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.rst.
   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_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
  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 int 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#define TRACE_DEFINE_SIZEOF(x)
  92
  93#endif /* _LINUX_TRACEPOINT_H */
  94
  95/*
  96 * Note: we keep the TRACE_EVENT and DECLARE_TRACE outside the include
  97 *  file ifdef protection.
  98 *  This is due to the way trace events work. If a file includes two
  99 *  trace event headers under one "CREATE_TRACE_POINTS" the first include
 100 *  will override the TRACE_EVENT and break the second include.
 101 */
 102
 103#ifndef DECLARE_TRACE
 104
 105#define TP_PROTO(args...)       args
 106#define TP_ARGS(args...)        args
 107#define TP_CONDITION(args...)   args
 108
 109/*
 110 * Individual subsystem my have a separate configuration to
 111 * enable their tracepoints. By default, this file will create
 112 * the tracepoints if CONFIG_TRACEPOINT is defined. If a subsystem
 113 * wants to be able to disable its tracepoints from being created
 114 * it can define NOTRACE before including the tracepoint headers.
 115 */
 116#if defined(CONFIG_TRACEPOINTS) && !defined(NOTRACE)
 117#define TRACEPOINTS_ENABLED
 118#endif
 119
 120#ifdef TRACEPOINTS_ENABLED
 121
 122/*
 123 * it_func[0] is never NULL because there is at least one element in the array
 124 * when the array itself is non NULL.
 125 *
 126 * Note, the proto and args passed in includes "__data" as the first parameter.
 127 * The reason for this is to handle the "void" prototype. If a tracepoint
 128 * has a "void" prototype, then it is invalid to declare a function
 129 * as "(void *, void)". The DECLARE_TRACE_NOARGS() will pass in just
 130 * "void *data", where as the DECLARE_TRACE() will pass in "void *data, proto".
 131 */
 132#define __DO_TRACE(tp, proto, args, cond, rcucheck)                     \
 133        do {                                                            \
 134                struct tracepoint_func *it_func_ptr;                    \
 135                void *it_func;                                          \
 136                void *__data;                                           \
 137                                                                        \
 138                if (!(cond))                                            \
 139                        return;                                         \
 140                if (rcucheck)                                           \
 141                        rcu_irq_enter_irqson();                         \
 142                rcu_read_lock_sched_notrace();                          \
 143                it_func_ptr = rcu_dereference_sched((tp)->funcs);       \
 144                if (it_func_ptr) {                                      \
 145                        do {                                            \
 146                                it_func = (it_func_ptr)->func;          \
 147                                __data = (it_func_ptr)->data;           \
 148                                ((void(*)(proto))(it_func))(args);      \
 149                        } while ((++it_func_ptr)->func);                \
 150                }                                                       \
 151                rcu_read_unlock_sched_notrace();                        \
 152                if (rcucheck)                                           \
 153                        rcu_irq_exit_irqson();                          \
 154        } while (0)
 155
 156#ifndef MODULE
 157#define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args) \
 158        static inline void trace_##name##_rcuidle(proto)                \
 159        {                                                               \
 160                if (static_key_false(&__tracepoint_##name.key))         \
 161                        __DO_TRACE(&__tracepoint_##name,                \
 162                                TP_PROTO(data_proto),                   \
 163                                TP_ARGS(data_args),                     \
 164                                TP_CONDITION(cond), 1);                 \
 165        }
 166#else
 167#define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args)
 168#endif
 169
 170/*
 171 * Make sure the alignment of the structure in the __tracepoints section will
 172 * not add unwanted padding between the beginning of the section and the
 173 * structure. Force alignment to the same alignment as the section start.
 174 *
 175 * When lockdep is enabled, we make sure to always do the RCU portions of
 176 * the tracepoint code, regardless of whether tracing is on. However,
 177 * don't check if the condition is false, due to interaction with idle
 178 * instrumentation. This lets us find RCU issues triggered with tracepoints
 179 * even when this tracepoint is off. This code has no purpose other than
 180 * poking RCU a bit.
