at v4.0 15 kB view raw
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 22struct module; 23struct tracepoint; 24struct notifier_block; 25 26struct tracepoint_func { 27 void *func; 28 void *data; 29}; 30 31struct tracepoint { 32 const char *name; /* Tracepoint name */ 33 struct static_key key; 34 void (*regfunc)(void); 35 void (*unregfunc)(void); 36 struct tracepoint_func __rcu *funcs; 37}; 38 39extern int 40tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data); 41extern int 42tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data); 43extern void 44for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv), 45 void *priv); 46 47#ifdef CONFIG_MODULES 48struct tp_module { 49 struct list_head list; 50 struct module *mod; 51}; 52 53bool trace_module_has_bad_taint(struct module *mod); 54extern int register_tracepoint_module_notifier(struct notifier_block *nb); 55extern int unregister_tracepoint_module_notifier(struct notifier_block *nb); 56#else 57static inline bool trace_module_has_bad_taint(struct module *mod) 58{ 59 return false; 60} 61static inline 62int register_tracepoint_module_notifier(struct notifier_block *nb) 63{ 64 return 0; 65} 66static inline 67int unregister_tracepoint_module_notifier(struct notifier_block *nb) 68{ 69 return 0; 70} 71#endif /* CONFIG_MODULES */ 72 73/* 74 * tracepoint_synchronize_unregister must be called between the last tracepoint 75 * probe unregistration and the end of module exit to make sure there is no 76 * caller executing a probe when it is freed. 77 */ 78static inline void tracepoint_synchronize_unregister(void) 79{ 80 synchronize_sched(); 81} 82 83#ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS 84extern void syscall_regfunc(void); 85extern void syscall_unregfunc(void); 86#endif /* CONFIG_HAVE_SYSCALL_TRACEPOINTS */ 87 88#define PARAMS(args...) args 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#ifdef CONFIG_TRACEPOINTS 107 108/* 109 * it_func[0] is never NULL because there is at least one element in the array 110 * when the array itself is non NULL. 111 * 112 * Note, the proto and args passed in includes "__data" as the first parameter. 113 * The reason for this is to handle the "void" prototype. If a tracepoint 114 * has a "void" prototype, then it is invalid to declare a function 115 * as "(void *, void)". The DECLARE_TRACE_NOARGS() will pass in just 116 * "void *data", where as the DECLARE_TRACE() will pass in "void *data, proto". 117 */ 118#define __DO_TRACE(tp, proto, args, cond, prercu, postrcu) \ 119 do { \ 120 struct tracepoint_func *it_func_ptr; \ 121 void *it_func; \ 122 void *__data; \ 123 \ 124 if (!(cond)) \ 125 return; \ 126 prercu; \ 127 rcu_read_lock_sched_notrace(); \ 128 it_func_ptr = rcu_dereference_sched((tp)->funcs); \ 129 if (it_func_ptr) { \ 130 do { \ 131 it_func = (it_func_ptr)->func; \ 132 __data = (it_func_ptr)->data; \ 133 ((void(*)(proto))(it_func))(args); \ 134 } while ((++it_func_ptr)->func); \ 135 } \ 136 rcu_read_unlock_sched_notrace(); \ 137 postrcu; \ 138 } while (0) 139 140#ifndef MODULE 141#define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args) \ 142 static inline void trace_##name##_rcuidle(proto) \ 143 { \ 144 if (static_key_false(&__tracepoint_##name.key)) \ 145 __DO_TRACE(&__tracepoint_##name, \ 146 TP_PROTO(data_proto), \ 147 TP_ARGS(data_args), \ 148 TP_CONDITION(cond), \ 149 rcu_irq_enter(), \ 150 rcu_irq_exit()); \ 151 } 152#else 153#define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args) 154#endif 155 156/* 157 * Make sure the alignment of the structure in the __tracepoints section will 158 * not add unwanted padding between the beginning of the section and the 159 * structure. Force alignment to the same alignment as the section start. 