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/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/jump_label.h>
  23
  24 struct module;
  25 struct tracepoint;
  26
  27 struct tracepoint_func {
  28         void *func;
  29         void *data;
  30 };
  31
  32 struct tracepoint {
  33         const char *name;               /* Tracepoint name */
  34         struct jump_label_key key;
  35         void (*regfunc)(void);
  36         void (*unregfunc)(void);
  37         struct tracepoint_func __rcu *funcs;
  38 };
  39
  40 /*
  41  * Connect a probe to a tracepoint.
  42  * Internal API, should not be used directly.
  43  */
  44 extern int tracepoint_probe_register(const char *name, void *probe, void *data);
  45
  46 /*
  47  * Disconnect a probe from a tracepoint.
  48  * Internal API, should not be used directly.
  49  */
  50 extern int
  51 tracepoint_probe_unregister(const char *name, void *probe, void *data);
  52
  53 extern int tracepoint_probe_register_noupdate(const char *name, void *probe,
  54                                               void *data);
  55 extern int tracepoint_probe_unregister_noupdate(const char *name, void *probe,
  56                                                 void *data);
  57 extern void tracepoint_probe_update_all(void);
  58
  59 #ifdef CONFIG_MODULES
  60 struct tp_module {
  61         struct list_head list;
  62         unsigned int num_tracepoints;
  63         struct tracepoint * const *tracepoints_ptrs;
  64 };
  65 bool trace_module_has_bad_taint(struct module *mod);
  66 #else
  67 static inline bool trace_module_has_bad_taint(struct module *mod)
  68 {
  69         return false;
  70 }
  71 #endif /* CONFIG_MODULES */
  72
  73 struct tracepoint_iter {
  74 #ifdef CONFIG_MODULES
  75         struct tp_module *module;
  76 #endif /* CONFIG_MODULES */
  77         struct tracepoint * const *tracepoint;
  78 };
  79
  80 extern void tracepoint_iter_start(struct tracepoint_iter *iter);
  81 extern void tracepoint_iter_next(struct tracepoint_iter *iter);
  82 extern void tracepoint_iter_stop(struct tracepoint_iter *iter);
  83 extern void tracepoint_iter_reset(struct tracepoint_iter *iter);
  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 #define PARAMS(args...) args
  96
  97 #endif /* _LINUX_TRACEPOINT_H */
  98
  99 /*
 100  * Note: we keep the TRACE_EVENT and DECLARE_TRACE outside the include
 101  *  file ifdef protection.
 102  *  This is due to the way trace events work. If a file includes two
 103  *  trace event headers under one "CREATE_TRACE_POINTS" the first include
 104  *  will override the TRACE_EVENT and break the second include.
 105  */
 106
 107 #ifndef DECLARE_TRACE
 108
 109 #define TP_PROTO(args...)       args
 110 #define TP_ARGS(args...)        args
 111 #define TP_CONDITION(args...)   args
 112
 113 #ifdef CONFIG_TRACEPOINTS
 114
 115 /*
 116  * it_func[0] is never NULL because there is at least one element in the array
 117  * when the array itself is non NULL.
 118  *
 119  * Note, the proto and args passed in includes "__data" as the first parameter.
 120  * The reason for this is to handle the "void" prototype. If a tracepoint
 121  * has a "void" prototype, then it is invalid to declare a function
 122  * as "(void *, void)". The DECLARE_TRACE_NOARGS() will pass in just
 123  * "void *data", where as the DECLARE_TRACE() will pass in "void *data, proto".
 124  */
 125 #define __DO_TRACE(tp, proto, args, cond)                               \
 126         do {                                                            \
 127                 struct tracepoint_func *it_func_ptr;                    \
 128                 void *it_func;                                          \
 129                 void *__data;                                           \
 130                                                                         \
 131                 if (!cpu_online(raw_smp_processor_id()))                \
 132                         return;                                         \
 133                                                                         \
 134                 if (!(cond))                                            \
 135                         return;                                         \
 136                 rcu_read_lock_sched_notrace();                          \
 137                 it_func_ptr = rcu_dereference_sched((tp)->funcs);       \
 138                 if (it_func_ptr) {                                      \
 139                         do {                                            \
 140                                 it_func = (it_func_ptr)->func;          \
 141                                 __data = (it_func_ptr)->data;           \
 142                                 ((void(*)(proto))(it_func))(args);      \
 143                         } while ((++it_func_ptr)->func);                \
 144                 }                                                       \
 145                 rcu_read_unlock_sched_notrace();                        \
 146         } while (0)
 147
 148 /*
 149  * Make sure the alignment of the structure in the __tracepoints section will
 150  * not add unwanted padding between the beginning of the section and the
 151  * structure. Force alignment to the same alignment as the section start.
