1/* SPDX-License-Identifier: GPL-2.0-only */ 2#ifndef _LINUX_TRACEPOINT_H 3#define _LINUX_TRACEPOINT_H 4 5/* 6 * Kernel Tracepoint API. 7 * 8 * See Documentation/trace/tracepoints.rst. 9 * 10 * Copyright (C) 2008-2014 Mathieu Desnoyers <mathieu.desnoyers@efficios.com> 11 * 12 * Heavily inspired from the Linux Kernel Markers. 13 */ 14 15#include <linux/smp.h> 16#include <linux/srcu.h> 17#include <linux/errno.h> 18#include <linux/types.h> 19#include <linux/cpumask.h> 20#include <linux/rcupdate.h> 21#include <linux/tracepoint-defs.h> 22#include <linux/static_call.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 struct srcu_struct tracepoint_srcu; 37 38extern int 39tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data); 40extern int 41tracepoint_probe_register_prio(struct tracepoint *tp, void *probe, void *data, 42 int prio); 43extern int 44tracepoint_probe_register_prio_may_exist(struct tracepoint *tp, void *probe, void *data, 45 int prio); 46extern int 47tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data); 48static inline int 49tracepoint_probe_register_may_exist(struct tracepoint *tp, void *probe, 50 void *data) 51{ 52 return tracepoint_probe_register_prio_may_exist(tp, probe, data, 53 TRACEPOINT_DEFAULT_PRIO); 54} 55extern void 56for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv), 57 void *priv); 58 59#ifdef CONFIG_MODULES 60struct tp_module { 61 struct list_head list; 62 struct module *mod; 63}; 64 65bool trace_module_has_bad_taint(struct module *mod); 66extern int register_tracepoint_module_notifier(struct notifier_block *nb); 67extern int unregister_tracepoint_module_notifier(struct notifier_block *nb); 68#else 69static inline bool trace_module_has_bad_taint(struct module *mod) 70{ 71 return false; 72} 73static inline 74int register_tracepoint_module_notifier(struct notifier_block *nb) 75{ 76 return 0; 77} 78static inline 79int unregister_tracepoint_module_notifier(struct notifier_block *nb) 80{ 81 return 0; 82} 83#endif /* CONFIG_MODULES */ 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#ifdef CONFIG_TRACEPOINTS 91static inline void tracepoint_synchronize_unregister(void) 92{ 93 synchronize_srcu(&tracepoint_srcu); 94 synchronize_rcu(); 95} 96#else 97static inline void tracepoint_synchronize_unregister(void) 98{ } 99#endif 100 101#ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS 102extern int syscall_regfunc(void); 103extern void syscall_unregfunc(void); 104#endif /* CONFIG_HAVE_SYSCALL_TRACEPOINTS */ 105 106#ifndef PARAMS 107#define PARAMS(args...) args 108#endif 109 110#define TRACE_DEFINE_ENUM(x) 111#define TRACE_DEFINE_SIZEOF(x) 112 113#ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS 114static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p) 115{ 116 return offset_to_ptr(p); 117} 118 119#define __TRACEPOINT_ENTRY(name) \ 120 asm(" .section \"__tracepoints_ptrs\", \"a\" \n" \ 121 " .balign 4 \n" \ 122 " .long __tracepoint_" #name " - . \n" \ 123 " .previous \n") 124#else 125static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p) 126{ 127 return *p; 128} 129 130#define __TRACEPOINT_ENTRY(name) \ 131 static tracepoint_ptr_t __tracepoint_ptr_##name __used \ 132 __section("__tracepoints_ptrs") = &__tracepoint_##name 133#endif 134 135#endif /* _LINUX_TRACEPOINT_H */ 136 137/* 138 * Note: we keep the TRACE_EVENT and DECLARE_TRACE outside the include 139 * file ifdef protection. 140 * This is due to the way trace events work. If a file includes two 141 * trace event headers under one "CREATE_TRACE_POINTS" the first include 142 * will override the TRACE_EVENT and break the second include. 143 */ 144 145#ifndef DECLARE_TRACE 146 147#define TP_PROTO(args...) args 148#define TP_ARGS(args...) args 149#define TP_CONDITION(args...) args 150 151/* 152 * Individual subsystem my have a separate configuration to 153 * enable their tracepoints. By default, this file will create 154 * the tracepoints if CONFIG_TRACEPOINTS is defined. If a subsystem 155 * wants to be able to disable its tracepoints from being created 156 * it can define NOTRACE before including the tracepoint headers. 