at v4.13 17 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/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/tracepoint-defs.h> 23 24struct module; 25struct tracepoint; 26struct notifier_block; 27 28struct trace_eval_map { 29 const char *system; 30 const char *eval_string; 31 unsigned long eval_value; 32}; 33 34#define TRACEPOINT_DEFAULT_PRIO 10 35 36extern int 37tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data); 38extern int 39tracepoint_probe_register_prio(struct tracepoint *tp, void *probe, void *data, 40 int prio); 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 int syscall_regfunc(void); 85extern void syscall_unregfunc(void); 86#endif /* CONFIG_HAVE_SYSCALL_TRACEPOINTS */ 87 88#define PARAMS(args...) args 89 90#define TRACE_DEFINE_ENUM(x) 91#define TRACE_DEFINE_SIZEOF(x) 92 93#endif /* _LINUX_TRACEPOINT_H */ 94 95/* 96 * Note: we keep the TRACE_EVENT and DECLARE_TRACE outside the include 97 * file ifdef protection. 98 * This is due to the way trace events work. If a file includes two 99 * trace event headers under one "CREATE_TRACE_POINTS" the first include 100 * will override the TRACE_EVENT and break the second include. 101 */ 102 103#ifndef DECLARE_TRACE 104 105#define TP_PROTO(args...) args 106#define TP_ARGS(args...) args 107#define TP_CONDITION(args...) args 108 109/* 110 * Individual subsystem my have a separate configuration to 111 * enable their tracepoints. By default, this file will create 112 * the tracepoints if CONFIG_TRACEPOINT is defined. If a subsystem 113 * wants to be able to disable its tracepoints from being created 114 * it can define NOTRACE before including the tracepoint headers. 115 */ 116#if defined(CONFIG_TRACEPOINTS) && !defined(NOTRACE) 117#define TRACEPOINTS_ENABLED 118#endif 119 120#ifdef TRACEPOINTS_ENABLED 121 122/* 123 * it_func[0] is never NULL because there is at least one element in the array 124 * when the array itself is non NULL. 125 * 126 * Note, the proto and args passed in includes "__data" as the first parameter. 127 * The reason for this is to handle the "void" prototype. If a tracepoint 128 * has a "void" prototype, then it is invalid to declare a function 129 * as "(void *, void)". The DECLARE_TRACE_NOARGS() will pass in just 130 * "void *data", where as the DECLARE_TRACE() will pass in "void *data, proto". 131 */ 132#define __DO_TRACE(tp, proto, args, cond, rcucheck) \ 133 do { \ 134 struct tracepoint_func *it_func_ptr; \ 135 void *it_func; \ 136 void *__data; \ 137 \ 138 if (!(cond)) \ 139 return; \ 140 if (rcucheck) { \ 141 if (WARN_ON_ONCE(rcu_irq_enter_disabled())) \ 142 return; \ 143 rcu_irq_enter_irqson(); \ 144 } \ 145 rcu_read_lock_sched_notrace(); \ 146 it_func_ptr = rcu_dereference_sched((tp)->funcs); \ 147 if (it_func_ptr) { \ 148 do { \ 149 it_func = (it_func_ptr)->func; \ 150 __data = (it_func_ptr)->data; \ 151 ((void(*)(proto))(it_func))(args); \ 152 } while ((++it_func_ptr)->func); \ 153 } \ 154 rcu_read_unlock_sched_notrace(); \ 155 if (rcucheck) \ 156 rcu_irq_exit_irqson(); \ 157 } while (0) 158 159#ifndef MODULE 160#define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args) \ 161 static inline void trace_##name##_rcuidle(proto) \ 162 { \ 163 if (static_key_false(&__tracepoint_##name.key)) \ 164 __DO_TRACE(&__tracepoint_##name, \ 165 TP_PROTO(data_proto), \ 166 TP_ARGS(data_args), \ 167 TP_CONDITION(cond), 1); \ 168 } 169#else 170#define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args) 171#endif 172 173/* 174 * Make sure the alignment of the structure in the __tracepoints section will 175 * not add unwanted padding between the beginning of the section and the 176 * structure. Force alignment to the same alignment as the section start. 