at v6.11 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/rcupdate.h> 20#include <linux/tracepoint-defs.h> 21#include <linux/static_call.h> 22 23struct module; 24struct tracepoint; 25struct notifier_block; 26 27struct trace_eval_map { 28 const char *system; 29 const char *eval_string; 30 unsigned long eval_value; 31}; 32 33#define TRACEPOINT_DEFAULT_PRIO 10 34 35extern struct srcu_struct tracepoint_srcu; 36 37extern int 38tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data); 39extern int 40tracepoint_probe_register_prio(struct tracepoint *tp, void *probe, void *data, 41 int prio); 42extern int 43tracepoint_probe_register_prio_may_exist(struct tracepoint *tp, void *probe, void *data, 44 int prio); 45extern int 46tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data); 47static inline int 48tracepoint_probe_register_may_exist(struct tracepoint *tp, void *probe, 49 void *data) 50{ 51 return tracepoint_probe_register_prio_may_exist(tp, probe, data, 52 TRACEPOINT_DEFAULT_PRIO); 53} 54extern void 55for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv), 56 void *priv); 57 58#ifdef CONFIG_MODULES 59struct tp_module { 60 struct list_head list; 61 struct module *mod; 62}; 63 64bool trace_module_has_bad_taint(struct module *mod); 65extern int register_tracepoint_module_notifier(struct notifier_block *nb); 66extern int unregister_tracepoint_module_notifier(struct notifier_block *nb); 67#else 68static inline bool trace_module_has_bad_taint(struct module *mod) 69{ 70 return false; 71} 72static inline 73int register_tracepoint_module_notifier(struct notifier_block *nb) 74{ 75 return 0; 76} 77static inline 78int unregister_tracepoint_module_notifier(struct notifier_block *nb) 79{ 80 return 0; 81} 82#endif /* CONFIG_MODULES */ 83 84/* 85 * tracepoint_synchronize_unregister must be called between the last tracepoint 86 * probe unregistration and the end of module exit to make sure there is no 87 * caller executing a probe when it is freed. 88 */ 89#ifdef CONFIG_TRACEPOINTS 90static inline void tracepoint_synchronize_unregister(void) 91{ 92 synchronize_srcu(&tracepoint_srcu); 93 synchronize_rcu(); 94} 95#else 96static inline void tracepoint_synchronize_unregister(void) 97{ } 98#endif 99 100#ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS 101extern int syscall_regfunc(void); 102extern void syscall_unregfunc(void); 103#endif /* CONFIG_HAVE_SYSCALL_TRACEPOINTS */ 104 105#ifndef PARAMS 106#define PARAMS(args...) args 107#endif 108 109#define TRACE_DEFINE_ENUM(x) 110#define TRACE_DEFINE_SIZEOF(x) 111 112#ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS 113static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p) 114{ 115 return offset_to_ptr(p); 116} 117 118#define __TRACEPOINT_ENTRY(name) \ 119 asm(" .section \"__tracepoints_ptrs\", \"a\" \n" \ 120 " .balign 4 \n" \ 121 " .long __tracepoint_" #name " - . \n" \ 122 " .previous \n") 123#else 124static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p) 125{ 126 return *p; 127} 128 129#define __TRACEPOINT_ENTRY(name) \ 130 static tracepoint_ptr_t __tracepoint_ptr_##name __used \ 131 __section("__tracepoints_ptrs") = &__tracepoint_##name 132#endif 133 134#endif /* _LINUX_TRACEPOINT_H */ 135 136/* 137 * Note: we keep the TRACE_EVENT and DECLARE_TRACE outside the include 138 * file ifdef protection. 139 * This is due to the way trace events work. If a file includes two 140 * trace event headers under one "CREATE_TRACE_POINTS" the first include 141 * will override the TRACE_EVENT and break the second include. 142 */ 143 144#ifndef DECLARE_TRACE 145 146#define TP_PROTO(args...) args 147#define TP_ARGS(args...) args 148#define TP_CONDITION(args...) args 149 150/* 151 * Individual subsystem my have a separate configuration to 152 * enable their tracepoints. By default, this file will create 153 * the tracepoints if CONFIG_TRACEPOINTS is defined. If a subsystem 154 * wants to be able to disable its tracepoints from being created 155 * it can define NOTRACE before including the tracepoint headers. 