160 * 161 * When lockdep is enabled, we make sure to always do the RCU portions of 162 * the tracepoint code, regardless of whether tracing is on or we match the 163 * condition. This lets us find RCU issues triggered with tracepoints even 164 * when this tracepoint is off. This code has no purpose other than poking 165 * RCU a bit. 166 */ 167#define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \ 168 extern struct tracepoint __tracepoint_##name; \ 169 static inline void trace_##name(proto) \ 170 { \ 171 if (static_key_false(&__tracepoint_##name.key)) \ 172 __DO_TRACE(&__tracepoint_##name, \ 173 TP_PROTO(data_proto), \ 174 TP_ARGS(data_args), \ 175 TP_CONDITION(cond),,); \ 176 if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) { \ 177 rcu_read_lock_sched_notrace(); \ 178 rcu_dereference_sched(__tracepoint_##name.funcs);\ 179 rcu_read_unlock_sched_notrace(); \ 180 } \ 181 } \ 182 __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args), \ 183 PARAMS(cond), PARAMS(data_proto), PARAMS(data_args)) \ 184 static inline int \ 185 register_trace_##name(void (*probe)(data_proto), void *data) \ 186 { \ 187 return tracepoint_probe_register(&__tracepoint_##name, \ 188 (void *)probe, data); \ 189 } \ 190 static inline int \ 191 unregister_trace_##name(void (*probe)(data_proto), void *data) \ 192 { \ 193 return tracepoint_probe_unregister(&__tracepoint_##name,\ 194 (void *)probe, data); \ 195 } \ 196 static inline void \ 197 check_trace_callback_type_##name(void (*cb)(data_proto)) \ 198 { \ 199 } \ 200 static inline bool \ 201 trace_##name##_enabled(void) \ 202 { \ 203 return static_key_false(&__tracepoint_##name.key); \ 204 } 205 206/* 207 * We have no guarantee that gcc and the linker won't up-align the tracepoint 208 * structures, so we create an array of pointers that will be used for iteration 209 * on the tracepoints. 210 */ 211#define DEFINE_TRACE_FN(name, reg, unreg) \ 212 static const char __tpstrtab_##name[] \ 213 __attribute__((section("__tracepoints_strings"))) = #name; \ 214 struct tracepoint __tracepoint_##name \ 215 __attribute__((section("__tracepoints"))) = \ 216 { __tpstrtab_##name, STATIC_KEY_INIT_FALSE, reg, unreg, NULL };\ 217 static struct tracepoint * const __tracepoint_ptr_##name __used \ 218 __attribute__((section("__tracepoints_ptrs"))) = \ 219 &__tracepoint_##name; 220 221#define DEFINE_TRACE(name) \ 222 DEFINE_TRACE_FN(name, NULL, NULL); 223 224#define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \ 225 EXPORT_SYMBOL_GPL(__tracepoint_##name) 226#define EXPORT_TRACEPOINT_SYMBOL(name) \ 227 EXPORT_SYMBOL(__tracepoint_##name) 228 229#else /* !CONFIG_TRACEPOINTS */ 230#define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \ 231 static inline void trace_##name(proto) \ 232 { } \ 233 static inline void trace_##name##_rcuidle(proto) \ 234 { } \ 235 static inline int \ 236 register_trace_##name(void (*probe)(data_proto), \ 237 void *data) \ 238 { \ 239 return -ENOSYS; \ 240 } \ 241 static inline int \ 242 unregister_trace_##name(void (*probe)(data_proto), \ 243 void *data) \ 244 { \ 245 return -ENOSYS; \ 246 } \ 247 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \ 248 { \ 249 } \ 250 static inline bool \ 251 trace_##name##_enabled(void) \ 252 { \ 253 return false; \ 254 } 255 256#define DEFINE_TRACE_FN(name, reg, unreg) 257#define DEFINE_TRACE(name) 258#define EXPORT_TRACEPOINT_SYMBOL_GPL(name) 259#define EXPORT_TRACEPOINT_SYMBOL(name) 260 261#endif /* CONFIG_TRACEPOINTS */ 262 263#ifdef CONFIG_TRACING 264/** 265 * tracepoint_string - register constant persistent string to trace system 266 * @str - a constant persistent string that will be referenced in tracepoints 267 * 268 * If constant strings are being used in tracepoints, it is faster and 269 * more efficient to just save the pointer to the string and reference 270 * that with a printf "%s" instead of saving the string in the ring buffer 271 * and wasting space and time. 272 * 273 * The problem with the above approach is that userspace tools that read 274 * the binary output of the trace buffers do not have access to the string. 