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