 181 */
 182#define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
 183        extern struct tracepoint __tracepoint_##name;                   \
 184        static inline void trace_##name(proto)                          \
 185        {                                                               \
 186                if (static_key_false(&__tracepoint_##name.key))         \
 187                        __DO_TRACE(&__tracepoint_##name,                \
 188                                TP_PROTO(data_proto),                   \
 189                                TP_ARGS(data_args),                     \
 190                                TP_CONDITION(cond), 0);                 \
 191                if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) {             \
 192                        rcu_read_lock_sched_notrace();                  \
 193                        rcu_dereference_sched(__tracepoint_##name.funcs);\
 194                        rcu_read_unlock_sched_notrace();                \
 195                }                                                       \
 196        }                                                               \
 197        __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args),          \
 198                PARAMS(cond), PARAMS(data_proto), PARAMS(data_args))    \
 199        static inline int                                               \
 200        register_trace_##name(void (*probe)(data_proto), void *data)    \
 201        {                                                               \
 202                return tracepoint_probe_register(&__tracepoint_##name,  \
 203                                                (void *)probe, data);   \
 204        }                                                               \
 205        static inline int                                               \
 206        register_trace_prio_##name(void (*probe)(data_proto), void *data,\
 207                                   int prio)                            \
 208        {                                                               \
 209                return tracepoint_probe_register_prio(&__tracepoint_##name, \
 210                                              (void *)probe, data, prio); \
 211        }                                                               \
 212        static inline int                                               \
 213        unregister_trace_##name(void (*probe)(data_proto), void *data)  \
 214        {                                                               \
 215                return tracepoint_probe_unregister(&__tracepoint_##name,\
 216                                                (void *)probe, data);   \
 217        }                                                               \
 218        static inline void                                              \
 219        check_trace_callback_type_##name(void (*cb)(data_proto))        \
 220        {                                                               \
 221        }                                                               \
 222        static inline bool                                              \
 223        trace_##name##_enabled(void)                                    \
 224        {                                                               \
 225                return static_key_false(&__tracepoint_##name.key);      \
 226        }
 227
 228/*
 229 * We have no guarantee that gcc and the linker won't up-align the tracepoint
 230 * structures, so we create an array of pointers that will be used for iteration
 231 * on the tracepoints.
 232 */
 233#define DEFINE_TRACE_FN(name, reg, unreg)                                \
 234        static const char __tpstrtab_##name[]                            \
 235        __attribute__((section("__tracepoints_strings"))) = #name;       \
 236        struct tracepoint __tracepoint_##name                            \
 237        __attribute__((section("__tracepoints"))) =                      \
 238                { __tpstrtab_##name, STATIC_KEY_INIT_FALSE, reg, unreg, NULL };\
 239        static struct tracepoint * const __tracepoint_ptr_##name __used  \
 240        __attribute__((section("__tracepoints_ptrs"))) =                 \
 241                &__tracepoint_##name;
 242
 243#define DEFINE_TRACE(name)                                              \
 244        DEFINE_TRACE_FN(name, NULL, NULL);
 245
 246#define EXPORT_TRACEPOINT_SYMBOL_GPL(name)                              \
 247        EXPORT_SYMBOL_GPL(__tracepoint_##name)
 248#define EXPORT_TRACEPOINT_SYMBOL(name)                                  \
 249        EXPORT_SYMBOL(__tracepoint_##name)
 250
 251#else /* !TRACEPOINTS_ENABLED */
 252#define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
 253        static inline void trace_##name(proto)                          \
 254        { }                                                             \
 255        static inline void trace_##name##_rcuidle(proto)                \
 256        { }                                                             \
 257        static inline int                                               \
 258        register_trace_##name(void (*probe)(data_proto),                \
 259                              void *data)                               \
 260        {                                                               \
 261                return -ENOSYS;                                         \
 262        }                                                               \
 263        static inline int                                               \
 264        unregister_trace_##name(void (*probe)(data_proto),              \
 265                                void *data)                             \
 266        {                                                               \
 267                return -ENOSYS;                                         \
 268        }                                                               \
 269        static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \
 270        {                                                               \
 271        }                                                               \
 272        static inline bool                                              \
 273        trace_##name##_enabled(void)                                    \
 274        {                                                               \
 275                return false;                                           \
 276        }
 277
 278#define DEFINE_TRACE_FN(name, reg, unreg)
 279#define DEFINE_TRACE(name)
 280#define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
 281#define EXPORT_TRACEPOINT_SYMBOL(name)
 282
 283#endif /* TRACEPOINTS_ENABLED */
 284
 285#ifdef CONFIG_TRACING
 286/**
 287 * tracepoint_string - register constant persistent string to trace system
 288 * @str - a constant persistent string that will be referenced in tracepoints
 289 *
 290 * If constant strings are being used in tracepoints, it is faster and
 291 * more efficient to just save the pointer to the string and reference
 292 * that with a printf "%s" instead of saving the string in the ring buffer
 293 * and wasting space and time.
