at v5.11 19 kB view raw
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_unregister(struct tracepoint *tp, void *probe, void *data); 45extern void 46for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv), 47 void *priv); 48 49#ifdef CONFIG_MODULES 50struct tp_module { 51 struct list_head list; 52 struct module *mod; 53}; 54 55bool trace_module_has_bad_taint(struct module *mod); 56extern int register_tracepoint_module_notifier(struct notifier_block *nb); 57extern int unregister_tracepoint_module_notifier(struct notifier_block *nb); 58#else 59static inline bool trace_module_has_bad_taint(struct module *mod) 60{ 61 return false; 62} 63static inline 64int register_tracepoint_module_notifier(struct notifier_block *nb) 65{ 66 return 0; 67} 68static inline 69int unregister_tracepoint_module_notifier(struct notifier_block *nb) 70{ 71 return 0; 72} 73#endif /* CONFIG_MODULES */ 74 75/* 76 * tracepoint_synchronize_unregister must be called between the last tracepoint 77 * probe unregistration and the end of module exit to make sure there is no 78 * caller executing a probe when it is freed. 79 */ 80#ifdef CONFIG_TRACEPOINTS 81static inline void tracepoint_synchronize_unregister(void) 82{ 83 synchronize_srcu(&tracepoint_srcu); 84 synchronize_rcu(); 85} 86#else 87static inline void tracepoint_synchronize_unregister(void) 88{ } 89#endif 90 91#ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS 92extern int syscall_regfunc(void); 93extern void syscall_unregfunc(void); 94#endif /* CONFIG_HAVE_SYSCALL_TRACEPOINTS */ 95 96#ifndef PARAMS 97#define PARAMS(args...) args 98#endif 99 100#define TRACE_DEFINE_ENUM(x) 101#define TRACE_DEFINE_SIZEOF(x) 102 103#ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS 104static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p) 105{ 106 return offset_to_ptr(p); 107} 108 109#define __TRACEPOINT_ENTRY(name) \ 110 asm(" .section \"__tracepoints_ptrs\", \"a\" \n" \ 111 " .balign 4 \n" \ 112 " .long __tracepoint_" #name " - . \n" \ 113 " .previous \n") 114#else 115static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p) 116{ 117 return *p; 118} 119 120#define __TRACEPOINT_ENTRY(name) \ 121 static tracepoint_ptr_t __tracepoint_ptr_##name __used \ 122 __section("__tracepoints_ptrs") = &__tracepoint_##name 123#endif 124 125#endif /* _LINUX_TRACEPOINT_H */ 126 127/* 128 * Note: we keep the TRACE_EVENT and DECLARE_TRACE outside the include 129 * file ifdef protection. 130 * This is due to the way trace events work. If a file includes two 131 * trace event headers under one "CREATE_TRACE_POINTS" the first include 132 * will override the TRACE_EVENT and break the second include. 133 */ 134 135#ifndef DECLARE_TRACE 136 137#define TP_PROTO(args...) args 138#define TP_ARGS(args...) args 139#define TP_CONDITION(args...) args 140 141/* 142 * Individual subsystem my have a separate configuration to 143 * enable their tracepoints. By default, this file will create 144 * the tracepoints if CONFIG_TRACEPOINT is defined. If a subsystem 145 * wants to be able to disable its tracepoints from being created 146 * it can define NOTRACE before including the tracepoint headers. 147 */ 148#if defined(CONFIG_TRACEPOINTS) && !defined(NOTRACE) 149#define TRACEPOINTS_ENABLED 150#endif 151 152#ifdef TRACEPOINTS_ENABLED 153 154#ifdef CONFIG_HAVE_STATIC_CALL 155#define __DO_TRACE_CALL(name) static_call(tp_func_##name) 156#else 157#define __DO_TRACE_CALL(name) __traceiter_##name 158#endif /* CONFIG_HAVE_STATIC_CALL */ 159 160/* 161 * it_func[0] is never NULL because there is at least one element in the array 162 * when the array itself is non NULL. 163 * 164 * Note, the proto and args passed in includes "__data" as the first parameter. 