156 */ 157#if defined(CONFIG_TRACEPOINTS) && !defined(NOTRACE) 158#define TRACEPOINTS_ENABLED 159#endif 160 161#ifdef TRACEPOINTS_ENABLED 162 163#ifdef CONFIG_HAVE_STATIC_CALL 164#define __DO_TRACE_CALL(name, args) \ 165 do { \ 166 struct tracepoint_func *it_func_ptr; \ 167 void *__data; \ 168 it_func_ptr = \ 169 rcu_dereference_raw((&__tracepoint_##name)->funcs); \ 170 if (it_func_ptr) { \ 171 __data = (it_func_ptr)->data; \ 172 static_call(tp_func_##name)(__data, args); \ 173 } \ 174 } while (0) 175#else 176#define __DO_TRACE_CALL(name, args) __traceiter_##name(NULL, args) 177#endif /* CONFIG_HAVE_STATIC_CALL */ 178 179/* 180 * ARCH_WANTS_NO_INSTR archs are expected to have sanitized entry and idle 181 * code that disallow any/all tracing/instrumentation when RCU isn't watching. 182 */ 183#ifdef CONFIG_ARCH_WANTS_NO_INSTR 184#define RCUIDLE_COND(rcuidle) (rcuidle) 185#else 186/* srcu can't be used from NMI */ 187#define RCUIDLE_COND(rcuidle) (rcuidle && in_nmi()) 188#endif 189 190/* 191 * it_func[0] is never NULL because there is at least one element in the array 192 * when the array itself is non NULL. 193 */ 194#define __DO_TRACE(name, args, cond, rcuidle) \ 195 do { \ 196 int __maybe_unused __idx = 0; \ 197 \ 198 if (!(cond)) \ 199 return; \ 200 \ 201 if (WARN_ONCE(RCUIDLE_COND(rcuidle), \ 202 "Bad RCU usage for tracepoint")) \ 203 return; \ 204 \ 205 /* keep srcu and sched-rcu usage consistent */ \ 206 preempt_disable_notrace(); \ 207 \ 208 /* \ 209 * For rcuidle callers, use srcu since sched-rcu \ 210 * doesn't work from the idle path. \ 211 */ \ 212 if (rcuidle) { \ 213 __idx = srcu_read_lock_notrace(&tracepoint_srcu);\ 214 ct_irq_enter_irqson(); \ 215 } \ 216 \ 217 __DO_TRACE_CALL(name, TP_ARGS(args)); \ 218 \ 219 if (rcuidle) { \ 220 ct_irq_exit_irqson(); \ 221 srcu_read_unlock_notrace(&tracepoint_srcu, __idx);\ 222 } \ 223 \ 224 preempt_enable_notrace(); \ 225 } while (0) 226 227#ifndef MODULE 228#define __DECLARE_TRACE_RCU(name, proto, args, cond) \ 229 static inline void trace_##name##_rcuidle(proto) \ 230 { \ 231 if (static_key_false(&__tracepoint_##name.key)) \ 232 __DO_TRACE(name, \ 233 TP_ARGS(args), \ 234 TP_CONDITION(cond), 1); \ 235 } 236#else 237#define __DECLARE_TRACE_RCU(name, proto, args, cond) 238#endif 239 240/* 241 * Make sure the alignment of the structure in the __tracepoints section will 242 * not add unwanted padding between the beginning of the section and the 243 * structure. Force alignment to the same alignment as the section start. 244 * 245 * When lockdep is enabled, we make sure to always test if RCU is 246 * "watching" regardless if the tracepoint is enabled or not. Tracepoints 247 * require RCU to be active, and it should always warn at the tracepoint 248 * site if it is not watching, as it will need to be active when the 249 * tracepoint is enabled. 250 */ 251#define __DECLARE_TRACE(name, proto, args, cond, data_proto) \ 252 extern int __traceiter_##name(data_proto); \ 253 DECLARE_STATIC_CALL(tp_func_##name, __traceiter_##name); \ 254 extern struct tracepoint __tracepoint_##name; \ 255 static inline void trace_##name(proto) \ 256 { \ 257 if (static_key_false(&__tracepoint_##name.key)) \ 258 __DO_TRACE(name, \ 259 TP_ARGS(args), \ 260 TP_CONDITION(cond), 0); \ 261 if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) { \ 262 WARN_ONCE(!rcu_is_watching(), \ 263 "RCU not watching for tracepoint"); \ 264 } \ 265 } \ 266 __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args), \ 267 PARAMS(cond)) \ 268 static inline int \ 269 register_trace_##name(void (*probe)(data_proto), void *data) \ 270 { \ 271 return tracepoint_probe_register(&__tracepoint_##name, \ 272 (void *)probe, data); \ 273 } \ 274 static inline