LXR linux/include/linux/tracepoint.h

   1#ifndef _LINUX_TRACEPOINT_H
   2#define _LINUX_TRACEPOINT_H
   3
   4/*
   5 * Kernel Tracepoint API.
   6 *
   7 * See Documentation/trace/tracepoints.txt.
   8 *
   9 * Copyright (C) 2008-2014 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
  10 *
  11 * Heavily inspired from the Linux Kernel Markers.
  12 *
  13 * This file is released under the GPLv2.
  14 * See the file COPYING for more details.
  15 */
  16
  17#include <linux/smp.h>
  18#include <linux/errno.h>
  19#include <linux/types.h>
  20#include <linux/cpumask.h>
  21#include <linux/rcupdate.h>
  22#include <linux/tracepoint-defs.h>
  23
  24struct module;
  25struct tracepoint;
  26struct notifier_block;
  27
  28struct trace_enum_map {
  29        const char              *system;
  30        const char              *enum_string;
  31        unsigned long           enum_value;
  32};
  33
  34#define TRACEPOINT_DEFAULT_PRIO 10
  35
  36extern int
  37tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data);
  38extern int
  39tracepoint_probe_register_prio(struct tracepoint *tp, void *probe, void *data,
  40                               int prio);
  41extern int
  42tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data);
  43extern void
  44for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv),
  45                void *priv);
  46
  47#ifdef CONFIG_MODULES
  48struct tp_module {
  49        struct list_head list;
  50        struct module *mod;
  51};
  52
  53bool trace_module_has_bad_taint(struct module *mod);
  54extern int register_tracepoint_module_notifier(struct notifier_block *nb);
  55extern int unregister_tracepoint_module_notifier(struct notifier_block *nb);
  56#else
  57static inline bool trace_module_has_bad_taint(struct module *mod)
  58{
  59        return false;
  60}
  61static inline
  62int register_tracepoint_module_notifier(struct notifier_block *nb)
  63{
  64        return 0;
  65}
  66static inline
  67int unregister_tracepoint_module_notifier(struct notifier_block *nb)
  68{
  69        return 0;
  70}
  71#endif /* CONFIG_MODULES */
  72
  73/*
  74 * tracepoint_synchronize_unregister must be called between the last tracepoint
  75 * probe unregistration and the end of module exit to make sure there is no
  76 * caller executing a probe when it is freed.
  77 */
  78static inline void tracepoint_synchronize_unregister(void)
  79{
  80        synchronize_sched();
  81}
  82
  83#ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
  84extern void syscall_regfunc(void);
  85extern void syscall_unregfunc(void);
  86#endif /* CONFIG_HAVE_SYSCALL_TRACEPOINTS */
  87
  88#define PARAMS(args...) args
  89
  90#define TRACE_DEFINE_ENUM(x)
  91
  92#endif /* _LINUX_TRACEPOINT_H */
  93
  94/*
  95 * Note: we keep the TRACE_EVENT and DECLARE_TRACE outside the include
  96 *  file ifdef protection.
  97 *  This is due to the way trace events work. If a file includes two
  98 *  trace event headers under one "CREATE_TRACE_POINTS" the first include
  99 *  will override the TRACE_EVENT and break the second include.
 100 */
 101
 102#ifndef DECLARE_TRACE
 103
 104#define TP_PROTO(args...)       args
 105#define TP_ARGS(args...)        args
 106#define TP_CONDITION(args...)   args
 107
 108/*
 109 * Individual subsystem my have a separate configuration to
 110 * enable their tracepoints. By default, this file will create
 111 * the tracepoints if CONFIG_TRACEPOINT is defined. If a subsystem
 112 * wants to be able to disable its tracepoints from being created
 113 * it can define NOTRACE before including the tracepoint headers.
 114 */
 115#if defined(CONFIG_TRACEPOINTS) && !defined(NOTRACE)
 116#define TRACEPOINTS_ENABLED
 117#endif
 118
 119#ifdef TRACEPOINTS_ENABLED
 120
 121/*
 122 * it_func[0] is never NULL because there is at least one element in the array
 123 * when the array itself is non NULL.
