \
199                         __idx = srcu_read_lock_notrace(&tracepoint_srcu);\
200                         rcu_irq_enter_irqson();                         \
201                 }                                                       \
202                                                                         \
203                 it_func_ptr =                                           \
204                         rcu_dereference_raw((&__tracepoint_##name)->funcs); \
205                 if (it_func_ptr) {                                      \
206                         __data = (it_func_ptr)->data;                   \
207                         __DO_TRACE_CALL(name)(args);                    \
208                 }                                                       \
209                                                                         \
210                 if (rcuidle) {                                          \
211                         rcu_irq_exit_irqson();                          \
212                         srcu_read_unlock_notrace(&tracepoint_srcu, __idx);\
213                 }                                                       \
214                                                                         \
215                 preempt_enable_notrace();                               \
216         } while (0)
217
218 #ifndef MODULE
219 #define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args) \
220         static inline void trace_##name##_rcuidle(proto)                \
221         {                                                               \
222                 if (static_key_false(&__tracepoint_##name.key))         \
223                         __DO_TRACE(name,                                \
224                                 TP_PROTO(data_proto),                   \
225                                 TP_ARGS(data_args),                     \
226                                 TP_CONDITION(cond), 1);                 \
227         }
228 #else
229 #define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args)
230 #endif
231
232 /*
233  * Make sure the alignment of the structure in the __tracepoints section will
234  * not add unwanted padding between the beginning of the section and the
235  * structure. Force alignment to the same alignment as the section start.
236  *
237  * When lockdep is enabled, we make sure to always test if RCU is
238  * "watching" regardless if the tracepoint is enabled or not. Tracepoints
239  * require RCU to be active, and it should always warn at the tracepoint
240  * site if it is not watching, as it will need to be active when the
241  * tracepoint is enabled.
242  */
243 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
244         extern int __traceiter_##name(data_proto);                      \
245         DECLARE_STATIC_CALL(tp_func_##name, __traceiter_##name);        \
246         extern struct tracepoint __tracepoint_##name;                   \
247         static inline void trace_##name(proto)                          \
248         {                                                               \
249                 if (static_key_false(&__tracepoint_##name.key))         \
250                         __DO_TRACE(name,                                \
251                                 TP_PROTO(data_proto),                   \
252                                 TP_ARGS(data_args),                     \
253                                 TP_CONDITION(cond), 0);                 \
254                 if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) {             \
255                         WARN_ON_ONCE(!rcu_is_watching());               \
256                 }                                                       \
257         }                                                               \
258         __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args),          \
259                 PARAMS(cond), PARAMS(data_proto), PARAMS(data_args))    \
260         static inline int                                               \
261         register_trace_##name(void (*probe)(data_proto), void *data)    \
262         {                                                               \
263                 return tracepoint_probe_register(&__tracepoint_##name,  \
264                                                 (void *)probe, data);   \
265         }                                                               \
266         static inline int                                               \
267         register_trace_prio_##name(void (*probe)(data_proto), void *data,\
268                                    int prio)                            \
269         {                                                               \
270                 return tracepoint_probe_register_prio(&__tracepoint_##name, \
271                                               (void *)probe, data, prio); \
272         }                                                               \
273         static inline int                                               \
274         unregister_trace_##name(void (*probe)(data_proto), void *data)  \
275         {                                                               \
276                 return tracepoint_probe_unregister(&__tracepoint_##name,\
277                                                 (void *)probe, data);   \
278         }                                                               \
279         static inline void                                              \
280         check_trace_callback_type_##name(void (*cb)(data_proto))        \
281         {                                                               \
282         }                                                               \
283         static inline bool                                              \
284         trace_##name##_enabled(void)                                    \
285         {                                                               \
286                 return static_key_false(&__tracepoint_##name.key);      \
287         }
288
289 /*
290  * We have no guarantee that gcc and the linker won't up-align the tracepoint
291  * structures, so we create an array of pointers that will be used for iteration
292  * on the tracepoints.