177 * 178 * When lockdep is enabled, we make sure to always do the RCU portions of 179 * the tracepoint code, regardless of whether tracing is on. However, 180 * don't check if the condition is false, due to interaction with idle 181 * instrumentation. This lets us find RCU issues triggered with tracepoints 182 * even when this tracepoint is off. This code has no purpose other than 183 * poking RCU a bit. 184 */ 185#define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \ 186 extern struct tracepoint __tracepoint_##name; \ 187 static inline void trace_##name(proto) \ 188 { \ 189 if (static_key_false(&__tracepoint_##name.key)) \ 190 __DO_TRACE(&__tracepoint_##name, \ 191 TP_PROTO(data_proto), \ 192 TP_ARGS(data_args), \ 193 TP_CONDITION(cond), 0); \ 194 if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) { \ 195 rcu_read_lock_sched_notrace(); \ 196 rcu_dereference_sched(__tracepoint_##name.funcs);\ 197 rcu_read_unlock_sched_notrace(); \ 198 } \ 199 } \ 200 __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args), \ 201 PARAMS(cond), PARAMS(data_proto), PARAMS(data_args)) \ 202 static inline int \ 203 register_trace_##name(void (*probe)(data_proto), void *data) \ 204 { \ 205 return tracepoint_probe_register(&__tracepoint_##name, \ 206 (void *)probe, data); \ 207 } \ 208 static inline int \ 209 register_trace_prio_##name(void (*probe)(data_proto), void *data,\ 210 int prio) \ 211 { \ 212 return tracepoint_probe_register_prio(&__tracepoint_##name, \ 213 (void *)probe, data, prio); \ 214 } \ 215 static inline int \ 216 unregister_trace_##name(void (*probe)(data_proto), void *data) \ 217 { \ 218 return tracepoint_probe_unregister(&__tracepoint_##name,\ 219 (void *)probe, data); \ 220 } \ 221 static inline void \ 222 check_trace_callback_type_##name(void (*cb)(data_proto)) \ 223 { \ 224 } \ 225 static inline bool \ 226 trace_##name##_enabled(void) \ 227 { \ 228 return static_key_false(&__tracepoint_##name.key); \ 229 } 230 231/* 232 * We have no guarantee that gcc and the linker won't up-align the tracepoint 233 * structures, so we create an array of pointers that will be used for iteration 234 * on the tracepoints. 235 */ 236#define DEFINE_TRACE_FN(name, reg, unreg) \ 237 static const char __tpstrtab_##name[] \ 238 __attribute__((section("__tracepoints_strings"))) = #name; \ 239 struct tracepoint __tracepoint_##name \ 240 __attribute__((section("__tracepoints"))) = \ 241 { __tpstrtab_##name, STATIC_KEY_INIT_FALSE, reg, unreg, NULL };\ 242 static struct tracepoint * const __tracepoint_ptr_##name __used \ 243 __attribute__((section("__tracepoints_ptrs"))) = \ 244 &__tracepoint_##name; 245 246#define DEFINE_TRACE(name) \ 247 DEFINE_TRACE_FN(name, NULL, NULL); 248 249#define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \ 250 EXPORT_SYMBOL_GPL(__tracepoint_##name) 251#define EXPORT_TRACEPOINT_SYMBOL(name) \ 252 EXPORT_SYMBOL(__tracepoint_##name) 253 254#else /* !TRACEPOINTS_ENABLED */ 255#define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \ 256 static inline void trace_##name(proto) \ 257 { } \ 258 static inline void trace_##name##_rcuidle(proto) \ 259 { } \ 260 static inline int \ 261 register_trace_##name(void (*probe)(data_proto), \ 262 void *data) \ 263 { \ 264 return -ENOSYS; \ 265 } \ 266 static inline int \ 267 unregister_trace_##name(void (*probe)(data_proto), \ 268 void *data) \ 269 { \ 270 return -ENOSYS; \ 271 } \ 272 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \ 273 { \ 274 } \ 275 static inline bool \ 276 trace_##name##_enabled(void) \ 277 { \ 278 return false; \ 279 } 280 281#define DEFINE_TRACE_FN(name, reg, unreg) 282#define DEFINE_TRACE(name) 283#define EXPORT_TRACEPOINT_SYMBOL_GPL(name) 284#define EXPORT_TRACEPOINT_SYMBOL(name) 285 286#endif /* TRACEPOINTS_ENABLED */ 287 288#ifdef CONFIG_TRACING 289/** 290 * tracepoint_string - register constant persistent string to trace system 291 * @str - a constant persistent string that will be referenced in tracepoints 292 * 293 * If constant strings are being used in tracepoints, it is faster and 294 * more efficient to just save the pointer to the string and reference 295 * that with a printf "%s" instead of saving the string in the ring buffer 296 * and wasting space and time. 297 * 298 * The problem with the above approach is that userspace tools that read 299 * the binary output of the trace buffers do not have access to the string. 