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/errno.h>
  18 #include <linux/types.h>
  19 #include <linux/rcupdate.h>
  20 #include <linux/static_key.h>
  21
  22 struct module;
  23 struct tracepoint;
  24 struct notifier_block;
  25
  26 struct tracepoint_func {
  27         void *func;
  28         void *data;
  29         int prio;
  30 };
  31
  32 struct tracepoint {
  33         const char *name;               /* Tracepoint name */
  34         struct static_key key;
  35         void (*regfunc)(void);
  36         void (*unregfunc)(void);
  37         struct tracepoint_func __rcu *funcs;
  38 };
  39
  40 struct trace_enum_map {
  41         const char              *system;
  42         const char              *enum_string;
  43         unsigned long           enum_value;
  44 };
  45
  46 #define TRACEPOINT_DEFAULT_PRIO 10
  47
  48 extern int
  49 tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data);
  50 extern int
  51 tracepoint_probe_register_prio(struct tracepoint *tp, void *probe, void *data,
  52                                int prio);
  53 extern int
  54 tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data);
  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 static inline void tracepoint_synchronize_unregister(void)
  91 {
  92         synchronize_sched();
  93 }
  94
  95 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
  96 extern void syscall_regfunc(void);
  97 extern void syscall_unregfunc(void);
  98 #endif /* CONFIG_HAVE_SYSCALL_TRACEPOINTS */
  99
 100 #define PARAMS(args...) args
 101
 102 #define TRACE_DEFINE_ENUM(x)
 103
 104 #endif /* _LINUX_TRACEPOINT_H */
 105
 106 /*
 107  * Note: we keep the TRACE_EVENT and DECLARE_TRACE outside the include
 108  *  file ifdef protection.
 109  *  This is due to the way trace events work. If a file includes two
 110  *  trace event headers under one "CREATE_TRACE_POINTS" the first include
 111  *  will override the TRACE_EVENT and break the second include.
 112  */
 113
 114 #ifndef DECLARE_TRACE
 115
 116 #define TP_PROTO(args...)       args
 117 #define TP_ARGS(args...)        args
 118 #define TP_CONDITION(args...)   args
 119
 120 /*
 121  * Individual subsystem my have a separate configuration to
 122  * enable their tracepoints. By default, this file will create
 123  * the tracepoints if CONFIG_TRACEPOINT is defined. If a subsystem
 124  * wants to be able to disable its tracepoints from being created
 125  * it can define NOTRACE before including the tracepoint headers.
 126  */
 127 #if defined(CONFIG_TRACEPOINTS) && !defined(NOTRACE)
 128 #define TRACEPOINTS_ENABLED
 129 #endif
 130
 131 #ifdef TRACEPOINTS_ENABLED
 132
 133 /*
 134  * it_func[0] is never NULL because there is at least one element in the array
 135  * when the array itself is non NULL.
 136  *
 137  * Note, the proto and args passed in includes "__data" as the first parameter.
 138  * The reason for this is to handle the "void" prototype. If a tracepoint
 139  * has a "void" prototype, then it is invalid to declare a function
 140  * as "(void *, void)". The DECLARE_TRACE_NOARGS() will pass in just
 141  * "void *data", where as the DECLARE_TRACE() will pass in "void *data, proto".
 142  */
 143 #define __DO_TRACE(tp, proto, args, cond, prercu, postrcu)              \
 144         do {                                                            \
 145                 struct tracepoint_func *it_func_ptr;                    \
 146                 void *it_func;                                          \
 147                 void *__data;                                           \
 148                                                                         \
 149                 if (!(cond))                                            \
 150                         return;                                         \
 151                 prercu;                                                 \
 152                 rcu_read_lock_sched_notrace();                          \
 153                 it_func_ptr = rcu_dereference_sched((tp)->funcs);       \
 154                 if (it_func_ptr) {                                      \
 155                         do {                                            \
 156                                 it_func = (it_func_ptr)->func;          \
 157                                 __data = (it_func_ptr)->data;           \
 158                                 ((void(*)(proto))(it_func))(args);      \
 159                         } while ((++it_func_ptr)->func);                \
 160                 }                                                       \
 161                 rcu_read_unlock_sched_notrace();                        \
 162                 postrcu;                                                \
 163         } while (0)
 164
 165 #ifndef MODULE
 166 #define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args)     \
 167         static inline void trace_##name##_rcuidle(proto)                \
 168         {                                                               \
 169                 if (static_key_false(&__tracepoint_##name.key))         \
 170                         __DO_TRACE(&__tracepoint_##name,                \
 171                                 TP_PROTO(data_proto),                   \
 172                                 TP_ARGS(data_args),                     \
 173                                 TP_CONDITION(cond),                     \
 174                                 rcu_irq_enter(),                        \
 175                                 rcu_irq_exit());                        \
 176         }
 177 #else
 178 #define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args)
 179 #endif
 180
 181 /*
 182  * Make sure the alignment of the structure in the __tracepoints section will
 183  * not add unwanted padding between the beginning of the section and the
 184  * structure. Force alignment to the same alignment as the section start.