int \ 275 register_trace_prio_##name(void (*probe)(data_proto), void *data,\ 276 int prio) \ 277 { \ 278 return tracepoint_probe_register_prio(&__tracepoint_##name, \ 279 (void *)probe, data, prio); \ 280 } \ 281 static inline int \ 282 unregister_trace_##name(void (*probe)(data_proto), void *data) \ 283 { \ 284 return tracepoint_probe_unregister(&__tracepoint_##name,\ 285 (void *)probe, data); \ 286 } \ 287 static inline void \ 288 check_trace_callback_type_##name(void (*cb)(data_proto)) \ 289 { \ 290 } \ 291 static inline bool \ 292 trace_##name##_enabled(void) \ 293 { \ 294 return static_key_false(&__tracepoint_##name.key); \ 295 } 296 297/* 298 * We have no guarantee that gcc and the linker won't up-align the tracepoint 299 * structures, so we create an array of pointers that will be used for iteration 300 * on the tracepoints. 301 */ 302#define DEFINE_TRACE_FN(_name, _reg, _unreg, proto, args) \ 303 static const char __tpstrtab_##_name[] \ 304 __section("__tracepoints_strings") = #_name; \ 305 extern struct static_call_key STATIC_CALL_KEY(tp_func_##_name); \ 306 int __traceiter_##_name(void *__data, proto); \ 307 void __probestub_##_name(void *__data, proto); \ 308 struct tracepoint __tracepoint_##_name __used \ 309 __section("__tracepoints") = { \ 310 .name = __tpstrtab_##_name, \ 311 .key = STATIC_KEY_INIT_FALSE, \ 312 .static_call_key = &STATIC_CALL_KEY(tp_func_##_name), \ 313 .static_call_tramp = STATIC_CALL_TRAMP_ADDR(tp_func_##_name), \ 314 .iterator = &__traceiter_##_name, \ 315 .probestub = &__probestub_##_name, \ 316 .regfunc = _reg, \ 317 .unregfunc = _unreg, \ 318 .funcs = NULL }; \ 319 __TRACEPOINT_ENTRY(_name); \ 320 int __traceiter_##_name(void *__data, proto) \ 321 { \ 322 struct tracepoint_func *it_func_ptr; \ 323 void *it_func; \ 324 \ 325 it_func_ptr = \ 326 rcu_dereference_raw((&__tracepoint_##_name)->funcs); \ 327 if (it_func_ptr) { \ 328 do { \ 329 it_func = READ_ONCE((it_func_ptr)->func); \ 330 __data = (it_func_ptr)->data; \ 331 ((void(*)(void *, proto))(it_func))(__data, args); \ 332 } while ((++it_func_ptr)->func); \ 333 } \ 334 return 0; \ 335 } \ 336 void __probestub_##_name(void *__data, proto) \ 337 { \ 338 } \ 339 DEFINE_STATIC_CALL(tp_func_##_name, __traceiter_##_name); 340 341#define DEFINE_TRACE(name, proto, args) \ 342 DEFINE_TRACE_FN(name, NULL, NULL, PARAMS(proto), PARAMS(args)); 343 344#define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \ 345 EXPORT_SYMBOL_GPL(__tracepoint_##name); \ 346 EXPORT_SYMBOL_GPL(__traceiter_##name); \ 347 EXPORT_STATIC_CALL_GPL(tp_func_##name) 348#define EXPORT_TRACEPOINT_SYMBOL(name) \ 349 EXPORT_SYMBOL(__tracepoint_##name); \ 350 EXPORT_SYMBOL(__traceiter_##name); \ 351 EXPORT_STATIC_CALL(tp_func_##name) 352 353 354#else /* !TRACEPOINTS_ENABLED */ 355#define __DECLARE_TRACE(name, proto, args, cond, data_proto) \ 356 static inline void trace_##name(proto) \ 357 { } \ 358 static inline void trace_##name##_rcuidle(proto) \ 359 { } \ 360 static inline int \ 361 register_trace_##name(void (*probe)(data_proto), \ 362 void *data) \ 363 { \ 364 return -ENOSYS; \ 365 } \ 366 static inline int \ 367 unregister_trace_##name(void (*probe)(data_proto), \ 368 void *data) \ 369 { \ 370 return -ENOSYS; \ 371 } \ 372 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \ 373 { \ 374 } \ 375 static inline bool \ 376 trace_##name##_enabled(void) \ 377 { \ 378 return false; \ 379 } 380 381#define DEFINE_TRACE_FN(name, reg, unreg, proto, args) 382#define DEFINE_TRACE(name, proto, args) 383#define EXPORT_TRACEPOINT_SYMBOL_GPL(name) 384#define EXPORT_TRACEPOINT_SYMBOL(name) 385 386#endif /* TRACEPOINTS_ENABLED */ 387 388#ifdef CONFIG_TRACING 389/** 390 * tracepoint_string - register constant persistent string to trace system 391 * @str - a constant persistent string that will be referenced in tracepoints 392 * 393 * If constant strings are being used in tracepoints, it is faster and 394 * more efficient to just save the pointer to the string and reference 395 * that with a printf "%s" instead of saving the string in the ring buffer 396 * and wasting space and time. 397 * 398 * The problem with the above approach is that userspace tools that read 399 * the binary output of the trace buffers do not have access to the string. 