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 <[email protected]> 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 24 struct module; 25 struct tracepoint; 26 struct notifier_block; 27 28 struct 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 36 extern struct srcu_struct tracepoint_srcu; 37 38 extern int 39 tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data); 40 extern int 41 tracepoint_probe_register_prio(struct tracepoint *tp, void *probe, void *data, 42 int prio); 43 extern int 44 tracepoint_probe_register_prio_may_exist(struct tracepoint *tp, void *probe, void *data, 45 int prio); 46 extern int 47 tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data); 48 static inline int 49 tracepoint_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 } 55 extern void 56 for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv), 57 void *priv); 58 59 #ifdef CONFIG_MODULES 60 struct tp_module { 61 struct list_head list; 62 struct module *mod; 63 }; 64 65 bool trace_module_has_bad_taint(struct module *mod); 66 extern int register_tracepoint_module_notifier(struct notifier_block *nb); 67 extern int unregister_tracepoint_module_notifier(struct notifier_block *nb); 68 #else 69 static inline bool trace_module_has_bad_taint(struct module *mod) 70 { 71 return false; 72 } 73 static inline 74 int register_tracepoint_module_notifier(struct notifier_block *nb) 75 { 76 return 0; 77 } 78 static inline 79 int 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 91 static inline void tracepoint_synchronize_unregister(void) 92 { 93 synchronize_srcu(&tracepoint_srcu); 94 synchronize_rcu(); 95 } 96 #else 97 static inline void tracepoint_synchronize_unregister(void) 98 { } 99 #endif 100 101 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS 102 extern int syscall_regfunc(void); 103 extern 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 114 static 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 125 static 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_ON_ONCE(RCUIDLE_COND(rcuidle))) \ 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_ON_ONCE(!rcu_is_watching()); \ 263 } \ 264 } \ 265 __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args), \ 266 PARAMS(cond)) \ 267 static inline int \ 268 register_trace_##name(void (*probe)(data_proto), void *data) \ 269 { \ 270 return tracepoint_probe_register(&__tracepoint_##name, \ 271 (void *)probe, data); \ 272 } \ 273 static inline int \ 274 register_trace_prio_##name(void (*probe)(data_proto), void *data,\ 275 int prio) \ 276 { \ 277 return tracepoint_probe_register_prio(&__tracepoint_##name, \ 278 (void *)probe, data, prio); \ 279 } \ 280 static inline int \ 281 unregister_trace_##name(void (*probe)(data_proto), void *data) \ 282 { \ 283 return tracepoint_probe_unregister(&__tracepoint_##name,\ 284 (void *)probe, data); \ 285 } \ 286 static inline void \ 287 check_trace_callback_type_##name(void (*cb)(data_proto)) \ 288 { \ 289 } \ 290 static inline bool \ 291 trace_##name##_enabled(void) \ 292 { \ 293 return static_key_false(&__tracepoint_##name.key); \ 294 } 295 296 /* 297 * We have no guarantee that gcc and the linker won't up-align the tracepoint 298 * structures, so we create an array of pointers that will be used for iteration 299 * on the tracepoints. 300 */ 301 #define DEFINE_TRACE_FN(_name, _reg, _unreg, proto, args) \ 302 static const char __tpstrtab_##_name[] \ 303 __section("__tracepoints_strings") = #_name; \ 304 extern struct static_call_key STATIC_CALL_KEY(tp_func_##_name); \ 305 int __traceiter_##_name(void *__data, proto); \ 306 struct tracepoint __tracepoint_##_name __used \ 307 __section("__tracepoints") = { \ 308 .name = __tpstrtab_##_name, \ 309 .key = STATIC_KEY_INIT_FALSE, \ 310 .static_call_key = &STATIC_CALL_KEY(tp_func_##_name), \ 311 .static_call_tramp = STATIC_CALL_TRAMP_ADDR(tp_func_##_name), \ 312 .iterator = &__traceiter_##_name, \ 313 .regfunc = _reg, \ 314 .unregfunc = _unreg, \ 315 .funcs = NULL }; \ 316 __TRACEPOINT_ENTRY(_name); \ 317 int __traceiter_##_name(void *__data, proto) \ 318 { \ 319 struct tracepoint_func *it_func_ptr; \ 320 void *it_func; \ 321 \ 322 it_func_ptr = \ 323 rcu_dereference_raw((&__tracepoint_##_name)->funcs); \ 324 if (it_func_ptr) { \ 325 do { \ 326 it_func = READ_ONCE((it_func_ptr)->func); \ 327 __data = (it_func_ptr)->data; \ 328 ((void(*)(void *, proto))(it_func))(__data, args); \ 329 } while ((++it_func_ptr)->func); \ 330 } \ 331 return 0; \ 332 } \ 333 DEFINE_STATIC_CALL(tp_func_##_name, __traceiter_##_name); 334 335 #define DEFINE_TRACE(name, proto, args) \ 336 DEFINE_TRACE_FN(name, NULL, NULL, PARAMS(proto), PARAMS(args)); 