157 */ 158#if defined(CONFIG_TRACEPOINTS) && !defined(NOTRACE) 159#define TRACEPOINTS_ENABLED 160#endif 161 162#ifdef TRACEPOINTS_ENABLED 163 164#ifdef CONFIG_HAVE_STATIC_CALL 165#define __DO_TRACE_CALL(name, args) \ 166 do { \ 167 struct tracepoint_func *it_func_ptr; \ 168 void *__data; \ 169 it_func_ptr = \ 170 rcu_dereference_raw((&__tracepoint_##name)->funcs); \ 171 if (it_func_ptr) { \ 172 __data = (it_func_ptr)->data; \ 173 static_call(tp_func_##name)(__data, args); \ 174 } \ 175 } while (0) 176#else 177#define __DO_TRACE_CALL(name, args) __traceiter_##name(NULL, args) 178#endif /* CONFIG_HAVE_STATIC_CALL */ 179 180/* 181 * ARCH_WANTS_NO_INSTR archs are expected to have sanitized entry and idle 182 * code that disallow any/all tracing/instrumentation when RCU isn't watching. 183 */ 184#ifdef CONFIG_ARCH_WANTS_NO_INSTR 185#define RCUIDLE_COND(rcuidle) (rcuidle) 186#else 187/* srcu can't be used from NMI */ 188#define RCUIDLE_COND(rcuidle) (rcuidle && in_nmi()) 189#endif 190 191/* 192 * it_func[0] is never NULL because there is at least one element in the array 193 * when the array itself is non NULL. 194 */ 195#define __DO_TRACE(name, args, cond, rcuidle) \ 196 do { \ 197 int __maybe_unused __idx = 0; \ 198 \ 199 if (!(cond)) \ 200 return; \ 201 \ 202 if (WARN_ONCE(RCUIDLE_COND(rcuidle), \ 203 "Bad RCU usage for tracepoint")) \ 204 return; \ 205 \ 206 /* keep srcu and sched-rcu usage consistent */ \ 207 preempt_disable_notrace(); \ 208 \ 209 /* \ 210 * For rcuidle callers, use srcu since sched-rcu \ 211 * doesn't work from the idle path. \ 212 */ \ 213 if (rcuidle) { \ 214 __idx = srcu_read_lock_notrace(&tracepoint_srcu);\ 215 ct_irq_enter_irqson(); \ 216 } \ 217 \ 218 __DO_TRACE_CALL(name, TP_ARGS(args)); \ 219 \ 220 if (rcuidle) { \ 221 ct_irq_exit_irqson(); \ 222 srcu_read_unlock_notrace(&tracepoint_srcu, __idx);\ 223 } \ 224 \ 225 preempt_enable_notrace(); \ 226 } while (0) 227 228#ifndef MODULE 229#define __DECLARE_TRACE_RCU(name, proto, args, cond) \ 230 static inline void trace_##name##_rcuidle(proto) \ 231 { \ 232 if (static_key_false(&__tracepoint_##name.key)) \ 233 __DO_TRACE(name, \ 234 TP_ARGS(args), \ 235 TP_CONDITION(cond), 1); \ 236 } 237#else 238#define __DECLARE_TRACE_RCU(name, proto, args, cond) 239#endif 240 241/* 242 * Make sure the alignment of the structure in the __tracepoints section will 243 * not add unwanted padding between the beginning of the section and the 244 * structure. Force alignment to the same alignment as the section start. 245 * 246 * When lockdep is enabled, we make sure to always test if RCU is 247 * "watching" regardless if the tracepoint is enabled or not. Tracepoints 248 * require RCU to be active, and it should always warn at the tracepoint 249 * site if it is not watching, as it will need to be active when the 250 * tracepoint is enabled. 251 */ 252#define __DECLARE_TRACE(name, proto, args, cond, data_proto) \ 253 extern int __traceiter_##name(data_proto); \ 254 DECLARE_STATIC_CALL(tp_func_##name, __traceiter_##name); \ 255 extern struct tracepoint __tracepoint_##name; \ 256 static inline void trace_##name(proto) \ 257 { \ 258 if (static_key_false(&__tracepoint_##name.key)) \ 259 __DO_TRACE(name, \ 260 TP_ARGS(args), \ 261 TP_CONDITION(cond), 0); \ 262 if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) { \ 263 WARN_ONCE(!rcu_is_watching(), \ 264 "RCU not watching for tracepoint"); \ 265 } \ 266 } \ 267 __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args), \ 268 PARAMS(cond)) \ 269 static inline int \ 270 register_trace_##name(void (*probe)(data_proto), void *data) \ 271 { \ 272 return tracepoint_probe_register(&__tracepoint_##name, \ 273 (void *)probe, data); \ 274 } \ 275 static inline int \ 276 register_trace_prio_##name(void (*probe)(data_proto), void *data,\ 277 int prio) \ 278 { \ 279 return tracepoint_probe_register_prio(&__tracepoint_##name, \ 280 (void *)probe, data, prio); \ 281 } \ 282 static inline int \ 283 unregister_trace_##name(void (*probe)(data_proto), void *data) \ 284 { \ 285 return tracepoint_probe_unregister(&__tracepoint_##name,\ 286 (void *)probe, data); \ 287 } \ 288 static inline void \ 289 check_trace_callback_type_##name(void (*cb)(data_proto)) \ 290 { \ 291 } \ 292 static inline bool \ 293 trace_##name##_enabled(void) \ 294 { \ 295 return static_key_false(&__tracepoint_##name.key); \ 296 } 297 298/* 299 * We have no guarantee that gcc and the linker won't up-align the tracepoint 300 * structures, so we create an array of pointers that will be used for iteration 301 * on the tracepoints. 302 */ 303#define DEFINE_TRACE_FN(_name, _reg, _unreg, proto, args) \ 304 static const char __tpstrtab_##_name[] \ 305 __section("__tracepoints_strings") = #_name; \ 306 extern struct static_call_key STATIC_CALL_KEY(tp_func_##_name); \ 307 int __traceiter_##_name(void *__data, proto); \ 308 void __probestub_##_name(void *__data, proto); \ 309 struct tracepoint __tracepoint_##_name __used \ 310 __section("__tracepoints") = { \ 311 .name = __tpstrtab_##_name, \ 312 .key = STATIC_KEY_INIT_FALSE, \ 313 .static_call_key = &STATIC_CALL_KEY(tp_func_##_name), \ 314 .static_call_tramp = STATIC_CALL_TRAMP_ADDR(tp_func_##_name), \ 315 .iterator = &__traceiter_##_name, \ 316 .probestub = &__probestub_##_name, \ 317 .regfunc = _reg, \ 318 .unregfunc = _unreg, \ 319 .funcs = NULL }; \ 320 __TRACEPOINT_ENTRY(_name); \ 321 int __traceiter_##_name(void *__data, proto) \ 322 { \ 323 struct tracepoint_func *it_func_ptr; \ 324 void *it_func; \ 325 \ 326 it_func_ptr = \ 327 rcu_dereference_raw((&__tracepoint_##_name)->funcs); \ 328 if (it_func_ptr) { \ 329 do { \ 330 it_func = READ_ONCE((it_func_ptr)->func); \ 331 __data = (it_func_ptr)->data; \ 332 ((void(*)(void *, proto))(it_func))(__data, args); \ 333 } while ((++it_func_ptr)->func); \ 334 } \ 335 return 0; \ 336 } \ 337 void __probestub_##_name(void *__data, proto) \ 338 { \ 339 } \ 340 DEFINE_STATIC_CALL(tp_func_##_name, __traceiter_##_name); 341 342#define DEFINE_TRACE(name, proto, args) \ 343 DEFINE_TRACE_FN(name, NULL, NULL, PARAMS(proto), PARAMS(args)); 344 345#define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \ 346 EXPORT_SYMBOL_GPL(__tracepoint_##name); \ 347 EXPORT_SYMBOL_GPL(__traceiter_##name); \ 348 EXPORT_STATIC_CALL_GPL(tp_func_##name) 349#define EXPORT_TRACEPOINT_SYMBOL(name) \ 350 EXPORT_SYMBOL(__tracepoint_##name); \ 351 EXPORT_SYMBOL(__traceiter_##name); \ 352 EXPORT_STATIC_CALL(tp_func_##name) 353 354 355#else /* !TRACEPOINTS_ENABLED */ 356#define __DECLARE_TRACE(name, proto, args, cond, data_proto) \ 357 static inline void trace_##name(proto) \ 358 { } \ 359 static inline void trace_##name##_rcuidle(proto) \ 360 { } \ 361 static inline int \ 362 register_trace_##name(void (*probe)(data_proto), \ 363 void *data) \ 364 { \ 365 return -ENOSYS; \ 366 } \ 367 static inline int \ 368 unregister_trace_##name(void (*probe)(data_proto), \ 369 void *data) \ 370 { \ 371 return -ENOSYS; \ 372 } \ 373 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \ 374 { \ 375 } \ 376 static inline bool \ 377 trace_##name##_enabled(void) \ 378 { \ 379 return false; \ 380 } 381 382#define DEFINE_TRACE_FN(name, reg, unreg, proto, args) 383#define DEFINE_TRACE(name, proto, args) 384#define EXPORT_TRACEPOINT_SYMBOL_GPL(name) 385#define EXPORT_TRACEPOINT_SYMBOL(name) 386 387#endif /* TRACEPOINTS_ENABLED */ 388 389#ifdef CONFIG_TRACING 390/** 391 * tracepoint_string - register constant persistent string to trace system 392 * @str - a constant persistent string that will be referenced in tracepoints 393 * 394 * If constant strings are being used in tracepoints, it is faster and 395 * more efficient to just save the pointer to the string and reference 396 * that with a printf "%s" instead of saving the string in the ring buffer 397 * and wasting space and time. 398 * 399 * The problem with the above approach is that userspace tools that read 400 * the binary output of the trace buffers do not have access to the string. 