 185  *
 186  * When lockdep is enabled, we make sure to always do the RCU portions of
 187  * the tracepoint code, regardless of whether tracing is on. However,
 188  * don't check if the condition is false, due to interaction with idle
 189  * instrumentation. This lets us find RCU issues triggered with tracepoints
 190  * even when this tracepoint is off. This code has no purpose other than
 191  * poking RCU a bit.
 192  */
 193 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
 194         extern struct tracepoint __tracepoint_##name;                   \
 195         static inline void trace_##name(proto)                          \
 196         {                                                               \
 197                 if (static_key_false(&__tracepoint_##name.key))         \
 198                         __DO_TRACE(&__tracepoint_##name,                \
 199                                 TP_PROTO(data_proto),                   \
 200                                 TP_ARGS(data_args),                     \
 201                                 TP_CONDITION(cond),,);                  \
 202                 if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) {             \
 203                         rcu_read_lock_sched_notrace();                  \
 204                         rcu_dereference_sched(__tracepoint_##name.funcs);\
 205                         rcu_read_unlock_sched_notrace();                \
 206                 }                                                       \
 207         }                                                               \
 208         __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args),          \
 209                 PARAMS(cond), PARAMS(data_proto), PARAMS(data_args))    \
 210         static inline int                                               \
 211         register_trace_##name(void (*probe)(data_proto), void *data)    \
 212         {                                                               \
 213                 return tracepoint_probe_register(&__tracepoint_##name,  \
 214                                                 (void *)probe, data);   \
 215         }                                                               \
 216         static inline int                                               \
 217         register_trace_prio_##name(void (*probe)(data_proto), void *data,\
 218                                    int prio)                            \
 219         {                                                               \
 220                 return tracepoint_probe_register_prio(&__tracepoint_##name, \
 221                                               (void *)probe, data, prio); \
 222         }                                                               \
 223         static inline int                                               \
 224         unregister_trace_##name(void (*probe)(data_proto), void *data)  \
 225         {                                                               \
 226                 return tracepoint_probe_unregister(&__tracepoint_##name,\
 227                                                 (void *)probe, data);   \
 228         }                                                               \
 229         static inline void                                              \
 230         check_trace_callback_type_##name(void (*cb)(data_proto))        \
 231         {                                                               \
 232         }                                                               \
 233         static inline bool                                              \
 234         trace_##name##_enabled(void)                                    \
 235         {                                                               \
 236                 return static_key_false(&__tracepoint_##name.key);      \
 237         }
 238
 239 /*
 240  * We have no guarantee that gcc and the linker won't up-align the tracepoint
 241  * structures, so we create an array of pointers that will be used for iteration
 242  * on the tracepoints.
 243  */
 244 #define DEFINE_TRACE_FN(name, reg, unreg)                                \
 245         static const char __tpstrtab_##name[]                            \
 246         __attribute__((section("__tracepoints_strings"))) = #name;       \
 247         struct tracepoint __tracepoint_##name                            \
 248         __attribute__((section("__tracepoints"))) =                      \
 249                 { __tpstrtab_##name, STATIC_KEY_INIT_FALSE, reg, unreg, NULL };\
 250         static struct tracepoint * const __tracepoint_ptr_##name __used  \
 251         __attribute__((section("__tracepoints_ptrs"))) =                 \
 252                 &__tracepoint_##name;
 253
 254 #define DEFINE_TRACE(name)                                              \
 255         DEFINE_TRACE_FN(name, NULL, NULL);
 256
 257 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)                              \
 258         EXPORT_SYMBOL_GPL(__tracepoint_##name)
 259 #define EXPORT_TRACEPOINT_SYMBOL(name)                                  \
 260         EXPORT_SYMBOL(__tracepoint_##name)
 261
 262 #else /* !TRACEPOINTS_ENABLED */
 263 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
 264         static inline void trace_##name(proto)                          \
 265         { }                                                             \
 266         static inline void trace_##name##_rcuidle(proto)                \
 267         { }                                                             \
 268         static inline int                                               \
 269         register_trace_##name(void (*probe)(data_proto),                \
 270                               void *data)                               \
 271         {                                                               \
 272                 return -ENOSYS;                                         \
 273         }                                                               \
 274         static inline int                                               \
 275         unregister_trace_##name(void (*probe)(data_proto),              \
 276                                 void *data)                             \
 277         {                                                               \
 278                 return -ENOSYS;                                         \
 279         }                                                               \
 280         static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \
 281         {                                                               \
 282         }                                                               \
 283         static inline bool                                              \
 284         trace_##name##_enabled(void)                                    \
 285         {                                                               \
 286                 return false;                                           \
 287         }
 288
 289 #define DEFINE_TRACE_FN(name, reg, unreg)
 290 #define DEFINE_TRACE(name)
 291 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
 292 #define EXPORT_TRACEPOINT_SYMBOL(name)
 293
 294 #endif /* TRACEPOINTS_ENABLED */
 295
 296 #ifdef CONFIG_TRACING
 297 /**
 298  * tracepoint_string - register constant persistent string to trace system
 299  * @str - a constant persistent string that will be referenced in tracepoints
 300  *
 301  * If constant strings are being used in tracepoints, it is faster and
 302  * more efficient to just save the pointer to the string and reference
 303  * that with a printf "%s" instead of saving the string in the ring buffer
 304  * and wasting space and time.