400 * Instead they just show the address of the string which is not very 401 * useful to users. 402 * 403 * With tracepoint_string(), the string will be registered to the tracing 404 * system and exported to userspace via the debugfs/tracing/printk_formats 405 * file that maps the string address to the string text. This way userspace 406 * tools that read the binary buffers have a way to map the pointers to 407 * the ASCII strings they represent. 408 * 409 * The @str used must be a constant string and persistent as it would not 410 * make sense to show a string that no longer exists. But it is still fine 411 * to be used with modules, because when modules are unloaded, if they 412 * had tracepoints, the ring buffers are cleared too. As long as the string 413 * does not change during the life of the module, it is fine to use 414 * tracepoint_string() within a module. 415 */ 416#define tracepoint_string(str) \ 417 ({ \ 418 static const char *___tp_str __tracepoint_string = str; \ 419 ___tp_str; \ 420 }) 421#define __tracepoint_string __used __section("__tracepoint_str") 422#else 423/* 424 * tracepoint_string() is used to save the string address for userspace 425 * tracing tools. When tracing isn't configured, there's no need to save 426 * anything. 427 */ 428# define tracepoint_string(str) str 429# define __tracepoint_string 430#endif 431 432#define DECLARE_TRACE(name, proto, args) \ 433 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \ 434 cpu_online(raw_smp_processor_id()), \ 435 PARAMS(void *__data, proto)) 436 437#define DECLARE_TRACE_CONDITION(name, proto, args, cond) \ 438 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \ 439 cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \ 440 PARAMS(void *__data, proto)) 441 442#define TRACE_EVENT_FLAGS(event, flag) 443 444#define TRACE_EVENT_PERF_PERM(event, expr...) 445 446#endif /* DECLARE_TRACE */ 447 448#ifndef TRACE_EVENT 449/* 450 * For use with the TRACE_EVENT macro: 451 * 452 * We define a tracepoint, its arguments, its printk format 453 * and its 'fast binary record' layout. 454 * 455 * Firstly, name your tracepoint via TRACE_EVENT(name : the 456 * 'subsystem_event' notation is fine. 457 * 458 * Think about this whole construct as the 459 * 'trace_sched_switch() function' from now on. 460 * 461 * 462 * TRACE_EVENT(sched_switch, 463 * 464 * * 465 * * A function has a regular function arguments 466 * * prototype, declare it via TP_PROTO(): 467 * * 468 * 469 * TP_PROTO(struct rq *rq, struct task_struct *prev, 470 * struct task_struct *next), 471 * 472 * * 473 * * Define the call signature of the 'function'. 474 * * (Design sidenote: we use this instead of a 475 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.) 476 * * 477 * 478 * TP_ARGS(rq, prev, next), 479 * 480 * * 481 * * Fast binary tracing: define the trace record via 482 * * TP_STRUCT__entry(). You can think about it like a 483 * * regular C structure local variable definition. 484 * * 485 * * This is how the trace record is structured and will 486 * * be saved into the ring buffer. These are the fields 487 * * that will be exposed to user-space in 488 * * /sys/kernel/tracing/events/<*>/format. 