 124 *
 125 * Note, the proto and args passed in includes "__data" as the first parameter.
 126 * The reason for this is to handle the "void" prototype. If a tracepoint
 127 * has a "void" prototype, then it is invalid to declare a function
 128 * as "(void *, void)". The DECLARE_TRACE_NOARGS() will pass in just
 129 * "void *data", where as the DECLARE_TRACE() will pass in "void *data, proto".
 130 */
 131#define __DO_TRACE(tp, proto, args, cond, prercu, postrcu)              \
 132        do {                                                            \
 133                struct tracepoint_func *it_func_ptr;                    \
 134                void *it_func;                                          \
 135                void *__data;                                           \
 136                                                                        \
 137                if (!(cond))                                            \
 138                        return;                                         \
 139                prercu;                                                 \
 140                rcu_read_lock_sched_notrace();                          \
 141                it_func_ptr = rcu_dereference_sched((tp)->funcs);       \
 142                if (it_func_ptr) {                                      \
 143                        do {                                            \
 144                                it_func = (it_func_ptr)->func;          \
 145                                __data = (it_func_ptr)->data;           \
 146                                ((void(*)(proto))(it_func))(args);      \
 147                        } while ((++it_func_ptr)->func);                \
 148                }                                                       \
 149                rcu_read_unlock_sched_notrace();                        \
 150                postrcu;                                                \
 151        } while (0)
 152
 153#ifndef MODULE
 154#define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args)     \
 155        static inline void trace_##name##_rcuidle(proto)                \
 156        {                                                               \
 157                if (static_key_false(&__tracepoint_##name.key))         \
 158                        __DO_TRACE(&__tracepoint_##name,                \
 159                                TP_PROTO(data_proto),                   \
 160                                TP_ARGS(data_args),                     \
 161                                TP_CONDITION(cond),                     \
 162                                rcu_irq_enter_irqson(),                 \
 163                                rcu_irq_exit_irqson());                 \
 164        }
 165#else
 166#define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args)
 167#endif
 168
 169/*
 170 * Make sure the alignment of the structure in the __tracepoints section will
 171 * not add unwanted padding between the beginning of the section and the
 172 * structure. Force alignment to the same alignment as the section start.
 173 *
 174 * When lockdep is enabled, we make sure to always do the RCU portions of
 175 * the tracepoint code, regardless of whether tracing is on. However,
 176 * don't check if the condition is false, due to interaction with idle
 177 * instrumentation. This lets us find RCU issues triggered with tracepoints
 178 * even when this tracepoint is off. This code has no purpose other than
 179 * poking RCU a bit.
 180 */
 181#define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
 182        extern struct tracepoint __tracepoint_##name;                   \
 183        static inline void trace_##name(proto)                          \
 184        {                                                               \
 185                if (static_key_false(&__tracepoint_##name.key))         \
 186                        __DO_TRACE(&__tracepoint_##name,                \
 187                                TP_PROTO(data_proto),                   \
 188                                TP_ARGS(data_args),                     \
 189                                TP_CONDITION(cond),,);                  \
 190                if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) {             \
 191                        rcu_read_lock_sched_notrace();                  \
 192                        rcu_dereference_sched(__tracepoint_##name.funcs);\
 193                        rcu_read_unlock_sched_notrace();                \
 194                }                                                       \
 195        }                                                               \
 196        __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args),          \
 197                PARAMS(cond), PARAMS(data_proto), PARAMS(data_args))    \
 198        static inline int                                               \
 199        register_trace_##name(void (*probe)(data_proto), void *data)    \
 200        {                                                               \
 201                return tracepoint_probe_register(&__tracepoint_##name,  \
 202                                                (void *)probe, data);   \
 203        }                                                               \
 204        static inline int                                               \
 205        register_trace_prio_##name(void (*probe)(data_proto), void *data,\
 206                                   int prio)                            \
 207        {                                                               \
 208                return tracepoint_probe_register_prio(&__tracepoint_##name, \
 209                                              (void *)probe, data, prio); \
 210        }                                                               \
 211        static inline int                                               \
 212        unregister_trace_##name(void (*probe)(data_proto), void *data)  \
 213        {                                                               \
 214                return tracepoint_probe_unregister(&__tracepoint_##name,\
 215                                                (void *)probe, data);   \
 216        }                                                               \
 217        static inline void                                              \
 218        check_trace_callback_type_##name(void (*cb)(data_proto))        \
 219        {                                                               \
 220        }                                                               \
 221        static inline bool                                              \
 222        trace_##name##_enabled(void)                                    \
 223        {                                                               \
 224                return static_key_false(&__tracepoint_##name.key);      \
 225        }
 226
 227/*
 228 * We have no guarantee that gcc and the linker won't up-align the tracepoint
 229 * structures, so we create an array of pointers that will be used for iteration
 230 * on the tracepoints.