 305  *
 306  * The problem with the above approach is that userspace tools that read
 307  * the binary output of the trace buffers do not have access to the string.
 308  * Instead they just show the address of the string which is not very
 309  * useful to users.
 310  *
 311  * With tracepoint_string(), the string will be registered to the tracing
 312  * system and exported to userspace via the debugfs/tracing/printk_formats
 313  * file that maps the string address to the string text. This way userspace
 314  * tools that read the binary buffers have a way to map the pointers to
 315  * the ASCII strings they represent.
 316  *
 317  * The @str used must be a constant string and persistent as it would not
 318  * make sense to show a string that no longer exists. But it is still fine
 319  * to be used with modules, because when modules are unloaded, if they
 320  * had tracepoints, the ring buffers are cleared too. As long as the string
 321  * does not change during the life of the module, it is fine to use
 322  * tracepoint_string() within a module.
 323  */
 324 #define tracepoint_string(str)                                          \
 325         ({                                                              \
 326                 static const char *___tp_str __tracepoint_string = str; \
 327                 ___tp_str;                                              \
 328         })
 329 #define __tracepoint_string     __attribute__((section("__tracepoint_str")))
 330 #else
 331 /*
 332  * tracepoint_string() is used to save the string address for userspace
 333  * tracing tools. When tracing isn't configured, there's no need to save
 334  * anything.
 335  */
 336 # define tracepoint_string(str) str
 337 # define __tracepoint_string
 338 #endif
 339
 340 /*
 341  * The need for the DECLARE_TRACE_NOARGS() is to handle the prototype
 342  * (void). "void" is a special value in a function prototype and can
 343  * not be combined with other arguments. Since the DECLARE_TRACE()
 344  * macro adds a data element at the beginning of the prototype,
 345  * we need a way to differentiate "(void *data, proto)" from
 346  * "(void *data, void)". The second prototype is invalid.
 347  *
 348  * DECLARE_TRACE_NOARGS() passes "void" as the tracepoint prototype
 349  * and "void *__data" as the callback prototype.
 350  *
 351  * DECLARE_TRACE() passes "proto" as the tracepoint protoype and
 352  * "void *__data, proto" as the callback prototype.
 353  */
 354 #define DECLARE_TRACE_NOARGS(name)                                      \
 355                 __DECLARE_TRACE(name, void, , 1, void *__data, __data)
 356
 357 #define DECLARE_TRACE(name, proto, args)                                \
 358                 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), 1,   \
 359                                 PARAMS(void *__data, proto),            \
 360                                 PARAMS(__data, args))
 361
 362 #define DECLARE_TRACE_CONDITION(name, proto, args, cond)                \
 363         __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), PARAMS(cond), \
 364                         PARAMS(void *__data, proto),                    \
 365                         PARAMS(__data, args))
 366
 367 #define TRACE_EVENT_FLAGS(event, flag)
 368
 369 #define TRACE_EVENT_PERF_PERM(event, expr...)
 370
 371 #endif /* DECLARE_TRACE */
 372
 373 #ifndef TRACE_EVENT
 374 /*
 375  * For use with the TRACE_EVENT macro:
 376  *
 377  * We define a tracepoint, its arguments, its printk format
 378  * and its 'fast binary record' layout.
 379  *
 380  * Firstly, name your tracepoint via TRACE_EVENT(name : the
 381  * 'subsystem_event' notation is fine.
 382  *
 383  * Think about this whole construct as the
 384  * 'trace_sched_switch() function' from now on.