300 * Instead they just show the address of the string which is not very 301 * useful to users. 302 * 303 * With tracepoint_string(), the string will be registered to the tracing 304 * system and exported to userspace via the debugfs/tracing/printk_formats 305 * file that maps the string address to the string text. This way userspace 306 * tools that read the binary buffers have a way to map the pointers to 307 * the ASCII strings they represent. 308 * 309 * The @str used must be a constant string and persistent as it would not 310 * make sense to show a string that no longer exists. But it is still fine 311 * to be used with modules, because when modules are unloaded, if they 312 * had tracepoints, the ring buffers are cleared too. As long as the string 313 * does not change during the life of the module, it is fine to use 314 * tracepoint_string() within a module. 315 */ 316#define tracepoint_string(str) \ 317 ({ \ 318 static const char *___tp_str __tracepoint_string = str; \ 319 ___tp_str; \ 320 }) 321#define __tracepoint_string __attribute__((section("__tracepoint_str"))) 322#else 323/* 324 * tracepoint_string() is used to save the string address for userspace 325 * tracing tools. When tracing isn't configured, there's no need to save 326 * anything. 327 */ 328# define tracepoint_string(str) str 329# define __tracepoint_string 330#endif 331 332/* 333 * The need for the DECLARE_TRACE_NOARGS() is to handle the prototype 334 * (void). "void" is a special value in a function prototype and can 335 * not be combined with other arguments. Since the DECLARE_TRACE() 336 * macro adds a data element at the beginning of the prototype, 337 * we need a way to differentiate "(void *data, proto)" from 338 * "(void *data, void)". The second prototype is invalid. 339 * 340 * DECLARE_TRACE_NOARGS() passes "void" as the tracepoint prototype 341 * and "void *__data" as the callback prototype. 342 * 343 * DECLARE_TRACE() passes "proto" as the tracepoint protoype and 344 * "void *__data, proto" as the callback prototype. 345 */ 346#define DECLARE_TRACE_NOARGS(name) \ 347 __DECLARE_TRACE(name, void, , \ 348 cpu_online(raw_smp_processor_id()), \ 349 void *__data, __data) 350 351#define DECLARE_TRACE(name, proto, args) \ 352 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \ 353 cpu_online(raw_smp_processor_id()), \ 354 PARAMS(void *__data, proto), \ 355 PARAMS(__data, args)) 356 357#define DECLARE_TRACE_CONDITION(name, proto, args, cond) \ 358 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \ 359 cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \ 360 PARAMS(void *__data, proto), \ 361 PARAMS(__data, args)) 362 363#define TRACE_EVENT_FLAGS(event, flag) 364 365#define TRACE_EVENT_PERF_PERM(event, expr...) 366 367#endif /* DECLARE_TRACE */ 368 369#ifndef TRACE_EVENT 370/* 371 * For use with the TRACE_EVENT macro: 372 * 373 * We define a tracepoint, its arguments, its printk format 374 * and its 'fast binary record' layout. 375 * 376 * Firstly, name your tracepoint via TRACE_EVENT(name : the 377 * 'subsystem_event' notation is fine. 378 * 379 * Think about this whole construct as the 380 * 'trace_sched_switch() function' from now on. 381 * 382 * 383 * TRACE_EVENT(sched_switch, 384 * 385 * * 386 * * A function has a regular function arguments 387 * * prototype, declare it via TP_PROTO(): 388 * * 389 * 390 * TP_PROTO(struct rq *rq, struct task_struct *prev, 391 * struct task_struct *next), 392 * 393 * * 394 * * Define the call signature of the 'function'. 395 * * (Design sidenote: we use this instead of a 396 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.) 397 * * 398 * 399 * TP_ARGS(rq, prev, next), 400 * 401 * * 402 * * Fast binary tracing: define the trace record via 403 * * TP_STRUCT__entry(). You can think about it like a 404 * * regular C structure local variable definition. 