 231 */
 232#define DEFINE_TRACE_FN(name, reg, unreg)                                \
 233        static const char __tpstrtab_##name[]                            \
 234        __attribute__((section("__tracepoints_strings"))) = #name;       \
 235        struct tracepoint __tracepoint_##name                            \
 236        __attribute__((section("__tracepoints"))) =                      \
 237                { __tpstrtab_##name, STATIC_KEY_INIT_FALSE, reg, unreg, NULL };\
 238        static struct tracepoint * const __tracepoint_ptr_##name __used  \
 239        __attribute__((section("__tracepoints_ptrs"))) =                 \
 240                &__tracepoint_##name;
 241
 242#define DEFINE_TRACE(name)                                              \
 243        DEFINE_TRACE_FN(name, NULL, NULL);
 244
 245#define EXPORT_TRACEPOINT_SYMBOL_GPL(name)                              \
 246        EXPORT_SYMBOL_GPL(__tracepoint_##name)
 247#define EXPORT_TRACEPOINT_SYMBOL(name)                                  \
 248        EXPORT_SYMBOL(__tracepoint_##name)
 249
 250#else /* !TRACEPOINTS_ENABLED */
 251#define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
 252        static inline void trace_##name(proto)                          \
 253        { }                                                             \
 254        static inline void trace_##name##_rcuidle(proto)                \
 255        { }                                                             \
 256        static inline int                                               \
 257        register_trace_##name(void (*probe)(data_proto),                \
 258                              void *data)                               \
 259        {                                                               \
 260                return -ENOSYS;                                         \
 261        }                                                               \
 262        static inline int                                               \
 263        unregister_trace_##name(void (*probe)(data_proto),              \
 264                                void *data)                             \
 265        {                                                               \
 266                return -ENOSYS;                                         \
 267        }                                                               \
 268        static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \
 269        {                                                               \
 270        }                                                               \
 271        static inline bool                                              \
 272        trace_##name##_enabled(void)                                    \
 273        {                                                               \
 274                return false;                                           \
 275        }
 276
 277#define DEFINE_TRACE_FN(name, reg, unreg)
 278#define DEFINE_TRACE(name)
 279#define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
 280#define EXPORT_TRACEPOINT_SYMBOL(name)
 281
 282#endif /* TRACEPOINTS_ENABLED */
 283
 284#ifdef CONFIG_TRACING
 285/**
 286 * tracepoint_string - register constant persistent string to trace system
 287 * @str - a constant persistent string that will be referenced in tracepoints
 288 *
 289 * If constant strings are being used in tracepoints, it is faster and
 290 * more efficient to just save the pointer to the string and reference
 291 * that with a printf "%s" instead of saving the string in the ring buffer
 292 * and wasting space and time.
 293 *
 294 * The problem with the above approach is that userspace tools that read
 295 * the binary output of the trace buffers do not have access to the string.
 296 * Instead they just show the address of the string which is not very
 297 * useful to users.
 298 *
 299 * With tracepoint_string(), the string will be registered to the tracing
 300 * system and exported to userspace via the debugfs/tracing/printk_formats
 301 * file that maps the string address to the string text. This way userspace
 302 * tools that read the binary buffers have a way to map the pointers to
 303 * the ASCII strings they represent.