275 * Instead they just show the address of the string which is not very 276 * useful to users. 277 * 278 * With tracepoint_string(), the string will be registered to the tracing 279 * system and exported to userspace via the debugfs/tracing/printk_formats 280 * file that maps the string address to the string text. This way userspace 281 * tools that read the binary buffers have a way to map the pointers to 282 * the ASCII strings they represent. 283 * 284 * The @str used must be a constant string and persistent as it would not 285 * make sense to show a string that no longer exists. But it is still fine 286 * to be used with modules, because when modules are unloaded, if they 287 * had tracepoints, the ring buffers are cleared too. As long as the string 288 * does not change during the life of the module, it is fine to use 289 * tracepoint_string() within a module. 290 */ 291#define tracepoint_string(str) \ 292 ({ \ 293 static const char *___tp_str __tracepoint_string = str; \ 294 ___tp_str; \ 295 }) 296#define __tracepoint_string __attribute__((section("__tracepoint_str"))) 297#else 298/* 299 * tracepoint_string() is used to save the string address for userspace 300 * tracing tools. When tracing isn't configured, there's no need to save 301 * anything. 302 */ 303# define tracepoint_string(str) str 304# define __tracepoint_string 305#endif 306 307/* 308 * The need for the DECLARE_TRACE_NOARGS() is to handle the prototype 309 * (void). "void" is a special value in a function prototype and can 310 * not be combined with other arguments. Since the DECLARE_TRACE() 311 * macro adds a data element at the beginning of the prototype, 312 * we need a way to differentiate "(void *data, proto)" from 313 * "(void *data, void)". The second prototype is invalid. 314 * 315 * DECLARE_TRACE_NOARGS() passes "void" as the tracepoint prototype 316 * and "void *__data" as the callback prototype. 317 * 318 * DECLARE_TRACE() passes "proto" as the tracepoint protoype and 319 * "void *__data, proto" as the callback prototype. 320 */ 321#define DECLARE_TRACE_NOARGS(name) \ 322 __DECLARE_TRACE(name, void, , 1, void *__data, __data) 323 324#define DECLARE_TRACE(name, proto, args) \ 325 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), 1, \ 326 PARAMS(void *__data, proto), \ 327 PARAMS(__data, args)) 328 329#define DECLARE_TRACE_CONDITION(name, proto, args, cond) \ 330 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), PARAMS(cond), \ 331 PARAMS(void *__data, proto), \ 332 PARAMS(__data, args)) 333 334#define TRACE_EVENT_FLAGS(event, flag) 335 336#define TRACE_EVENT_PERF_PERM(event, expr...) 337 338#endif /* DECLARE_TRACE */ 339 340#ifndef TRACE_EVENT 341/* 342 * For use with the TRACE_EVENT macro: 343 * 344 * We define a tracepoint, its arguments, its printk format 345 * and its 'fast binary record' layout. 346 * 347 * Firstly, name your tracepoint via TRACE_EVENT(name : the 348 * 'subsystem_event' notation is fine. 349 * 350 * Think about this whole construct as the 351 * 'trace_sched_switch() function' from now on. 352 * 353 * 354 * TRACE_EVENT(sched_switch, 355 * 356 * * 357 * * A function has a regular function arguments 358 * * prototype, declare it via TP_PROTO(): 359 * * 360 * 361 * TP_PROTO(struct rq *rq, struct task_struct *prev, 362 * struct task_struct *next), 363 * 364 * * 365 * * Define the call signature of the 'function'. 366 * * (Design sidenote: we use this instead of a 367 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.) 