293  */
294 #define DEFINE_TRACE_FN(_name, _reg, _unreg, proto, args)               \
295         static const char __tpstrtab_##_name[]                          \
296         __section("__tracepoints_strings") = #_name;                    \
297         extern struct static_call_key STATIC_CALL_KEY(tp_func_##_name); \
298         int __traceiter_##_name(void *__data, proto);                   \
299         struct tracepoint __tracepoint_##_name  __used                  \
300         __section("__tracepoints") = {                                  \
301                 .name = __tpstrtab_##_name,                             \
302                 .key = STATIC_KEY_INIT_FALSE,                           \
303                 .static_call_key = &STATIC_CALL_KEY(tp_func_##_name),   \
304                 .static_call_tramp = STATIC_CALL_TRAMP_ADDR(tp_func_##_name), \
305                 .iterator = &__traceiter_##_name,                       \
306                 .regfunc = _reg,                                        \
307                 .unregfunc = _unreg,                                    \
308                 .funcs = NULL };                                        \
309         __TRACEPOINT_ENTRY(_name);                                      \
310         int __traceiter_##_name(void *__data, proto)                    \
311         {                                                               \
312                 struct tracepoint_func *it_func_ptr;                    \
313                 void *it_func;                                          \
314                                                                         \
315                 it_func_ptr =                                           \
316                         rcu_dereference_raw((&__tracepoint_##_name)->funcs); \
317                 if (it_func_ptr) {                                      \
318                         do {                                            \
319                                 it_func = (it_func_ptr)->func;          \
320                                 __data = (it_func_ptr)->data;           \
321                                 ((void(*)(void *, proto))(it_func))(__data, args); \
322                         } while ((++it_func_ptr)->func);                \
323                 }                                                       \
324                 return 0;                                               \
325         }                                                               \
326         DEFINE_STATIC_CALL(tp_func_##_name, __traceiter_##_name);
327
328 #define DEFINE_TRACE(name, proto, args)         \
329         DEFINE_TRACE_FN(name, NULL, NULL, PARAMS(proto), PARAMS(args));
330
331 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)                              \
332         EXPORT_SYMBOL_GPL(__tracepoint_##name);                         \
333         EXPORT_SYMBOL_GPL(__traceiter_##name);                          \
334         EXPORT_STATIC_CALL_GPL(tp_func_##name)
335 #define EXPORT_TRACEPOINT_SYMBOL(name)                                  \
336         EXPORT_SYMBOL(__tracepoint_##name);                             \
337         EXPORT_SYMBOL(__traceiter_##name);                              \
338         EXPORT_STATIC_CALL(tp_func_##name)
339
340
341 #else /* !TRACEPOINTS_ENABLED */
342 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
343         static inline void trace_##name(proto)                          \
344         { }                                                             \
345         static inline void trace_##name##_rcuidle(proto)                \
346         { }                                                             \
347         static inline int                                               \
348         register_trace_##name(void (*probe)(data_proto),                \
349                               void *data)                               \
350         {                                                               \
351                 return -ENOSYS;                                         \
352         }                                                               \
353         static inline int                                               \
354         unregister_trace_##name(void (*probe)(data_proto),              \
355                                 void *data)                             \
356         {                                                               \
357                 return -ENOSYS;                                         \
358         }                                                               \
359         static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \
360         {                                                               \
361         }                                                               \
362         static inline bool                                              \
363         trace_##name##_enabled(void)                                    \
364         {                                                               \
365                 return false;                                           \
366         }
367
368 #define DEFINE_TRACE_FN(name, reg, unreg, proto, args)
369 #define DEFINE_TRACE(name, proto, args)
370 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
371 #define EXPORT_TRACEPOINT_SYMBOL(name)
372
373 #endif /* TRACEPOINTS_ENABLED */
374
375 #ifdef CONFIG_TRACING
376 /**
377  * tracepoint_string - register constant persistent string to trace system
378  * @str - a constant persistent string that will be referenced in tracepoints
379  *
380  * If constant strings are being used in tracepoints, it is faster and
381  * more efficient to just save the pointer to the string and reference
382  * that with a printf "%s" instead of saving the string in the ring buffer
383  * and wasting space and time.