 304 *
 305 * The @str used must be a constant string and persistent as it would not
 306 * make sense to show a string that no longer exists. But it is still fine
 307 * to be used with modules, because when modules are unloaded, if they
 308 * had tracepoints, the ring buffers are cleared too. As long as the string
 309 * does not change during the life of the module, it is fine to use
 310 * tracepoint_string() within a module.
 311 */
 312#define tracepoint_string(str)                                          \
 313        ({                                                              \
 314                static const char *___tp_str __tracepoint_string = str; \
 315                ___tp_str;                                              \
 316        })
 317#define __tracepoint_string     __attribute__((section("__tracepoint_str")))
 318#else
 319/*
 320 * tracepoint_string() is used to save the string address for userspace
 321 * tracing tools. When tracing isn't configured, there's no need to save
 322 * anything.
 323 */
 324# define tracepoint_string(str) str
 325# define __tracepoint_string
 326#endif
 327
 328/*
 329 * The need for the DECLARE_TRACE_NOARGS() is to handle the prototype
 330 * (void). "void" is a special value in a function prototype and can
 331 * not be combined with other arguments. Since the DECLARE_TRACE()
 332 * macro adds a data element at the beginning of the prototype,
 333 * we need a way to differentiate "(void *data, proto)" from
 334 * "(void *data, void)". The second prototype is invalid.
 335 *
 336 * DECLARE_TRACE_NOARGS() passes "void" as the tracepoint prototype
 337 * and "void *__data" as the callback prototype.
 338 *
 339 * DECLARE_TRACE() passes "proto" as the tracepoint protoype and
 340 * "void *__data, proto" as the callback prototype.
 341 */
 342#define DECLARE_TRACE_NOARGS(name)                                      \
 343        __DECLARE_TRACE(name, void, ,                                   \
 344                        cpu_online(raw_smp_processor_id()),             \
 345                        void *__data, __data)
 346
 347#define DECLARE_TRACE(name, proto, args)                                \
 348        __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args),              \
 349                        cpu_online(raw_smp_processor_id()),             \
 350                        PARAMS(void *__data, proto),                    \
 351                        PARAMS(__data, args))
 352
 353#define DECLARE_TRACE_CONDITION(name, proto, args, cond)                \
 354        __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args),              \
 355                        cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \
 356                        PARAMS(void *__data, proto),                    \
 357                        PARAMS(__data, args))
 358
 359#define TRACE_EVENT_FLAGS(event, flag)
 360
 361#define TRACE_EVENT_PERF_PERM(event, expr...)
 362
 363#endif /* DECLARE_TRACE */
 364
 365#ifndef TRACE_EVENT
 366/*
 367 * For use with the TRACE_EVENT macro:
 368 *
 369 * We define a tracepoint, its arguments, its printk format
 370 * and its 'fast binary record' layout.
 371 *
 372 * Firstly, name your tracepoint via TRACE_EVENT(name : the
 373 * 'subsystem_event' notation is fine.
 374 *
 375 * Think about this whole construct as the
 376 * 'trace_sched_switch() function' from now on.
 377 *
 378 *
 379 *  TRACE_EVENT(sched_switch,
 380 *
 381 *      *
 382 *      * A function has a regular function arguments
 383 *      * prototype, declare it via TP_PROTO():
 384 *      *
 385 *
 386 *      TP_PROTO(struct rq *rq, struct task_struct *prev,
 387 *               struct task_struct *next),
 388 *
 389 *      *
 390 *      * Define the call signature of the 'function'.
 391 *      * (Design sidenote: we use this instead of a
 392 *      *  TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
 393 *      *
 394 *
 395 *      TP_ARGS(rq, prev, next),
 396 *
 397 *      *
 398 *      * Fast binary tracing: define the trace record via
 399 *      * TP_STRUCT__entry(). You can think about it like a
 400 *      * regular C structure local variable definition.
 401 *      *
 402 *      * This is how the trace record is structured and will
 403 *      * be saved into the ring buffer. These are the fields
 404 *      * that will be exposed to user-space in
 405 *      * /sys/kernel/debug/tracing/events/<*>/format.
