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