 294 *
 295 * The problem with the above approach is that userspace tools that read
 296 * the binary output of the trace buffers do not have access to the string.
 297 * Instead they just show the address of the string which is not very
 298 * useful to users.
 299 *
 300 * With tracepoint_string(), the string will be registered to the tracing
 301 * system and exported to userspace via the debugfs/tracing/printk_formats
 302 * file that maps the string address to the string text. This way userspace
 303 * tools that read the binary buffers have a way to map the pointers to
 304 * the ASCII strings they represent.
 305 *
 306 * The @str used must be a constant string and persistent as it would not
 307 * make sense to show a string that no longer exists. But it is still fine
 308 * to be used with modules, because when modules are unloaded, if they
 309 * had tracepoints, the ring buffers are cleared too. As long as the string
 310 * does not change during the life of the module, it is fine to use
 311 * tracepoint_string() within a module.
 312 */
 313#define tracepoint_string(str)                                          \
 314        ({                                                              \
 315                static const char *___tp_str __tracepoint_string = str; \
 316                ___tp_str;                                              \
 317        })
 318#define __tracepoint_string     __attribute__((section("__tracepoint_str")))
 319#else
 320/*
 321 * tracepoint_string() is used to save the string address for userspace
 322 * tracing tools. When tracing isn't configured, there's no need to save
 323 * anything.
 324 */
 325# define tracepoint_string(str) str
 326# define __tracepoint_string
 327#endif
 328
 329/*
 330 * The need for the DECLARE_TRACE_NOARGS() is to handle the prototype
 331 * (void). "void" is a special value in a function prototype and can
 332 * not be combined with other arguments. Since the DECLARE_TRACE()
 333 * macro adds a data element at the beginning of the prototype,
 334 * we need a way to differentiate "(void *data, proto)" from
 335 * "(void *data, void)". The second prototype is invalid.
 336 *
 337 * DECLARE_TRACE_NOARGS() passes "void" as the tracepoint prototype
 338 * and "void *__data" as the callback prototype.
 339 *
 340 * DECLARE_TRACE() passes "proto" as the tracepoint protoype and
 341 * "void *__data, proto" as the callback prototype.
 342 */
 343#define DECLARE_TRACE_NOARGS(name)                                      \
 344        __DECLARE_TRACE(name, void, ,                                   \
 345                        cpu_online(raw_smp_processor_id()),             \
 346                        void *__data, __data)
 347
 348#define DECLARE_TRACE(name, proto, args)                                \
 349        __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args),              \
 350                        cpu_online(raw_smp_processor_id()),             \
 351                        PARAMS(void *__data, proto),                    \
 352                        PARAMS(__data, args))
 353
 354#define DECLARE_TRACE_CONDITION(name, proto, args, cond)                \
 355        __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args),              \
 356                        cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \
 357                        PARAMS(void *__data, proto),                    \
 358                        PARAMS(__data, args))
 359
 360#define TRACE_EVENT_FLAGS(event, flag)
 361
 362#define TRACE_EVENT_PERF_PERM(event, expr...)
 363
 364#endif /* DECLARE_TRACE */
 365
 366#ifndef TRACE_EVENT
 367/*
 368 * For use with the TRACE_EVENT macro:
 369 *
 370 * We define a tracepoint, its arguments, its printk format
 371 * and its 'fast binary record' layout.
 372 *
 373 * Firstly, name your tracepoint via TRACE_EVENT(name : the
 374 * 'subsystem_event' notation is fine.
 375 *
 376 * Think about this whole construct as the
 377 * 'trace_sched_switch() function' from now on.
 378 *
 379 *
 380 *  TRACE_EVENT(sched_switch,
 381 *
 382 *      *
 383 *      * A function has a regular function arguments
 384 *      * prototype, declare it via TP_PROTO():
 385 *      *
 386 *
 387 *      TP_PROTO(struct rq *rq, struct task_struct *prev,
 388 *               struct task_struct *next),
 389 *
 390 *      *
 391 *      * Define the call signature of the 'function'.
 392 *      * (Design sidenote: we use this instead of a
 393 *      *  TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
 394 *      *
 395 *
 396 *      TP_ARGS(rq, prev, next),
 397 *
 398 *      *
 399 *      * Fast binary tracing: define the trace record via
 400 *      * TP_STRUCT__entry(). You can think about it like a
 401 *      * regular C structure local variable definition.