368 * * 369 * 370 * TP_ARGS(rq, prev, next), 371 * 372 * * 373 * * Fast binary tracing: define the trace record via 374 * * TP_STRUCT__entry(). You can think about it like a 375 * * regular C structure local variable definition. 376 * * 377 * * This is how the trace record is structured and will 378 * * be saved into the ring buffer. These are the fields 379 * * that will be exposed to user-space in 380 * * /sys/kernel/debug/tracing/events/<*>/format. 381 * * 382 * * The declared 'local variable' is called '__entry' 383 * * 384 * * __field(pid_t, prev_prid) is equivalent to a standard declariton: 385 * * 386 * * pid_t prev_pid; 387 * * 388 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to: 389 * * 390 * * char prev_comm[TASK_COMM_LEN]; 391 * * 392 * 393 * TP_STRUCT__entry( 394 * __array( char, prev_comm, TASK_COMM_LEN ) 395 * __field( pid_t, prev_pid ) 396 * __field( int, prev_prio ) 397 * __array( char, next_comm, TASK_COMM_LEN ) 398 * __field( pid_t, next_pid ) 399 * __field( int, next_prio ) 400 * ), 401 * 402 * * 403 * * Assign the entry into the trace record, by embedding 404 * * a full C statement block into TP_fast_assign(). You 405 * * can refer to the trace record as '__entry' - 406 * * otherwise you can put arbitrary C code in here. 407 * * 408 * * Note: this C code will execute every time a trace event 409 * * happens, on an active tracepoint. 410 * * 411 * 412 * TP_fast_assign( 413 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN); 414 * __entry->prev_pid = prev->pid; 415 * __entry->prev_prio = prev->prio; 416 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN); 417 * __entry->next_pid = next->pid; 418 * __entry->next_prio = next->prio; 419 * ), 420 * 421 * * 422 * * Formatted output of a trace record via TP_printk(). 423 * * This is how the tracepoint will appear under ftrace 424 * * plugins that make use of this tracepoint. 425 * * 426 * * (raw-binary tracing wont actually perform this step.) 427 * * 428 * 429 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]", 430 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio, 431 * __entry->next_comm, __entry->next_pid, __entry->next_prio), 432 * 433 * ); 434 * 435 * This macro construct is thus used for the regular printk format 436 * tracing setup, it is used to construct a function pointer based 437 * tracepoint callback (this is used by programmatic plugins and 438 * can also by used by generic instrumentation like SystemTap), and 439 * it is also used to expose a structured trace record in 440 * /sys/kernel/debug/tracing/events/. 441 * 442 * A set of (un)registration functions can be passed to the variant 443 * TRACE_EVENT_FN to perform any (un)registration work. 444 */ 445 446#define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print) 447#define DEFINE_EVENT(template, name, proto, args) \ 448 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 449#define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\ 450 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 451#define DEFINE_EVENT_PRINT(template, name, proto, args, print) \ 452 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 453#define DEFINE_EVENT_CONDITION(template, name, proto, \ 454 args, cond) \ 455 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 456 PARAMS(args), PARAMS(cond)) 457 458#define TRACE_EVENT(name, proto, args, struct, assign, print) \ 459 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 460#define TRACE_EVENT_FN(name, proto, args, struct, \ 461 assign, print, reg, unreg) \ 462 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 463#define TRACE_EVENT_CONDITION(name, proto, args, cond, \ 464 struct, assign, print) \ 465 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 466 PARAMS(args), PARAMS(cond)) 467 468#define TRACE_EVENT_FLAGS(event, flag) 469 470#define TRACE_EVENT_PERF_PERM(event, expr...) 471 472#endif /* ifdef TRACE_EVENT (see note above) */