 406 *      *
 407 *      * The declared 'local variable' is called '__entry'
 408 *      *
 409 *      * __field(pid_t, prev_prid) is equivalent to a standard declariton:
 410 *      *
 411 *      *       pid_t   prev_pid;
 412 *      *
 413 *      * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
 414 *      *
 415 *      *       char    prev_comm[TASK_COMM_LEN];
 416 *      *
 417 *
 418 *      TP_STRUCT__entry(
 419 *              __array(        char,   prev_comm,      TASK_COMM_LEN   )
 420 *              __field(        pid_t,  prev_pid                        )
 421 *              __field(        int,    prev_prio                       )
 422 *              __array(        char,   next_comm,      TASK_COMM_LEN   )
 423 *              __field(        pid_t,  next_pid                        )
 424 *              __field(        int,    next_prio                       )
 425 *      ),
 426 *
 427 *      *
 428 *      * Assign the entry into the trace record, by embedding
 429 *      * a full C statement block into TP_fast_assign(). You
 430 *      * can refer to the trace record as '__entry' -
 431 *      * otherwise you can put arbitrary C code in here.
 432 *      *
 433 *      * Note: this C code will execute every time a trace event
 434 *      * happens, on an active tracepoint.
 435 *      *
 436 *
 437 *      TP_fast_assign(
 438 *              memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
 439 *              __entry->prev_pid       = prev->pid;
 440 *              __entry->prev_prio      = prev->prio;
 441 *              memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
 442 *              __entry->next_pid       = next->pid;
 443 *              __entry->next_prio      = next->prio;
 444 *      ),
 445 *
 446 *      *
 447 *      * Formatted output of a trace record via TP_printk().
 448 *      * This is how the tracepoint will appear under ftrace
 449 *      * plugins that make use of this tracepoint.
 450 *      *
 451 *      * (raw-binary tracing wont actually perform this step.)
 452 *      *
 453 *
 454 *      TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
 455 *              __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
 456 *              __entry->next_comm, __entry->next_pid, __entry->next_prio),
 457 *
 458 * );
 459 *
 460 * This macro construct is thus used for the regular printk format
 461 * tracing setup, it is used to construct a function pointer based
 462 * tracepoint callback (this is used by programmatic plugins and
 463 * can also by used by generic instrumentation like SystemTap), and
 464 * it is also used to expose a structured trace record in
 465 * /sys/kernel/debug/tracing/events/.
 466 *
 467 * A set of (un)registration functions can be passed to the variant
 468 * TRACE_EVENT_FN to perform any (un)registration work.
 469 */
 470
 471#define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
 472#define DEFINE_EVENT(template, name, proto, args)               \
 473        DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 474#define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\
 475        DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 476#define DEFINE_EVENT_PRINT(template, name, proto, args, print)  \
 477        DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 478#define DEFINE_EVENT_CONDITION(template, name, proto,           \
 479                               args, cond)                      \
 480        DECLARE_TRACE_CONDITION(name, PARAMS(proto),            \
 481                                PARAMS(args), PARAMS(cond))
 482
 483#define TRACE_EVENT(name, proto, args, struct, assign, print)   \
 484        DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 485#define TRACE_EVENT_FN(name, proto, args, struct,               \
 486                assign, print, reg, unreg)                      \
 487        DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 488#define TRACE_EVENT_FN_COND(name, proto, args, cond, struct,            \
 489                assign, print, reg, unreg)                      \
 490        DECLARE_TRACE_CONDITION(name, PARAMS(proto),    \
 491                        PARAMS(args), PARAMS(cond))
 492#define TRACE_EVENT_CONDITION(name, proto, args, cond,          \
 493                              struct, assign, print)            \
 494        DECLARE_TRACE_CONDITION(name, PARAMS(proto),            \
 495                                PARAMS(args), PARAMS(cond))
 496
 497#define TRACE_EVENT_FLAGS(event, flag)
 498
 499#define TRACE_EVENT_PERF_PERM(event, expr...)
 500
 501#endif /* ifdef TRACE_EVENT (see note above) */
 502