 402 *      *
 403 *      * This is how the trace record is structured and will
 404 *      * be saved into the ring buffer. These are the fields
 405 *      * that will be exposed to user-space in
 406 *      * /sys/kernel/debug/tracing/events/<*>/format.
 407 *      *
 408 *      * The declared 'local variable' is called '__entry'
 409 *      *
 410 *      * __field(pid_t, prev_prid) is equivalent to a standard declariton:
 411 *      *
 412 *      *       pid_t   prev_pid;
 413 *      *
 414 *      * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
 415 *      *
 416 *      *       char    prev_comm[TASK_COMM_LEN];
 417 *      *
 418 *
 419 *      TP_STRUCT__entry(
 420 *              __array(        char,   prev_comm,      TASK_COMM_LEN   )
 421 *              __field(        pid_t,  prev_pid                        )
 422 *              __field(        int,    prev_prio                       )
 423 *              __array(        char,   next_comm,      TASK_COMM_LEN   )
 424 *              __field(        pid_t,  next_pid                        )
 425 *              __field(        int,    next_prio                       )
 426 *      ),
 427 *
 428 *      *
 429 *      * Assign the entry into the trace record, by embedding
 430 *      * a full C statement block into TP_fast_assign(). You
 431 *      * can refer to the trace record as '__entry' -
 432 *      * otherwise you can put arbitrary C code in here.
 433 *      *
 434 *      * Note: this C code will execute every time a trace event
 435 *      * happens, on an active tracepoint.
 436 *      *
 437 *
 438 *      TP_fast_assign(
 439 *              memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
 440 *              __entry->prev_pid       = prev->pid;
 441 *              __entry->prev_prio      = prev->prio;
 442 *              memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
 443 *              __entry->next_pid       = next->pid;
 444 *              __entry->next_prio      = next->prio;
 445 *      ),
 446 *
 447 *      *
 448 *      * Formatted output of a trace record via TP_printk().
 449 *      * This is how the tracepoint will appear under ftrace
 450 *      * plugins that make use of this tracepoint.
 451 *      *
 452 *      * (raw-binary tracing wont actually perform this step.)
 453 *      *
 454 *
 455 *      TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
 456 *              __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
 457 *              __entry->next_comm, __entry->next_pid, __entry->next_prio),
 458 *
 459 * );
 460 *
 461 * This macro construct is thus used for the regular printk format
 462 * tracing setup, it is used to construct a function pointer based
 463 * tracepoint callback (this is used by programmatic plugins and
 464 * can also by used by generic instrumentation like SystemTap), and
 465 * it is also used to expose a structured trace record in
 466 * /sys/kernel/debug/tracing/events/.
 467 *
 468 * A set of (un)registration functions can be passed to the variant
 469 * TRACE_EVENT_FN to perform any (un)registration work.
 470 */
 471
 472#define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
 473#define DEFINE_EVENT(template, name, proto, args)               \
 474        DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 475#define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\
 476        DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 477#define DEFINE_EVENT_PRINT(template, name, proto, args, print)  \
 478        DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 479#define DEFINE_EVENT_CONDITION(template, name, proto,           \
 480                               args, cond)                      \
 481        DECLARE_TRACE_CONDITION(name, PARAMS(proto),            \
 482                                PARAMS(args), PARAMS(cond))
 483
 484#define TRACE_EVENT(name, proto, args, struct, assign, print)   \
 485        DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 486#define TRACE_EVENT_FN(name, proto, args, struct,               \
 487                assign, print, reg, unreg)                      \
 488        DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 489#define TRACE_EVENT_FN_COND(name, proto, args, cond, struct,            \
 490                assign, print, reg, unreg)                      \
 491        DECLARE_TRACE_CONDITION(name, PARAMS(proto),    \
 492                        PARAMS(args), PARAMS(cond))
 493#define TRACE_EVENT_CONDITION(name, proto, args, cond,          \
 494                              struct, assign, print)            \
 495        DECLARE_TRACE_CONDITION(name, PARAMS(proto),            \
 496                                PARAMS(args), PARAMS(cond))
 497
 498#define TRACE_EVENT_FLAGS(event, flag)
 499
 500#define TRACE_EVENT_PERF_PERM(event, expr...)
 501
 502#endif /* ifdef TRACE_EVENT (see note above) */
 503