401 * Instead they just show the address of the string which is not very 402 * useful to users. 403 * 404 * With tracepoint_string(), the string will be registered to the tracing 405 * system and exported to userspace via the debugfs/tracing/printk_formats 406 * file that maps the string address to the string text. This way userspace 407 * tools that read the binary buffers have a way to map the pointers to 408 * the ASCII strings they represent. 409 * 410 * The @str used must be a constant string and persistent as it would not 411 * make sense to show a string that no longer exists. But it is still fine 412 * to be used with modules, because when modules are unloaded, if they 413 * had tracepoints, the ring buffers are cleared too. As long as the string 414 * does not change during the life of the module, it is fine to use 415 * tracepoint_string() within a module. 416 */ 417#define tracepoint_string(str) \ 418 ({ \ 419 static const char *___tp_str __tracepoint_string = str; \ 420 ___tp_str; \ 421 }) 422#define __tracepoint_string __used __section("__tracepoint_str") 423#else 424/* 425 * tracepoint_string() is used to save the string address for userspace 426 * tracing tools. When tracing isn't configured, there's no need to save 427 * anything. 428 */ 429# define tracepoint_string(str) str 430# define __tracepoint_string 431#endif 432 433#define DECLARE_TRACE(name, proto, args) \ 434 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \ 435 cpu_online(raw_smp_processor_id()), \ 436 PARAMS(void *__data, proto)) 437 438#define DECLARE_TRACE_CONDITION(name, proto, args, cond) \ 439 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \ 440 cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \ 441 PARAMS(void *__data, proto)) 442 443#define TRACE_EVENT_FLAGS(event, flag) 444 445#define TRACE_EVENT_PERF_PERM(event, expr...) 446 447#endif /* DECLARE_TRACE */ 448 449#ifndef TRACE_EVENT 450/* 451 * For use with the TRACE_EVENT macro: 452 * 453 * We define a tracepoint, its arguments, its printk format 454 * and its 'fast binary record' layout. 455 * 456 * Firstly, name your tracepoint via TRACE_EVENT(name : the 457 * 'subsystem_event' notation is fine. 458 * 459 * Think about this whole construct as the 460 * 'trace_sched_switch() function' from now on. 461 * 462 * 463 * TRACE_EVENT(sched_switch, 464 * 465 * * 466 * * A function has a regular function arguments 467 * * prototype, declare it via TP_PROTO(): 468 * * 469 * 470 * TP_PROTO(struct rq *rq, struct task_struct *prev, 471 * struct task_struct *next), 472 * 473 * * 474 * * Define the call signature of the 'function'. 475 * * (Design sidenote: we use this instead of a 476 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.) 477 * * 478 * 479 * TP_ARGS(rq, prev, next), 480 * 481 * * 482 * * Fast binary tracing: define the trace record via 483 * * TP_STRUCT__entry(). You can think about it like a 484 * * regular C structure local variable definition. 485 * * 486 * * This is how the trace record is structured and will 487 * * be saved into the ring buffer. These are the fields 488 * * that will be exposed to user-space in 489 * * /sys/kernel/tracing/events/<*>/format. 