 385  *
 386  *
 387  *  TRACE_EVENT(sched_switch,
 388  *
 389  *      *
 390  *      * A function has a regular function arguments
 391  *      * prototype, declare it via TP_PROTO():
 392  *      *
 393  *
 394  *      TP_PROTO(struct rq *rq, struct task_struct *prev,
 395  *               struct task_struct *next),
 396  *
 397  *      *
 398  *      * Define the call signature of the 'function'.
 399  *      * (Design sidenote: we use this instead of a
 400  *      *  TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
 401  *      *
 402  *
 403  *      TP_ARGS(rq, prev, next),
 404  *
 405  *      *
 406  *      * Fast binary tracing: define the trace record via
 407  *      * TP_STRUCT__entry(). You can think about it like a
 408  *      * regular C structure local variable definition.
 409  *      *
 410  *      * This is how the trace record is structured and will
 411  *      * be saved into the ring buffer. These are the fields
 412  *      * that will be exposed to user-space in
 413  *      * /sys/kernel/debug/tracing/events/<*>/format.
 414  *      *
 415  *      * The declared 'local variable' is called '__entry'
 416  *      *
 417  *      * __field(pid_t, prev_prid) is equivalent to a standard declariton:
 418  *      *
 419  *      *       pid_t   prev_pid;
 420  *      *
 421  *      * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
 422  *      *
 423  *      *       char    prev_comm[TASK_COMM_LEN];
 424  *      *
 425  *
 426  *      TP_STRUCT__entry(
 427  *              __array(        char,   prev_comm,      TASK_COMM_LEN   )
 428  *              __field(        pid_t,  prev_pid                        )
 429  *              __field(        int,    prev_prio                       )
 430  *              __array(        char,   next_comm,      TASK_COMM_LEN   )
 431  *              __field(        pid_t,  next_pid                        )
 432  *              __field(        int,    next_prio                       )
 433  *      ),
 434  *
 435  *      *
 436  *      * Assign the entry into the trace record, by embedding
 437  *      * a full C statement block into TP_fast_assign(). You
 438  *      * can refer to the trace record as '__entry' -
 439  *      * otherwise you can put arbitrary C code in here.
 440  *      *
 441  *      * Note: this C code will execute every time a trace event
 442  *      * happens, on an active tracepoint.
 443  *      *
 444  *
 445  *      TP_fast_assign(
 446  *              memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
 447  *              __entry->prev_pid       = prev->pid;
 448  *              __entry->prev_prio      = prev->prio;
 449  *              memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
 450  *              __entry->next_pid       = next->pid;
 451  *              __entry->next_prio      = next->prio;
 452  *      ),
 453  *
 454  *      *
 455  *      * Formatted output of a trace record via TP_printk().
 456  *      * This is how the tracepoint will appear under ftrace
 457  *      * plugins that make use of this tracepoint.
 458  *      *
 459  *      * (raw-binary tracing wont actually perform this step.)
 460  *      *
 461  *
 462  *      TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
 463  *              __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
 464  *              __entry->next_comm, __entry->next_pid, __entry->next_prio),
 465  *
 466  * );
 467  *
 468  * This macro construct is thus used for the regular printk format
 469  * tracing setup, it is used to construct a function pointer based
 470  * tracepoint callback (this is used by programmatic plugins and
 471  * can also by used by generic instrumentation like SystemTap), and
 472  * it is also used to expose a structured trace record in
 473  * /sys/kernel/debug/tracing/events/.
 474  *
 475  * A set of (un)registration functions can be passed to the variant
 476  * TRACE_EVENT_FN to perform any (un)registration work.
 477  */
 478
 479 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
 480 #define DEFINE_EVENT(template, name, proto, args)               \
 481         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 482 #define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\
 483         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 484 #define DEFINE_EVENT_PRINT(template, name, proto, args, print)  \
 485         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 486 #define DEFINE_EVENT_CONDITION(template, name, proto,           \
 487                                args, cond)                      \
 488         DECLARE_TRACE_CONDITION(name, PARAMS(proto),            \
 489                                 PARAMS(args), PARAMS(cond))
 490
 491 #define TRACE_EVENT(name, proto, args, struct, assign, print)   \
 492         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 493 #define TRACE_EVENT_FN(name, proto, args, struct,               \
 494                 assign, print, reg, unreg)                      \
 495         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 496 #define TRACE_EVENT_CONDITION(name, proto, args, cond,          \
 497                               struct, assign, print)            \
 498         DECLARE_TRACE_CONDITION(name, PARAMS(proto),            \
 499                                 PARAMS(args), PARAMS(cond))
 500
 501 #define TRACE_EVENT_FLAGS(event, flag)
 502
 503 #define TRACE_EVENT_PERF_PERM(event, expr...)
 504
 505 #endif /* ifdef TRACE_EVENT (see note above) */