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  * (C) Copyright 2008 Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
  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/jump_label.h>
  21
  22 struct module;
  23 struct tracepoint;
  24
  25 struct tracepoint_func {
  26         void *func;
  27         void *data;
  28 };
  29
  30 struct tracepoint {
  31         const char *name;               /* Tracepoint name */
  32         int state;                      /* State. */
  33         void (*regfunc)(void);
  34         void (*unregfunc)(void);
  35         struct tracepoint_func __rcu *funcs;
  36 };
  37
  38 /*
  39  * Connect a probe to a tracepoint.
  40  * Internal API, should not be used directly.
  41  */
  42 extern int tracepoint_probe_register(const char *name, void *probe, void *data);
  43
  44 /*
  45  * Disconnect a probe from a tracepoint.
  46  * Internal API, should not be used directly.
  47  */
  48 extern int
  49 tracepoint_probe_unregister(const char *name, void *probe, void *data);
  50
  51 extern int tracepoint_probe_register_noupdate(const char *name, void *probe,
  52                                               void *data);
  53 extern int tracepoint_probe_unregister_noupdate(const char *name, void *probe,
  54                                                 void *data);
  55 extern void tracepoint_probe_update_all(void);
  56
  57 struct tracepoint_iter {
  58         struct module *module;
  59         struct tracepoint * const *tracepoint;
  60 };
  61
  62 extern void tracepoint_iter_start(struct tracepoint_iter *iter);
  63 extern void tracepoint_iter_next(struct tracepoint_iter *iter);
  64 extern void tracepoint_iter_stop(struct tracepoint_iter *iter);
  65 extern void tracepoint_iter_reset(struct tracepoint_iter *iter);
  66 extern int tracepoint_get_iter_range(struct tracepoint * const **tracepoint,
  67         struct tracepoint * const *begin, struct tracepoint * const *end);
  68
  69 /*
  70  * tracepoint_synchronize_unregister must be called between the last tracepoint
  71  * probe unregistration and the end of module exit to make sure there is no
  72  * caller executing a probe when it is freed.
  73  */
  74 static inline void tracepoint_synchronize_unregister(void)
  75 {
  76         synchronize_sched();
  77 }
  78
  79 #define PARAMS(args...) args
  80
  81 #ifdef CONFIG_TRACEPOINTS
  82 extern
  83 void tracepoint_update_probe_range(struct tracepoint * const *begin,
  84         struct tracepoint * const *end);
  85 #else
  86 static inline
  87 void tracepoint_update_probe_range(struct tracepoint * const *begin,
  88         struct tracepoint * const *end)
  89 { }
  90 #endif /* CONFIG_TRACEPOINTS */
  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 #ifdef CONFIG_TRACEPOINTS
 109
 110 /*
 111  * it_func[0] is never NULL because there is at least one element in the array
 112  * when the array itself is non NULL.
 113  *
 114  * Note, the proto and args passed in includes "__data" as the first parameter.
 115  * The reason for this is to handle the "void" prototype. If a tracepoint
 116  * has a "void" prototype, then it is invalid to declare a function
 117  * as "(void *, void)". The DECLARE_TRACE_NOARGS() will pass in just
 118  * "void *data", where as the DECLARE_TRACE() will pass in "void *data, proto".
 119  */
 120 #define __DO_TRACE(tp, proto, args, cond)                               \
 121         do {                                                            \
 122                 struct tracepoint_func *it_func_ptr;                    \
 123                 void *it_func;                                          \
 124                 void *__data;                                           \
 125                                                                         \
 126                 if (!(cond))                                            \
 127                         return;                                         \
 128                 rcu_read_lock_sched_notrace();                          \
 129                 it_func_ptr = rcu_dereference_sched((tp)->funcs);       \
 130                 if (it_func_ptr) {                                      \
 131                         do {                                            \
 132                                 it_func = (it_func_ptr)->func;          \
 133                                 __data = (it_func_ptr)->data;           \
 134                                 ((void(*)(proto))(it_func))(args);      \
 135                         } while ((++it_func_ptr)->func);                \
 136                 }                                                       \
 137                 rcu_read_unlock_sched_notrace();                        \
 138         } while (0)
 139
 140 /*
 141  * Make sure the alignment of the structure in the __tracepoints section will
 142  * not add unwanted padding between the beginning of the section and the
 143  * structure. Force alignment to the same alignment as the section start.