490 * * 491 * * The declared 'local variable' is called '__entry' 492 * * 493 * * __field(pid_t, prev_pid) is equivalent to a standard declaration: 494 * * 495 * * pid_t prev_pid; 496 * * 497 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to: 498 * * 499 * * char prev_comm[TASK_COMM_LEN]; 500 * * 501 * 502 * TP_STRUCT__entry( 503 * __array( char, prev_comm, TASK_COMM_LEN ) 504 * __field( pid_t, prev_pid ) 505 * __field( int, prev_prio ) 506 * __array( char, next_comm, TASK_COMM_LEN ) 507 * __field( pid_t, next_pid ) 508 * __field( int, next_prio ) 509 * ), 510 * 511 * * 512 * * Assign the entry into the trace record, by embedding 513 * * a full C statement block into TP_fast_assign(). You 514 * * can refer to the trace record as '__entry' - 515 * * otherwise you can put arbitrary C code in here. 516 * * 517 * * Note: this C code will execute every time a trace event 518 * * happens, on an active tracepoint. 519 * * 520 * 521 * TP_fast_assign( 522 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN); 523 * __entry->prev_pid = prev->pid; 524 * __entry->prev_prio = prev->prio; 525 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN); 526 * __entry->next_pid = next->pid; 527 * __entry->next_prio = next->prio; 528 * ), 529 * 530 * * 531 * * Formatted output of a trace record via TP_printk(). 532 * * This is how the tracepoint will appear under ftrace 533 * * plugins that make use of this tracepoint. 534 * * 535 * * (raw-binary tracing wont actually perform this step.) 536 * * 537 * 538 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]", 539 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio, 540 * __entry->next_comm, __entry->next_pid, __entry->next_prio), 541 * 542 * ); 543 * 544 * This macro construct is thus used for the regular printk format 545 * tracing setup, it is used to construct a function pointer based 546 * tracepoint callback (this is used by programmatic plugins and 547 * can also by used by generic instrumentation like SystemTap), and 548 * it is also used to expose a structured trace record in 549 * /sys/kernel/tracing/events/. 550 * 551 * A set of (un)registration functions can be passed to the variant 552 * TRACE_EVENT_FN to perform any (un)registration work. 553 */ 554 555#define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print) 556#define DEFINE_EVENT(template, name, proto, args) \ 557 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 558#define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\ 559 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 560#define DEFINE_EVENT_PRINT(template, name, proto, args, print) \ 561 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 562#define DEFINE_EVENT_CONDITION(template, name, proto, \ 563 args, cond) \ 564 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 565 PARAMS(args), PARAMS(cond)) 566 567#define TRACE_EVENT(name, proto, args, struct, assign, print) \ 568 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 569#define TRACE_EVENT_FN(name, proto, args, struct, \ 570 assign, print, reg, unreg) \ 571 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 572#define TRACE_EVENT_FN_COND(name, proto, args, cond, struct, \ 573 assign, print, reg, unreg) \ 574 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 575 PARAMS(args), PARAMS(cond)) 576#define TRACE_EVENT_CONDITION(name, proto, args, cond, \ 577 struct, assign, print) \ 578 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 579 PARAMS(args), PARAMS(cond)) 580 581#define TRACE_EVENT_FLAGS(event, flag) 582 583#define TRACE_EVENT_PERF_PERM(event, expr...) 584 585#define DECLARE_EVENT_NOP(name, proto, args) \ 586 static inline void trace_##name(proto) \ 587 { } \ 588 static inline bool trace_##name##_enabled(void) \ 589 { \ 590 return false; \ 591 } 592 593#define TRACE_EVENT_NOP(name, proto, args, struct, assign, print) \ 594 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args)) 595 596#define DECLARE_EVENT_CLASS_NOP(name, proto, args, tstruct, assign, print) 597#define DEFINE_EVENT_NOP(template, name, proto, args) \ 598 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args)) 599 600#endif /* ifdef TRACE_EVENT (see note above) */ 601