165 * The reason for this is to handle the "void" prototype. If a tracepoint 166 * has a "void" prototype, then it is invalid to declare a function 167 * as "(void *, void)". 168 */ 169#define __DO_TRACE(name, proto, args, cond, rcuidle) \ 170 do { \ 171 struct tracepoint_func *it_func_ptr; \ 172 int __maybe_unused __idx = 0; \ 173 void *__data; \ 174 \ 175 if (!(cond)) \ 176 return; \ 177 \ 178 /* srcu can't be used from NMI */ \ 179 WARN_ON_ONCE(rcuidle && in_nmi()); \ 180 \ 181 /* keep srcu and sched-rcu usage consistent */ \ 182 preempt_disable_notrace(); \ 183 \ 184 /* \ 185 * For rcuidle callers, use srcu since sched-rcu \ 186 * doesn't work from the idle path. \ 187 */ \ 188 if (rcuidle) { \ 189 __idx = srcu_read_lock_notrace(&tracepoint_srcu);\ 190 rcu_irq_enter_irqson(); \ 191 } \ 192 \ 193 it_func_ptr = \ 194 rcu_dereference_raw((&__tracepoint_##name)->funcs); \ 195 if (it_func_ptr) { \ 196 __data = (it_func_ptr)->data; \ 197 __DO_TRACE_CALL(name)(args); \ 198 } \ 199 \ 200 if (rcuidle) { \ 201 rcu_irq_exit_irqson(); \ 202 srcu_read_unlock_notrace(&tracepoint_srcu, __idx);\ 203 } \ 204 \ 205 preempt_enable_notrace(); \ 206 } while (0) 207 208#ifndef MODULE 209#define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args) \ 210 static inline void trace_##name##_rcuidle(proto) \ 211 { \ 212 if (static_key_false(&__tracepoint_##name.key)) \ 213 __DO_TRACE(name, \ 214 TP_PROTO(data_proto), \ 215 TP_ARGS(data_args), \ 216 TP_CONDITION(cond), 1); \ 217 } 218#else 219#define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args) 220#endif 221 222/* 223 * Make sure the alignment of the structure in the __tracepoints section will 224 * not add unwanted padding between the beginning of the section and the 225 * structure. Force alignment to the same alignment as the section start. 226 * 227 * When lockdep is enabled, we make sure to always do the RCU portions of 228 * the tracepoint code, regardless of whether tracing is on. However, 229 * don't check if the condition is false, due to interaction with idle 230 * instrumentation. This lets us find RCU issues triggered with tracepoints 231 * even when this tracepoint is off. This code has no purpose other than 232 * poking RCU a bit. 233 */ 234#define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \ 235 extern int __traceiter_##name(data_proto); \ 236 DECLARE_STATIC_CALL(tp_func_##name, __traceiter_##name); \ 237 extern struct tracepoint __tracepoint_##name; \ 238 static inline void trace_##name(proto) \ 239 { \ 240 if (static_key_false(&__tracepoint_##name.key)) \ 241 __DO_TRACE(name, \ 242 TP_PROTO(data_proto), \ 243 TP_ARGS(data_args), \ 244 TP_CONDITION(cond), 0); \ 245 if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) { \ 246 rcu_read_lock_sched_notrace(); \ 247 rcu_dereference_sched(__tracepoint_##name.funcs);\ 248 rcu_read_unlock_sched_notrace(); \ 249 } \ 250 } \ 251 __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args), \ 252 PARAMS(cond), PARAMS(data_proto), PARAMS(data_args)) \ 253 static inline int \ 254 register_trace_##name(void (*probe)(data_proto), void *data) \ 255 { \ 256 return tracepoint_probe_register(&__tracepoint_##name, \ 257 (void *)probe, data); \ 258 } \ 259 static inline int \ 260 register_trace_prio_##name(void (*probe)(data_proto), void *data,\ 261 int prio) \ 262 { \ 263 return tracepoint_probe_register_prio(&__tracepoint_##name, \ 264 (void *)probe, data, prio); \ 265 } \ 266 static inline int \ 267 unregister_trace_##name(void (*probe)(data_proto), void *data) \ 268 { \ 269 return tracepoint_probe_unregister(&__tracepoint_##name,\ 270 (void *)probe, data); \ 271 } \ 272 static inline void \ 273 check_trace_callback_type_##name(void (*cb)(data_proto)) \ 274 { \ 275 } \ 276 static inline bool \ 277 trace_##name##_enabled(void) \ 278 { \ 279 return static_key_false(&__tracepoint_##name.key); \ 280 } 281 282/* 283 * We have no guarantee that gcc and the linker won't up-align the tracepoint 284 * structures, so we create an array of pointers that will be used for iteration 285 * on the tracepoints. 