 152  */
 153 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
 154         extern struct tracepoint __tracepoint_##name;                   \
 155         static inline void trace_##name(proto)                          \
 156         {                                                               \
 157                 if (static_branch(&__tracepoint_##name.key))            \
 158                         __DO_TRACE(&__tracepoint_##name,                \
 159                                 TP_PROTO(data_proto),                   \
 160                                 TP_ARGS(data_args),                     \
 161                                 TP_CONDITION(cond));                    \
 162         }                                                               \
 163         static inline int                                               \
 164         register_trace_##name(void (*probe)(data_proto), void *data)    \
 165         {                                                               \
 166                 return tracepoint_probe_register(#name, (void *)probe,  \
 167                                                  data);                 \
 168         }                                                               \
 169         static inline int                                               \
 170         unregister_trace_##name(void (*probe)(data_proto), void *data)  \
 171         {                                                               \
 172                 return tracepoint_probe_unregister(#name, (void *)probe, \
 173                                                    data);               \
 174         }                                                               \
 175         static inline void                                              \
 176         check_trace_callback_type_##name(void (*cb)(data_proto))        \
 177         {                                                               \
 178         }
 179
 180 /*
 181  * We have no guarantee that gcc and the linker won't up-align the tracepoint
 182  * structures, so we create an array of pointers that will be used for iteration
 183  * on the tracepoints.
 184  */
 185 #define DEFINE_TRACE_FN(name, reg, unreg)                                \
 186         static const char __tpstrtab_##name[]                            \
 187         __attribute__((section("__tracepoints_strings"))) = #name;       \
 188         struct tracepoint __tracepoint_##name                            \
 189         __attribute__((section("__tracepoints"))) =                      \
 190                 { __tpstrtab_##name, JUMP_LABEL_INIT, reg, unreg, NULL };\
 191         static struct tracepoint * const __tracepoint_ptr_##name __used  \
 192         __attribute__((section("__tracepoints_ptrs"))) =                 \
 193                 &__tracepoint_##name;
 194
 195 #define DEFINE_TRACE(name)                                              \
 196         DEFINE_TRACE_FN(name, NULL, NULL);
 197
 198 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)                              \
 199         EXPORT_SYMBOL_GPL(__tracepoint_##name)
 200 #define EXPORT_TRACEPOINT_SYMBOL(name)                                  \
 201         EXPORT_SYMBOL(__tracepoint_##name)
 202
 203 #else /* !CONFIG_TRACEPOINTS */
 204 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \
 205         static inline void trace_##name(proto)                          \
 206         { }                                                             \
 207         static inline int                                               \
 208         register_trace_##name(void (*probe)(data_proto),                \
 209                               void *data)                               \
 210         {                                                               \
 211                 return -ENOSYS;                                         \
 212         }                                                               \
 213         static inline int                                               \
 214         unregister_trace_##name(void (*probe)(data_proto),              \
 215                                 void *data)                             \
 216         {                                                               \
 217                 return -ENOSYS;                                         \
 218         }                                                               \
 219         static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \
 220         {                                                               \
 221         }
 222
 223 #define DEFINE_TRACE_FN(name, reg, unreg)
 224 #define DEFINE_TRACE(name)
 225 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
 226 #define EXPORT_TRACEPOINT_SYMBOL(name)
 227
 228 #endif /* CONFIG_TRACEPOINTS */
 229
 230 /*
 231  * The need for the DECLARE_TRACE_NOARGS() is to handle the prototype
 232  * (void). "void" is a special value in a function prototype and can
 233  * not be combined with other arguments. Since the DECLARE_TRACE()
 234  * macro adds a data element at the beginning of the prototype,
 235  * we need a way to differentiate "(void *data, proto)" from
 236  * "(void *data, void)". The second prototype is invalid.
 237  *
 238  * DECLARE_TRACE_NOARGS() passes "void" as the tracepoint prototype
 239  * and "void *__data" as the callback prototype.
 240  *
 241  * DECLARE_TRACE() passes "proto" as the tracepoint protoype and
 242  * "void *__data, proto" as the callback prototype.
 243  */
 244 #define DECLARE_TRACE_NOARGS(name)                                      \
 245                 __DECLARE_TRACE(name, void, , 1, void *__data, __data)
 246
 247 #define DECLARE_TRACE(name, proto, args)                                \
 248                 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), 1,   \
 249                                 PARAMS(void *__data, proto),            \
 250                                 PARAMS(__data, args))
 251
 252 #define DECLARE_TRACE_CONDITION(name, proto, args, cond)                \
 253         __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), PARAMS(cond), \
 254                         PARAMS(void *__data, proto),                    \
 255                         PARAMS(__data, args))
 256
 257 #define TRACE_EVENT_FLAGS(event, flag)
 258
 259 #endif /* DECLARE_TRACE */
 260
 261 #ifndef TRACE_EVENT
 262 /*
 263  * For use with the TRACE_EVENT macro:
 264  *
 265  * We define a tracepoint, its arguments, its printk format
 266  * and its 'fast binay record' layout.