384  *
385  * The problem with the above approach is that userspace tools that read
386  * the binary output of the trace buffers do not have access to the string.
387  * Instead they just show the address of the string which is not very
388  * useful to users.
389  *
390  * With tracepoint_string(), the string will be registered to the tracing
391  * system and exported to userspace via the debugfs/tracing/printk_formats
392  * file that maps the string address to the string text. This way userspace
393  * tools that read the binary buffers have a way to map the pointers to
394  * the ASCII strings they represent.
395  *
396  * The @str used must be a constant string and persistent as it would not
397  * make sense to show a string that no longer exists. But it is still fine
398  * to be used with modules, because when modules are unloaded, if they
399  * had tracepoints, the ring buffers are cleared too. As long as the string
400  * does not change during the life of the module, it is fine to use
401  * tracepoint_string() within a module.
402  */
403 #define tracepoint_string(str)                                          \
404         ({                                                              \
405                 static const char *___tp_str __tracepoint_string = str; \
406                 ___tp_str;                                              \
407         })
408 #define __tracepoint_string     __used __section("__tracepoint_str")
409 #else
410 /*
411  * tracepoint_string() is used to save the string address for userspace
412  * tracing tools. When tracing isn't configured, there's no need to save
413  * anything.
414  */
415 # define tracepoint_string(str) str
416 # define __tracepoint_string
417 #endif
418
419 #define DECLARE_TRACE(name, proto, args)                                \
420         __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args),              \
421                         cpu_online(raw_smp_processor_id()),             \
422                         PARAMS(void *__data, proto),                    \
423                         PARAMS(__data, args))
424
425 #define DECLARE_TRACE_CONDITION(name, proto, args, cond)                \
426         __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args),              \
427                         cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \
428                         PARAMS(void *__data, proto),                    \
429                         PARAMS(__data, args))
430
431 #define TRACE_EVENT_FLAGS(event, flag)
432
433 #define TRACE_EVENT_PERF_PERM(event, expr...)
434
435 #endif /* DECLARE_TRACE */
436
437 #ifndef TRACE_EVENT
438 /*
439  * For use with the TRACE_EVENT macro:
440  *
441  * We define a tracepoint, its arguments, its printk format
442  * and its 'fast binary record' layout.
443  *
444  * Firstly, name your tracepoint via TRACE_EVENT(name : the
445  * 'subsystem_event' notation is fine.
446  *
447  * Think about this whole construct as the
448  * 'trace_sched_switch() function' from now on.
449  *
450  *
451  *  TRACE_EVENT(sched_switch,
452  *
453  *      *
454  *      * A function has a regular function arguments
455  *      * prototype, declare it via TP_PROTO():
456  *      *
457  *
458  *      TP_PROTO(struct rq *rq, struct task_struct *prev,
459  *               struct task_struct *next),
460  *
461  *      *
462  *      * Define the call signature of the 'function'.
463  *      * (Design sidenote: we use this instead of a
464  *      *  TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
465  *      *
466  *
467  *      TP_ARGS(rq, prev, next),
468  *
469  *      *
470  *      * Fast binary tracing: define the trace record via
471  *      * TP_STRUCT__entry(). You can think about it like a
472  *      * regular C structure local variable definition.
473  *      *
474  *      * This is how the trace record is structured and will
475  *      * be saved into the ring buffer. These are the fields
476  *      * that will be exposed to user-space in
477  *      * /sys/kernel/debug/tracing/events/<*>/format.