 144  */
 145 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
 146         extern struct tracepoint __tracepoint_##name;                   \
 147         static inline void trace_##name(proto)                          \
 148         {                                                               \
 149                 JUMP_LABEL(&__tracepoint_##name.state, do_trace);       \
 150                 return;                                                 \
 151 do_trace:                                                               \
 152                         __DO_TRACE(&__tracepoint_##name,                \
 153                                 TP_PROTO(data_proto),                   \
 154                                 TP_ARGS(data_args),                     \
 155                                 TP_CONDITION(cond));                    \
 156         }                                                               \
 157         static inline int                                               \
 158         register_trace_##name(void (*probe)(data_proto), void *data)    \
 159         {                                                               \
 160                 return tracepoint_probe_register(#name, (void *)probe,  \
 161                                                  data);                 \
 162         }                                                               \
 163         static inline int                                               \
 164         unregister_trace_##name(void (*probe)(data_proto), void *data)  \
 165         {                                                               \
 166                 return tracepoint_probe_unregister(#name, (void *)probe, \
 167                                                    data);               \
 168         }                                                               \
 169         static inline void                                              \
 170         check_trace_callback_type_##name(void (*cb)(data_proto))        \
 171         {                                                               \
 172         }
 173
 174 /*
 175  * We have no guarantee that gcc and the linker won't up-align the tracepoint
 176  * structures, so we create an array of pointers that will be used for iteration
 177  * on the tracepoints.
 178  */
 179 #define DEFINE_TRACE_FN(name, reg, unreg)                               \
 180         static const char __tpstrtab_##name[]                           \
 181         __attribute__((section("__tracepoints_strings"))) = #name;      \
 182         struct tracepoint __tracepoint_##name                           \
 183         __attribute__((section("__tracepoints"))) =                     \
 184                 { __tpstrtab_##name, 0, reg, unreg, NULL };             \
 185         static struct tracepoint * const __tracepoint_ptr_##name __used \
 186         __attribute__((section("__tracepoints_ptrs"))) =                \
 187                 &__tracepoint_##name;
 188
 189 #define DEFINE_TRACE(name)                                              \
 190         DEFINE_TRACE_FN(name, NULL, NULL);
 191
 192 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)                              \
 193         EXPORT_SYMBOL_GPL(__tracepoint_##name)
 194 #define EXPORT_TRACEPOINT_SYMBOL(name)                                  \
 195         EXPORT_SYMBOL(__tracepoint_##name)
 196
 197 #else /* !CONFIG_TRACEPOINTS */
 198 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
 199         static inline void trace_##name(proto)                          \
 200         { }                                                             \
 201         static inline int                                               \
 202         register_trace_##name(void (*probe)(data_proto),                \
 203                               void *data)                               \
 204         {                                                               \
 205                 return -ENOSYS;                                         \
 206         }                                                               \
 207         static inline int                                               \
 208         unregister_trace_##name(void (*probe)(data_proto),              \
 209                                 void *data)                             \
 210         {                                                               \
 211                 return -ENOSYS;                                         \
 212         }                                                               \
 213         static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \
 214         {                                                               \
 215         }
 216
 217 #define DEFINE_TRACE_FN(name, reg, unreg)
 218 #define DEFINE_TRACE(name)
 219 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
 220 #define EXPORT_TRACEPOINT_SYMBOL(name)
 221
 222 #endif /* CONFIG_TRACEPOINTS */
 223
 224 /*
 225  * The need for the DECLARE_TRACE_NOARGS() is to handle the prototype
 226  * (void). "void" is a special value in a function prototype and can
 227  * not be combined with other arguments. Since the DECLARE_TRACE()
 228  * macro adds a data element at the beginning of the prototype,
 229  * we need a way to differentiate "(void *data, proto)" from
 230  * "(void *data, void)". The second prototype is invalid.
 231  *
 232  * DECLARE_TRACE_NOARGS() passes "void" as the tracepoint prototype
 233  * and "void *__data" as the callback prototype.
 234  *
 235  * DECLARE_TRACE() passes "proto" as the tracepoint protoype and
 236  * "void *__data, proto" as the callback prototype.
 237  */
 238 #define DECLARE_TRACE_NOARGS(name)                                      \
 239                 __DECLARE_TRACE(name, void, , 1, void *__data, __data)
 240
 241 #define DECLARE_TRACE(name, proto, args)                                \
 242                 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), 1,   \
 243                                 PARAMS(void *__data, proto),            \
 244                                 PARAMS(__data, args))
 245
 246 #define DECLARE_TRACE_CONDITION(name, proto, args, cond)                \
 247         __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), PARAMS(cond), \
 248                         PARAMS(void *__data, proto),                    \
 249                         PARAMS(__data, args))
 250
 251 #define TRACE_EVENT_FLAGS(event, flag)
 252
 253 #endif /* DECLARE_TRACE */
 254
 255 #ifndef TRACE_EVENT
 256 /*
 257  * For use with the TRACE_EVENT macro:
 258  *
 259  * We define a tracepoint, its arguments, its printk format
 260  * and its 'fast binay record' layout.
 261  *
 262  * Firstly, name your tracepoint via TRACE_EVENT(name : the
 263  * 'subsystem_event' notation is fine.