 267  *
 268  * Firstly, name your tracepoint via TRACE_EVENT(name : the
 269  * 'subsystem_event' notation is fine.
 270  *
 271  * Think about this whole construct as the
 272  * 'trace_sched_switch() function' from now on.
 273  *
 274  *
 275  *  TRACE_EVENT(sched_switch,
 276  *
 277  *      *
 278  *      * A function has a regular function arguments
 279  *      * prototype, declare it via TP_PROTO():
 280  *      *
 281  *
 282  *      TP_PROTO(struct rq *rq, struct task_struct *prev,
 283  *               struct task_struct *next),
 284  *
 285  *      *
 286  *      * Define the call signature of the 'function'.
 287  *      * (Design sidenote: we use this instead of a
 288  *      *  TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
 289  *      *
 290  *
 291  *      TP_ARGS(rq, prev, next),
 292  *
 293  *      *
 294  *      * Fast binary tracing: define the trace record via
 295  *      * TP_STRUCT__entry(). You can think about it like a
 296  *      * regular C structure local variable definition.
 297  *      *
 298  *      * This is how the trace record is structured and will
 299  *      * be saved into the ring buffer. These are the fields
 300  *      * that will be exposed to user-space in
 301  *      * /sys/kernel/debug/tracing/events/<*>/format.
 302  *      *
 303  *      * The declared 'local variable' is called '__entry'
 304  *      *
 305  *      * __field(pid_t, prev_prid) is equivalent to a standard declariton:
 306  *      *
 307  *      *       pid_t   prev_pid;
 308  *      *
 309  *      * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
 310  *      *
 311  *      *       char    prev_comm[TASK_COMM_LEN];
 312  *      *
 313  *
 314  *      TP_STRUCT__entry(
 315  *              __array(        char,   prev_comm,      TASK_COMM_LEN   )
 316  *              __field(        pid_t,  prev_pid                        )
 317  *              __field(        int,    prev_prio                       )
 318  *              __array(        char,   next_comm,      TASK_COMM_LEN   )
 319  *              __field(        pid_t,  next_pid                        )
 320  *              __field(        int,    next_prio                       )
 321  *      ),
 322  *
 323  *      *
 324  *      * Assign the entry into the trace record, by embedding
 325  *      * a full C statement block into TP_fast_assign(). You
 326  *      * can refer to the trace record as '__entry' -
 327  *      * otherwise you can put arbitrary C code in here.
 328  *      *
 329  *      * Note: this C code will execute every time a trace event
 330  *      * happens, on an active tracepoint.
 331  *      *
 332  *
 333  *      TP_fast_assign(
 334  *              memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
 335  *              __entry->prev_pid       = prev->pid;
 336  *              __entry->prev_prio      = prev->prio;
 337  *              memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
 338  *              __entry->next_pid       = next->pid;
 339  *              __entry->next_prio      = next->prio;
 340  *      ),
 341  *
 342  *      *
 343  *      * Formatted output of a trace record via TP_printk().
 344  *      * This is how the tracepoint will appear under ftrace
 345  *      * plugins that make use of this tracepoint.
 346  *      *
 347  *      * (raw-binary tracing wont actually perform this step.)
 348  *      *
 349  *
 350  *      TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
 351  *              __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
 352  *              __entry->next_comm, __entry->next_pid, __entry->next_prio),
 353  *
 354  * );
 355  *
 356  * This macro construct is thus used for the regular printk format
 357  * tracing setup, it is used to construct a function pointer based
 358  * tracepoint callback (this is used by programmatic plugins and
 359  * can also by used by generic instrumentation like SystemTap), and
 360  * it is also used to expose a structured trace record in
 361  * /sys/kernel/debug/tracing/events/.
 362  *
 363  * A set of (un)registration functions can be passed to the variant
 364  * TRACE_EVENT_FN to perform any (un)registration work.
 365  */
 366
 367 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print)
 368 #define DEFINE_EVENT(template, name, proto, args)               \
 369         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 370 #define DEFINE_EVENT_PRINT(template, name, proto, args, print)  \
 371         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 372 #define DEFINE_EVENT_CONDITION(template, name, proto,           \
 373                                args, cond)                      \
 374         DECLARE_TRACE_CONDITION(name, PARAMS(proto),            \
 375                                 PARAMS(args), PARAMS(cond))
 376
 377 #define TRACE_EVENT(name, proto, args, struct, assign, print)   \
 378         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 379 #define TRACE_EVENT_FN(name, proto, args, struct,               \
 380                 assign, print, reg, unreg)                      \
 381         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
 382 #define TRACE_EVENT_CONDITION(name, proto, args, cond,          \
 383                               struct, assign, print)            \
 384         DECLARE_TRACE_CONDITION(name, PARAMS(proto),            \
 385                                 PARAMS(args), PARAMS(cond))
 386
 387 #define TRACE_EVENT_FLAGS(event, flag)
 388
 389 #endif /* ifdef TRACE_EVENT (see note above) */