478  *      *
479  *      * The declared 'local variable' is called '__entry'
480  *      *
481  *      * __field(pid_t, prev_prid) is equivalent to a standard declariton:
482  *      *
483  *      *       pid_t   prev_pid;
484  *      *
485  *      * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
486  *      *
487  *      *       char    prev_comm[TASK_COMM_LEN];
488  *      *
489  *
490  *      TP_STRUCT__entry(
491  *              __array(        char,   prev_comm,      TASK_COMM_LEN   )
492  *              __field(        pid_t,  prev_pid                        )
493  *              __field(        int,    prev_prio                       )
494  *              __array(        char,   next_comm,      TASK_COMM_LEN   )
495  *              __field(        pid_t,  next_pid                        )
496  *              __field(        int,    next_prio                       )
497  *      ),
498  *
499  *      *
500  *      * Assign the entry into the trace record, by embedding
501  *      * a full C statement block into TP_fast_assign(). You
502  *      * can refer to the trace record as '__entry' -
503  *      * otherwise you can put arbitrary C code in here.
504  *      *
505  *      * Note: this C code will execute every time a trace event
506  *      * happens, on an active tracepoint.
507  *      *
508  *
509  *      TP_fast_assign(
510  *              memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
511  *              __entry->prev_pid       = prev->pid;
512  *              __entry->prev_prio      = prev->prio;
513  *              memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
514  *              __entry->next_pid       = next->pid;
515  *              __entry->next_prio      = next->prio;
516  *      ),
517  *
518  *      *
519  *      * Formatted output of a trace record via TP_printk().
520  *      * This is how the tracepoint will appear under ftrace
521  *      * plugins that make use of this tracepoint.
522  *      *
523  *      * (raw-binary tracing wont actually perform this step.)
524  *      *
525  *
526  *      TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
527  *              __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
528  *              __entry->next_comm, __entry->next_pid, __entry->next_prio),
529  *
530  * );
531  *
532  * This macro construct is thus used for the regular printk format
533  * tracing setup, it is used to construct a function pointer based
534  * tracepoint callback (this is used by programmatic plugins and
535  * can also by used by generic instrumentation like SystemTap), and
536  * it is also used to expose a structured trace record in
537  * /sys/kernel/debug/tracing/events/.
538  *
539  * A set of (un)registration functions can be passed to the variant
540  * TRACE_EVENT_FN to perform any (un)registration work.
541  */
542
543 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
544 #define DEFINE_EVENT(template, name, proto, args)               \
545         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
546 #define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\
547         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
548 #define DEFINE_EVENT_PRINT(template, name, proto, args, print)  \
549         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
550 #define DEFINE_EVENT_CONDITION(template, name, proto,           \
551                                args, cond)                      \
552         DECLARE_TRACE_CONDITION(name, PARAMS(proto),            \
553                                 PARAMS(args), PARAMS(cond))
554
555 #define TRACE_EVENT(name, proto, args, struct, assign, print)   \
556         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
557 #define TRACE_EVENT_FN(name, proto, args, struct,               \
558                 assign, print, reg, unreg)                      \
559         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
560 #define TRACE_EVENT_FN_COND(name, proto, args, cond, struct,            \
561                 assign, print, reg, unreg)                      \
562         DECLARE_TRACE_CONDITION(name, PARAMS(proto),    \
563                         PARAMS(args), PARAMS(cond))
564 #define TRACE_EVENT_CONDITION(name, proto, args, cond,          \
565                               struct, assign, print)            \
566         DECLARE_TRACE_CONDITION(name, PARAMS(proto),            \
567                                 PARAMS(args), PARAMS(cond))
568
569 #define TRACE_EVENT_FLAGS(event, flag)
570
571 #define TRACE_EVENT_PERF_PERM(event, expr...)
572
573 #define DECLARE_EVENT_NOP(name, proto, args)                            \
574         static inline void trace_##name(proto)                          \
575         { }                                                             \
576         static inline bool trace_##name##_enabled(void)                 \
577         {                                                               \
578                 return false;                                           \
579         }
580
581 #define TRACE_EVENT_NOP(name, proto, args, struct, assign, print)       \
582         DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args))
583
584 #define DECLARE_EVENT_CLASS_NOP(name, proto, args, tstruct, assign, print)
585 #define DEFINE_EVENT_NOP(template, name, proto, args)                   \
586         DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args))
587
588 #endif /* ifdef TRACE_EVENT (see note above) */