 264  *
 265  * Think about this whole construct as the
 266  * 'trace_sched_switch() function' from now on.
 267  *
 268  *
 269  *  TRACE_EVENT(sched_switch,
 270  *
 271  *      *
 272  *      * A function has a regular function arguments
 273  *      * prototype, declare it via TP_PROTO():
 274  *      *
 275  *
 276  *      TP_PROTO(struct rq *rq, struct task_struct *prev,
 277  *               struct task_struct *next),
 278  *
 279  *      *
 280  *      * Define the call signature of the 'function'.
 281  *      * (Design sidenote: we use this instead of a
 282  *      *  TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
 283  *      *
 284  *
 285  *      TP_ARGS(rq, prev, next),
 286  *
 287  *      *
 288  *      * Fast binary tracing: define the trace record via
 289  *      * TP_STRUCT__entry(). You can think about it like a
 290  *      * regular C structure local variable definition.
 291  *      *
 292  *      * This is how the trace record is structured and will
 293  *      * be saved into the ring buffer. These are the fields
 294  *      * that will be exposed to user-space in
 295  *      * /sys/kernel/debug/tracing/events/<*>/format.
 296  *      *
 297  *      * The declared 'local variable' is called '__entry'
 298  *      *
 299  *      * __field(pid_t, prev_prid) is equivalent to a standard declariton:
 300  *      *
 301  *      *       pid_t   prev_pid;
 302  *      *
 303  *      * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
 304  *      *
 305  *      *       char    prev_comm[TASK_COMM_LEN];
 306  *      *
 307  *
 308  *      TP_STRUCT__entry(
 309  *              __array(        char,   prev_comm,      TASK_COMM_LEN   )
 310  *              __field(        pid_t,  prev_pid                        )
 311  *              __field(        int,    prev_prio                       )
 312  *              __array(        char,   next_comm,      TASK_COMM_LEN   )
 313  *              __field(        pid_t,  next_pid                        )
 314  *              __field(        int,    next_prio                       )
 315  *      ),
 316  *
 317  *      *
 318  *      * Assign the entry into the trace record, by embedding
 319  *      * a full C statement block into TP_fast_assign(). You
 320  *      * can refer to the trace record as '__entry' -
 321  *      * otherwise you can put arbitrary C code in here.
 322  *      *
 323  *      * Note: this C code will execute every time a trace event
 324  *      * happens, on an active tracepoint.
 325  *      *
 326  *
 327  *      TP_fast_assign(
 328  *              memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
 329  *              __entry->prev_pid       = prev->pid;
 330  *              __entry->prev_prio      = prev->prio;
 331  *              memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
 332  *              __entry->next_pid       = next->pid;
 333  *              __entry->next_prio      = next->prio;
 334  *      ),
 335  *
 336  *      *
 337  *      * Formatted output of a trace record via TP_printk().
 338  *      * This is how the tracepoint will appear under ftrace
 339  *      * plugins that make use of this tracepoint.
 340  *      *
 341  *      * (raw-binary tracing wont actually perform this step.)
 342  *      *
 343  *
 344  *      TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
 345  *              __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
 346  *              __entry->next_comm, __entry->next_pid, __entry->next_prio),
 347  *
 348  * );
 349  *
 350  * This macro construct is thus used for the regular printk format
 351  * tracing setup, it is used to construct a function pointer based
 352  * tracepoint callback (this is used by programmatic plugins and
 353  * can also by used by generic instrumentation like SystemTap), and
 354  * it is also used to expose a structured trace record in
 355  * /sys/kernel/debug/tracing/events/.
 356  *
 357  * A set of (un)registration functions can be passed to the variant
 358  * TRACE_EVENT_FN to perform any (un)registration work.
 359  */
 360
 361 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
 362 #define DEFINE_EVENT(template, name, proto, args)               \
 363         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 364 #define DEFINE_EVENT_PRINT(template, name, proto, args, print)  \
 365         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 366 #define DEFINE_EVENT_CONDITION(template, name, proto,           \
 367                                args, cond)                      \
 368         DECLARE_TRACE_CONDITION(name, PARAMS(proto),            \
 369                                 PARAMS(args), PARAMS(cond))
 370
 371 #define TRACE_EVENT(name, proto, args, struct, assign, print)   \
 372         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 373 #define TRACE_EVENT_FN(name, proto, args, struct,               \
 374                 assign, print, reg, unreg)                      \
 375         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 376 #define TRACE_EVENT_CONDITION(name, proto, args, cond,          \
 377                               struct, assign, print)            \
 378         DECLARE_TRACE_CONDITION(name, PARAMS(proto),            \
 379                                 PARAMS(args), PARAMS(cond))
 380
 381 #define TRACE_EVENT_FLAGS(event, flag)
 382
 383 #endif /* ifdef TRACE_EVENT (see note above) */