337 338 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \ 339 EXPORT_SYMBOL_GPL(__tracepoint_##name); \ 340 EXPORT_SYMBOL_GPL(__traceiter_##name); \ 341 EXPORT_STATIC_CALL_GPL(tp_func_##name) 342 #define EXPORT_TRACEPOINT_SYMBOL(name) \ 343 EXPORT_SYMBOL(__tracepoint_##name); \ 344 EXPORT_SYMBOL(__traceiter_##name); \ 345 EXPORT_STATIC_CALL(tp_func_##name) 346 347 348 #else /* !TRACEPOINTS_ENABLED */ 349 #define __DECLARE_TRACE(name, proto, args, cond, data_proto) \ 350 static inline void trace_##name(proto) \ 351 { } \ 352 static inline void trace_##name##_rcuidle(proto) \ 353 { } \ 354 static inline int \ 355 register_trace_##name(void (*probe)(data_proto), \ 356 void *data) \ 357 { \ 358 return -ENOSYS; \ 359 } \ 360 static inline int \ 361 unregister_trace_##name(void (*probe)(data_proto), \ 362 void *data) \ 363 { \ 364 return -ENOSYS; \ 365 } \ 366 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \ 367 { \ 368 } \ 369 static inline bool \ 370 trace_##name##_enabled(void) \ 371 { \ 372 return false; \ 373 } 374 375 #define DEFINE_TRACE_FN(name, reg, unreg, proto, args) 376 #define DEFINE_TRACE(name, proto, args) 377 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) 378 #define EXPORT_TRACEPOINT_SYMBOL(name) 379 380 #endif /* TRACEPOINTS_ENABLED */ 381 382 #ifdef CONFIG_TRACING 383 /** 384 * tracepoint_string - register constant persistent string to trace system 385 * @str - a constant persistent string that will be referenced in tracepoints 386 * 387 * If constant strings are being used in tracepoints, it is faster and 388 * more efficient to just save the pointer to the string and reference 389 * that with a printf "%s" instead of saving the string in the ring buffer 390 * and wasting space and time. 391 * 392 * The problem with the above approach is that userspace tools that read 393 * the binary output of the trace buffers do not have access to the string. 394 * Instead they just show the address of the string which is not very 395 * useful to users. 396 * 397 * With tracepoint_string(), the string will be registered to the tracing 398 * system and exported to userspace via the debugfs/tracing/printk_formats 399 * file that maps the string address to the string text. This way userspace 400 * tools that read the binary buffers have a way to map the pointers to 401 * the ASCII strings they represent. 402 * 403 * The @str used must be a constant string and persistent as it would not 404 * make sense to show a string that no longer exists. But it is still fine 405 * to be used with modules, because when modules are unloaded, if they 406 * had tracepoints, the ring buffers are cleared too. As long as the string 407 * does not change during the life of the module, it is fine to use 408 * tracepoint_string() within a module. 409 */ 410 #define tracepoint_string(str) \ 411 ({ \ 412 static const char *___tp_str __tracepoint_string = str; \ 413 ___tp_str; \ 414 }) 415 #define __tracepoint_string __used __section("__tracepoint_str") 416 #else 417 /* 418 * tracepoint_string() is used to save the string address for userspace 419 * tracing tools. When tracing isn't configured, there's no need to save 420 * anything. 421 */ 422 # define tracepoint_string(str) str 423 # define __tracepoint_string 424 #endif 425 426 #define DECLARE_TRACE(name, proto, args) \ 427 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \ 428 cpu_online(raw_smp_processor_id()), \ 429 PARAMS(void *__data, proto)) 430 431 #define DECLARE_TRACE_CONDITION(name, proto, args, cond) \ 432 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \ 433 cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \ 434 PARAMS(void *__data, proto)) 435 436 #define TRACE_EVENT_FLAGS(event, flag) 437 438 #define TRACE_EVENT_PERF_PERM(event, expr...) 439 440 #endif /* DECLARE_TRACE */ 441 442 #ifndef TRACE_EVENT 443 /* 444 * For use with the TRACE_EVENT macro: 445 * 446 * We define a tracepoint, its arguments, its printk format 447 * and its 'fast binary record' layout. 448 * 449 * Firstly, name your tracepoint via TRACE_EVENT(name : the 450 * 'subsystem_event' notation is fine. 451 * 452 * Think about this whole construct as the 453 * 'trace_sched_switch() function' from now on. 454 * 455 * 456 * TRACE_EVENT(sched_switch, 457 * 458 * * 459 * * A function has a regular function arguments 460 * * prototype, declare it via TP_PROTO(): 461 * * 462 * 463 * TP_PROTO(struct rq *rq, struct task_struct *prev, 464 * struct task_struct *next), 465 * 466 * * 467 * * Define the call signature of the 'function'. 468 * * (Design sidenote: we use this instead of a 469 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.) 