286 */ 287#define DEFINE_TRACE_FN(_name, _reg, _unreg, proto, args) \ 288 static const char __tpstrtab_##_name[] \ 289 __section("__tracepoints_strings") = #_name; \ 290 extern struct static_call_key STATIC_CALL_KEY(tp_func_##_name); \ 291 int __traceiter_##_name(void *__data, proto); \ 292 struct tracepoint __tracepoint_##_name __used \ 293 __section("__tracepoints") = { \ 294 .name = __tpstrtab_##_name, \ 295 .key = STATIC_KEY_INIT_FALSE, \ 296 .static_call_key = &STATIC_CALL_KEY(tp_func_##_name), \ 297 .static_call_tramp = STATIC_CALL_TRAMP_ADDR(tp_func_##_name), \ 298 .iterator = &__traceiter_##_name, \ 299 .regfunc = _reg, \ 300 .unregfunc = _unreg, \ 301 .funcs = NULL }; \ 302 __TRACEPOINT_ENTRY(_name); \ 303 int __traceiter_##_name(void *__data, proto) \ 304 { \ 305 struct tracepoint_func *it_func_ptr; \ 306 void *it_func; \ 307 \ 308 it_func_ptr = \ 309 rcu_dereference_raw((&__tracepoint_##_name)->funcs); \ 310 if (it_func_ptr) { \ 311 do { \ 312 it_func = (it_func_ptr)->func; \ 313 __data = (it_func_ptr)->data; \ 314 ((void(*)(void *, proto))(it_func))(__data, args); \ 315 } while ((++it_func_ptr)->func); \ 316 } \ 317 return 0; \ 318 } \ 319 DEFINE_STATIC_CALL(tp_func_##_name, __traceiter_##_name); 320 321#define DEFINE_TRACE(name, proto, args) \ 322 DEFINE_TRACE_FN(name, NULL, NULL, PARAMS(proto), PARAMS(args)); 323 324#define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \ 325 EXPORT_SYMBOL_GPL(__tracepoint_##name); \ 326 EXPORT_SYMBOL_GPL(__traceiter_##name); \ 327 EXPORT_STATIC_CALL_GPL(tp_func_##name) 328#define EXPORT_TRACEPOINT_SYMBOL(name) \ 329 EXPORT_SYMBOL(__tracepoint_##name); \ 330 EXPORT_SYMBOL(__traceiter_##name); \ 331 EXPORT_STATIC_CALL(tp_func_##name) 332 333 334#else /* !TRACEPOINTS_ENABLED */ 335#define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \ 336 static inline void trace_##name(proto) \ 337 { } \ 338 static inline void trace_##name##_rcuidle(proto) \ 339 { } \ 340 static inline int \ 341 register_trace_##name(void (*probe)(data_proto), \ 342 void *data) \ 343 { \ 344 return -ENOSYS; \ 345 } \ 346 static inline int \ 347 unregister_trace_##name(void (*probe)(data_proto), \ 348 void *data) \ 349 { \ 350 return -ENOSYS; \ 351 } \ 352 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \ 353 { \ 354 } \ 355 static inline bool \ 356 trace_##name##_enabled(void) \ 357 { \ 358 return false; \ 359 } 360 361#define DEFINE_TRACE_FN(name, reg, unreg, proto, args) 362#define DEFINE_TRACE(name, proto, args) 363#define EXPORT_TRACEPOINT_SYMBOL_GPL(name) 364#define EXPORT_TRACEPOINT_SYMBOL(name) 365 366#endif /* TRACEPOINTS_ENABLED */ 367 368#ifdef CONFIG_TRACING 369/** 370 * tracepoint_string - register constant persistent string to trace system 371 * @str - a constant persistent string that will be referenced in tracepoints 372 * 373 * If constant strings are being used in tracepoints, it is faster and 374 * more efficient to just save the pointer to the string and reference 375 * that with a printf "%s" instead of saving the string in the ring buffer 376 * and wasting space and time. 377 * 378 * The problem with the above approach is that userspace tools that read 379 * the binary output of the trace buffers do not have access to the string. 