489 * * 490 * * The declared 'local variable' is called '__entry' 491 * * 492 * * __field(pid_t, prev_pid) is equivalent to a standard declaration: 493 * * 494 * * pid_t prev_pid; 495 * * 496 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to: 497 * * 498 * * char prev_comm[TASK_COMM_LEN]; 499 * * 500 * 501 * TP_STRUCT__entry( 502 * __array( char, prev_comm, TASK_COMM_LEN ) 503 * __field( pid_t, prev_pid ) 504 * __field( int, prev_prio ) 505 * __array( char, next_comm, TASK_COMM_LEN ) 506 * __field( pid_t, next_pid ) 507 * __field( int, next_prio ) 508 * ), 509 * 510 * * 511 * * Assign the entry into the trace record, by embedding 512 * * a full C statement block into TP_fast_assign(). You 513 * * can refer to the trace record as '__entry' - 514 * * otherwise you can put arbitrary C code in here. 515 * * 516 * * Note: this C code will execute every time a trace event 517 * * happens, on an active tracepoint. 518 * * 519 * 520 * TP_fast_assign( 521 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN); 522 * __entry->prev_pid = prev->pid; 523 * __entry->prev_prio = prev->prio; 524 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN); 525 * __entry->next_pid = next->pid; 526 * __entry->next_prio = next->prio; 527 * ), 528 * 529 * * 530 * * Formatted output of a trace record via TP_printk(). 531 * * This is how the tracepoint will appear under ftrace 532 * * plugins that make use of this tracepoint. 533 * * 534 * * (raw-binary tracing wont actually perform this step.) 535 * * 536 * 537 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]", 538 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio, 539 * __entry->next_comm, __entry->next_pid, __entry->next_prio), 540 * 541 * ); 542 * 543 * This macro construct is thus used for the regular printk format 544 * tracing setup, it is used to construct a function pointer based 545 * tracepoint callback (this is used by programmatic plugins and 546 * can also by used by generic instrumentation like SystemTap), and 547 * it is also used to expose a structured trace record in 548 * /sys/kernel/tracing/events/. 549 * 550 * A set of (un)registration functions can be passed to the variant 551 * TRACE_EVENT_FN to perform any (un)registration work. 552 */ 553 554#define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print) 555#define DEFINE_EVENT(template, name, proto, args) \ 556 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 557#define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\ 558 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 559#define DEFINE_EVENT_PRINT(template, name, proto, args, print) \ 560 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 561#define DEFINE_EVENT_CONDITION(template, name, proto, \ 562 args, cond) \ 563 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 564 PARAMS(args), PARAMS(cond)) 565 566#define TRACE_EVENT(name, proto, args, struct, assign, print) \ 567 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 568#define TRACE_EVENT_FN(name, proto, args, struct, \ 569 assign, print, reg, unreg) \ 570 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 571#define TRACE_EVENT_FN_COND(name, proto, args, cond, struct, \ 572 assign, print, reg, unreg) \ 573 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 574 PARAMS(args), PARAMS(cond)) 575#define TRACE_EVENT_CONDITION(name, proto, args, cond, \ 576 struct, assign, print) \ 577 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 578 PARAMS(args), PARAMS(cond)) 579 580#define TRACE_EVENT_FLAGS(event, flag) 581 582#define TRACE_EVENT_PERF_PERM(event, expr...) 583 584#define DECLARE_EVENT_NOP(name, proto, args) \ 585 static inline void trace_##name(proto) \ 586 { } \ 587 static inline bool trace_##name##_enabled(void) \ 588 { \ 589 return false; \ 590 } 591 592#define TRACE_EVENT_NOP(name, proto, args, struct, assign, print) \ 593 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args)) 594 595#define DECLARE_EVENT_CLASS_NOP(name, proto, args, tstruct, assign, print) 596#define DEFINE_EVENT_NOP(template, name, proto, args) \ 597 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args)) 598 599#endif /* ifdef TRACE_EVENT (see note above) */