470 * * 471 * 472 * TP_ARGS(rq, prev, next), 473 * 474 * * 475 * * Fast binary tracing: define the trace record via 476 * * TP_STRUCT__entry(). You can think about it like a 477 * * regular C structure local variable definition. 478 * * 479 * * This is how the trace record is structured and will 480 * * be saved into the ring buffer. These are the fields 481 * * that will be exposed to user-space in 482 * * /sys/kernel/tracing/events/<*>/format. 483 * * 484 * * The declared 'local variable' is called '__entry' 485 * * 486 * * __field(pid_t, prev_pid) is equivalent to a standard declaration: 487 * * 488 * * pid_t prev_pid; 489 * * 490 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to: 491 * * 492 * * char prev_comm[TASK_COMM_LEN]; 493 * * 494 * 495 * TP_STRUCT__entry( 496 * __array( char, prev_comm, TASK_COMM_LEN ) 497 * __field( pid_t, prev_pid ) 498 * __field( int, prev_prio ) 499 * __array( char, next_comm, TASK_COMM_LEN ) 500 * __field( pid_t, next_pid ) 501 * __field( int, next_prio ) 502 * ), 503 * 504 * * 505 * * Assign the entry into the trace record, by embedding 506 * * a full C statement block into TP_fast_assign(). You 507 * * can refer to the trace record as '__entry' - 508 * * otherwise you can put arbitrary C code in here. 509 * * 510 * * Note: this C code will execute every time a trace event 511 * * happens, on an active tracepoint. 512 * * 513 * 514 * TP_fast_assign( 515 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN); 516 * __entry->prev_pid = prev->pid; 517 * __entry->prev_prio = prev->prio; 518 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN); 519 * __entry->next_pid = next->pid; 520 * __entry->next_prio = next->prio; 521 * ), 522 * 523 * * 524 * * Formatted output of a trace record via TP_printk(). 525 * * This is how the tracepoint will appear under ftrace 526 * * plugins that make use of this tracepoint. 527 * * 528 * * (raw-binary tracing wont actually perform this step.) 529 * * 530 * 531 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]", 532 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio, 533 * __entry->next_comm, __entry->next_pid, __entry->next_prio), 534 * 535 * ); 536 * 537 * This macro construct is thus used for the regular printk format 538 * tracing setup, it is used to construct a function pointer based 539 * tracepoint callback (this is used by programmatic plugins and 540 * can also by used by generic instrumentation like SystemTap), and 541 * it is also used to expose a structured trace record in 542 * /sys/kernel/tracing/events/. 543 * 544 * A set of (un)registration functions can be passed to the variant 545 * TRACE_EVENT_FN to perform any (un)registration work. 546 */ 547 548 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print) 549 #define DEFINE_EVENT(template, name, proto, args) \ 550 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 551 #define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\ 552 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 553 #define DEFINE_EVENT_PRINT(template, name, proto, args, print) \ 554 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 555 #define DEFINE_EVENT_CONDITION(template, name, proto, \ 556 args, cond) \ 557 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 558 PARAMS(args), PARAMS(cond)) 559 560 #define TRACE_EVENT(name, proto, args, struct, assign, print) \ 561 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 562 #define TRACE_EVENT_FN(name, proto, args, struct, \ 563 assign, print, reg, unreg) \ 564 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 565 #define TRACE_EVENT_FN_COND(name, proto, args, cond, struct, \ 566 assign, print, reg, unreg) \ 567 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 568 PARAMS(args), PARAMS(cond)) 569 #define TRACE_EVENT_CONDITION(name, proto, args, cond, \ 570 struct, assign, print) \ 571 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 572 PARAMS(args), PARAMS(cond)) 573 574 #define TRACE_EVENT_FLAGS(event, flag) 575 576 #define TRACE_EVENT_PERF_PERM(event, expr...) 577 578 #define DECLARE_EVENT_NOP(name, proto, args) \ 579 static inline void trace_##name(proto) \ 580 { } \ 581 static inline bool trace_##name##_enabled(void) \ 582 { \ 583 return false; \ 584 } 585 586 #define TRACE_EVENT_NOP(name, proto, args, struct, assign, print) \ 587 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args)) 588 589 #define DECLARE_EVENT_CLASS_NOP(name, proto, args, tstruct, assign, print) 590 #define DEFINE_EVENT_NOP(template, name, proto, args) \ 591 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args)) 592 593 #endif /* ifdef TRACE_EVENT (see note above) */ 594