405 * * 406 * * This is how the trace record is structured and will 407 * * be saved into the ring buffer. These are the fields 408 * * that will be exposed to user-space in 409 * * /sys/kernel/debug/tracing/events/<*>/format. 410 * * 411 * * The declared 'local variable' is called '__entry' 412 * * 413 * * __field(pid_t, prev_prid) is equivalent to a standard declariton: 414 * * 415 * * pid_t prev_pid; 416 * * 417 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to: 418 * * 419 * * char prev_comm[TASK_COMM_LEN]; 420 * * 421 * 422 * TP_STRUCT__entry( 423 * __array( char, prev_comm, TASK_COMM_LEN ) 424 * __field( pid_t, prev_pid ) 425 * __field( int, prev_prio ) 426 * __array( char, next_comm, TASK_COMM_LEN ) 427 * __field( pid_t, next_pid ) 428 * __field( int, next_prio ) 429 * ), 430 * 431 * * 432 * * Assign the entry into the trace record, by embedding 433 * * a full C statement block into TP_fast_assign(). You 434 * * can refer to the trace record as '__entry' - 435 * * otherwise you can put arbitrary C code in here. 436 * * 437 * * Note: this C code will execute every time a trace event 438 * * happens, on an active tracepoint. 439 * * 440 * 441 * TP_fast_assign( 442 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN); 443 * __entry->prev_pid = prev->pid; 444 * __entry->prev_prio = prev->prio; 445 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN); 446 * __entry->next_pid = next->pid; 447 * __entry->next_prio = next->prio; 448 * ), 449 * 450 * * 451 * * Formatted output of a trace record via TP_printk(). 452 * * This is how the tracepoint will appear under ftrace 453 * * plugins that make use of this tracepoint. 454 * * 455 * * (raw-binary tracing wont actually perform this step.) 456 * * 457 * 458 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]", 459 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio, 460 * __entry->next_comm, __entry->next_pid, __entry->next_prio), 461 * 462 * ); 463 * 464 * This macro construct is thus used for the regular printk format 465 * tracing setup, it is used to construct a function pointer based 466 * tracepoint callback (this is used by programmatic plugins and 467 * can also by used by generic instrumentation like SystemTap), and 468 * it is also used to expose a structured trace record in 469 * /sys/kernel/debug/tracing/events/. 470 * 471 * A set of (un)registration functions can be passed to the variant 472 * TRACE_EVENT_FN to perform any (un)registration work. 473 */ 474 475#define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print) 476#define DEFINE_EVENT(template, name, proto, args) \ 477 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 478#define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\ 479 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 480#define DEFINE_EVENT_PRINT(template, name, proto, args, print) \ 481 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 482#define DEFINE_EVENT_CONDITION(template, name, proto, \ 483 args, cond) \ 484 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 485 PARAMS(args), PARAMS(cond)) 486 487#define TRACE_EVENT(name, proto, args, struct, assign, print) \ 488 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 489#define TRACE_EVENT_FN(name, proto, args, struct, \ 490 assign, print, reg, unreg) \ 491 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 492#define TRACE_EVENT_FN_COND(name, proto, args, cond, struct, \ 493 assign, print, reg, unreg) \ 494 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 495 PARAMS(args), PARAMS(cond)) 496#define TRACE_EVENT_CONDITION(name, proto, args, cond, \ 497 struct, assign, print) \ 498 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 499 PARAMS(args), PARAMS(cond)) 500 501#define TRACE_EVENT_FLAGS(event, flag) 502 503#define TRACE_EVENT_PERF_PERM(event, expr...) 504 505#endif /* ifdef TRACE_EVENT (see note above) */