380 * Instead they just show the address of the string which is not very 381 * useful to users. 382 * 383 * With tracepoint_string(), the string will be registered to the tracing 384 * system and exported to userspace via the debugfs/tracing/printk_formats 385 * file that maps the string address to the string text. This way userspace 386 * tools that read the binary buffers have a way to map the pointers to 387 * the ASCII strings they represent. 388 * 389 * The @str used must be a constant string and persistent as it would not 390 * make sense to show a string that no longer exists. But it is still fine 391 * to be used with modules, because when modules are unloaded, if they 392 * had tracepoints, the ring buffers are cleared too. As long as the string 393 * does not change during the life of the module, it is fine to use 394 * tracepoint_string() within a module. 395 */ 396#define tracepoint_string(str) \ 397 ({ \ 398 static const char *___tp_str __tracepoint_string = str; \ 399 ___tp_str; \ 400 }) 401#define __tracepoint_string __used __section("__tracepoint_str") 402#else 403/* 404 * tracepoint_string() is used to save the string address for userspace 405 * tracing tools. When tracing isn't configured, there's no need to save 406 * anything. 407 */ 408# define tracepoint_string(str) str 409# define __tracepoint_string 410#endif 411 412#define DECLARE_TRACE(name, proto, args) \ 413 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \ 414 cpu_online(raw_smp_processor_id()), \ 415 PARAMS(void *__data, proto), \ 416 PARAMS(__data, args)) 417 418#define DECLARE_TRACE_CONDITION(name, proto, args, cond) \ 419 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \ 420 cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \ 421 PARAMS(void *__data, proto), \ 422 PARAMS(__data, args)) 423 424#define TRACE_EVENT_FLAGS(event, flag) 425 426#define TRACE_EVENT_PERF_PERM(event, expr...) 427 428#endif /* DECLARE_TRACE */ 429 430#ifndef TRACE_EVENT 431/* 432 * For use with the TRACE_EVENT macro: 433 * 434 * We define a tracepoint, its arguments, its printk format 435 * and its 'fast binary record' layout. 436 * 437 * Firstly, name your tracepoint via TRACE_EVENT(name : the 438 * 'subsystem_event' notation is fine. 439 * 440 * Think about this whole construct as the 441 * 'trace_sched_switch() function' from now on. 442 * 443 * 444 * TRACE_EVENT(sched_switch, 445 * 446 * * 447 * * A function has a regular function arguments 448 * * prototype, declare it via TP_PROTO(): 449 * * 450 * 451 * TP_PROTO(struct rq *rq, struct task_struct *prev, 452 * struct task_struct *next), 453 * 454 * * 455 * * Define the call signature of the 'function'. 456 * * (Design sidenote: we use this instead of a 457 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.) 458 * * 459 * 460 * TP_ARGS(rq, prev, next), 461 * 462 * * 463 * * Fast binary tracing: define the trace record via 464 * * TP_STRUCT__entry(). You can think about it like a 465 * * regular C structure local variable definition. 466 * * 467 * * This is how the trace record is structured and will 468 * * be saved into the ring buffer. These are the fields 469 * * that will be exposed to user-space in 470 * * /sys/kernel/debug/tracing/events/<*>/format. 471 * * 472 * * The declared 'local variable' is called '__entry' 473 * * 474 * * __field(pid_t, prev_prid) is equivalent to a standard declariton: 475 * * 476 * * pid_t prev_pid; 477 * * 478 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to: 479 * * 480 * * char prev_comm[TASK_COMM_LEN]; 481 * * 482 * 483 * TP_STRUCT__entry( 484 * __array( char, prev_comm, TASK_COMM_LEN ) 485 * __field( pid_t, prev_pid ) 486 * __field( int, prev_prio ) 487 * __array( char, next_comm, TASK_COMM_LEN ) 488 * __field( pid_t, next_pid ) 489 * __field( int, next_prio ) 490 * ), 491 * 492 * * 493 * * Assign the entry into the trace record, by embedding 494 * * a full C statement block into TP_fast_assign(). You 495 * * can refer to the trace record as '__entry' - 496 * * otherwise you can put arbitrary C code in here. 497 * * 498 * * Note: this C code will execute every time a trace event 499 * * happens, on an active tracepoint. 500 * * 501 * 502 * TP_fast_assign( 503 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN); 504 * __entry->prev_pid = prev->pid; 505 * __entry->prev_prio = prev->prio; 506 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN); 507 * __entry->next_pid = next->pid; 508 * __entry->next_prio = next->prio; 509 * ), 510 * 511 * * 512 * * Formatted output of a trace record via TP_printk(). 513 * * This is how the tracepoint will appear under ftrace 514 * * plugins that make use of this tracepoint. 515 * * 516 * * (raw-binary tracing wont actually perform this step.) 517 * * 518 * 519 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]", 520 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio, 521 * __entry->next_comm, __entry->next_pid, __entry->next_prio), 522 * 523 * ); 524 * 525 * This macro construct is thus used for the regular printk format 526 * tracing setup, it is used to construct a function pointer based 527 * tracepoint callback (this is used by programmatic plugins and 528 * can also by used by generic instrumentation like SystemTap), and 529 * it is also used to expose a structured trace record in 530 * /sys/kernel/debug/tracing/events/. 531 * 532 * A set of (un)registration functions can be passed to the variant 533 * TRACE_EVENT_FN to perform any (un)registration work. 534 */ 535 536#define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print) 537#define DEFINE_EVENT(template, name, proto, args) \ 538 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 539#define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\ 540 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 541#define DEFINE_EVENT_PRINT(template, name, proto, args, print) \ 542 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 543#define DEFINE_EVENT_CONDITION(template, name, proto, \ 544 args, cond) \ 545 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 546 PARAMS(args), PARAMS(cond)) 547 548#define TRACE_EVENT(name, proto, args, struct, assign, print) \ 549 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 550#define TRACE_EVENT_FN(name, proto, args, struct, \ 551 assign, print, reg, unreg) \ 552 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 553#define TRACE_EVENT_FN_COND(name, proto, args, cond, struct, \ 554 assign, print, reg, unreg) \ 555 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 556 PARAMS(args), PARAMS(cond)) 557#define TRACE_EVENT_CONDITION(name, proto, args, cond, \ 558 struct, assign, print) \ 559 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 560 PARAMS(args), PARAMS(cond)) 561 562#define TRACE_EVENT_FLAGS(event, flag) 563 564#define TRACE_EVENT_PERF_PERM(event, expr...) 565 566#define DECLARE_EVENT_NOP(name, proto, args) \ 567 static inline void trace_##name(proto) \ 568 { } \ 569 static inline bool trace_##name##_enabled(void) \ 570 { \ 571 return false; \ 572 } 573 574#define TRACE_EVENT_NOP(name, proto, args, struct, assign, print) \ 575 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args)) 576 577#define DECLARE_EVENT_CLASS_NOP(name, proto, args, tstruct, assign, print) 578#define DEFINE_EVENT_NOP(template, name, proto, args) \ 579 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args)) 580 581#endif /* ifdef TRACE_EVENT (see note above) */