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 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_TRACEPOINT 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 * it_func[0] is never NULL because there is at least one element in the array 182 * when the array itself is non NULL. 183 */ 184 #define __DO_TRACE(name, args, cond, rcuidle) \ 185 do { \ 186 int __maybe_unused __idx = 0; \ 187 \ 188 if (!(cond)) \ 189 return; \ 190 \ 191 /* srcu can't be used from NMI */ \ 192 WARN_ON_ONCE(rcuidle && in_nmi()); \ 193 \ 194 /* keep srcu and sched-rcu usage consistent */ \ 195 preempt_disable_notrace(); \ 196 \ 197 /* \ 198 * For rcuidle callers, use srcu since sched-rcu \ 199 * doesn't work from the idle path. \ 200 */ \ 201 if (rcuidle) { \ 202 __idx = srcu_read_lock_notrace(&tracepoint_srcu);\ 203 rcu_irq_enter_irqson(); \ 204 } \ 205 \ 206 __DO_TRACE_CALL(name, TP_ARGS(args)); \ 207 \ 208 if (rcuidle) { \ 209 rcu_irq_exit_irqson(); \ 210 srcu_read_unlock_notrace(&tracepoint_srcu, __idx);\ 211 } \ 212 \ 213 preempt_enable_notrace(); \ 214 } while (0) 215 216 #ifndef MODULE 217 #define __DECLARE_TRACE_RCU(name, proto, args, cond) \ 218 static inline void trace_##name##_rcuidle(proto) \ 219 { \ 220 if (static_key_false(&__tracepoint_##name.key)) \ 221 __DO_TRACE(name, \ 222 TP_ARGS(args), \ 223 TP_CONDITION(cond), 1); \ 224 } 225 #else 226 #define __DECLARE_TRACE_RCU(name, proto, args, cond) 227 #endif 228 229 /* 230 * Make sure the alignment of the structure in the __tracepoints section will 231 * not add unwanted padding between the beginning of the section and the 232 * structure. Force alignment to the same alignment as the section start. 233 * 234 * When lockdep is enabled, we make sure to always do the RCU portions of 235 * the tracepoint code, regardless of whether tracing is on. However, 236 * don't check if the condition is false, due to interaction with idle 237 * instrumentation. This lets us find RCU issues triggered with tracepoints 238 * even when this tracepoint is off. This code has no purpose other than 239 * poking RCU a bit. 240 */ 241 #define __DECLARE_TRACE(name, proto, args, cond, data_proto) \ 242 extern int __traceiter_##name(data_proto); \ 243 DECLARE_STATIC_CALL(tp_func_##name, __traceiter_##name); \ 244 extern struct tracepoint __tracepoint_##name; \ 245 static inline void trace_##name(proto) \ 246 { \ 247 if (static_key_false(&__tracepoint_##name.key)) \ 248 __DO_TRACE(name, \ 249 TP_ARGS(args), \ 250 TP_CONDITION(cond), 0); \ 251 if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) { \ 252 rcu_read_lock_sched_notrace(); \ 253 rcu_dereference_sched(__tracepoint_##name.funcs);\ 254 rcu_read_unlock_sched_notrace(); \ 255 } \ 256 } \ 257 __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args), \ 258 PARAMS(cond)) \ 259 static inline int \ 260 register_trace_##name(void (*probe)(data_proto), void *data) \ 261 { \ 262 return tracepoint_probe_register(&__tracepoint_##name, \ 263 (void *)probe, data); \ 264 } \ 265 static inline int \ 266 register_trace_prio_##name(void (*probe)(data_proto), void *data,\ 267 int prio) \ 268 { \ 269 return tracepoint_probe_register_prio(&__tracepoint_##name, \ 270 (void *)probe, data, prio); \ 271 } \ 272 static inline int \ 273 unregister_trace_##name(void (*probe)(data_proto), void *data) \ 274 { \ 275 return tracepoint_probe_unregister(&__tracepoint_##name,\ 276 (void *)probe, data); \ 277 } \ 278 static inline void \ 279 check_trace_callback_type_##name(void (*cb)(data_proto)) \ 280 { \ 281 } \ 282 static inline bool \ 283 trace_##name##_enabled(void) \ 284 { \ 285 return static_key_false(&__tracepoint_##name.key); \ 286 } 287 288 /* 289 * We have no guarantee that gcc and the linker won't up-align the tracepoint 290 * structures, so we create an array of pointers that will be used for iteration 291 * on the tracepoints. 292 */ 293 #define DEFINE_TRACE_FN(_name, _reg, _unreg, proto, args) \ 294 static const char __tpstrtab_##_name[] \ 295 __section("__tracepoints_strings") = #_name; \ 296 extern struct static_call_key STATIC_CALL_KEY(tp_func_##_name); \ 297 int __traceiter_##_name(void *__data, proto); \ 298 struct tracepoint __tracepoint_##_name __used \ 299 __section("__tracepoints") = { \ 300 .name = __tpstrtab_##_name, \ 301 .key = STATIC_KEY_INIT_FALSE, \ 302 .static_call_key = &STATIC_CALL_KEY(tp_func_##_name), \ 303 .static_call_tramp = STATIC_CALL_TRAMP_ADDR(tp_func_##_name), \ 304 .iterator = &__traceiter_##_name, \ 305 .regfunc = _reg, \ 306 .unregfunc = _unreg, \ 307 .funcs = NULL }; \ 308 __TRACEPOINT_ENTRY(_name); \ 309 int __traceiter_##_name(void *__data, proto) \ 310 { \ 311 struct tracepoint_func *it_func_ptr; \ 312 void *it_func; \ 313 \ 314 it_func_ptr = \ 315 rcu_dereference_raw((&__tracepoint_##_name)->funcs); \ 316 if (it_func_ptr) { \ 317 do { \ 318 it_func = READ_ONCE((it_func_ptr)->func); \ 319 __data = (it_func_ptr)->data; \ 320 ((void(*)(void *, proto))(it_func))(__data, args); \ 321 } while ((++it_func_ptr)->func); \ 322 } \ 323 return 0; \ 324 } \ 325 DEFINE_STATIC_CALL(tp_func_##_name, __traceiter_##_name); 326 327 #define DEFINE_TRACE(name, proto, args) \ 328 DEFINE_TRACE_FN(name, NULL, NULL, PARAMS(proto), PARAMS(args)); 329 330 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \ 331 EXPORT_SYMBOL_GPL(__tracepoint_##name); \ 332 EXPORT_SYMBOL_GPL(__traceiter_##name); \ 333 EXPORT_STATIC_CALL_GPL(tp_func_##name) 334 #define EXPORT_TRACEPOINT_SYMBOL(name) \ 335 EXPORT_SYMBOL(__tracepoint_##name); \ 336 EXPORT_SYMBOL(__traceiter_##name); \ 337 EXPORT_STATIC_CALL(tp_func_##name) 338 339 340 #else /* !TRACEPOINTS_ENABLED */ 341 #define __DECLARE_TRACE(name, proto, args, cond, data_proto) \ 342 static inline void trace_##name(proto) \ 343 { } \ 344 static inline void trace_##name##_rcuidle(proto) \ 345 { } \ 346 static inline int \ 347 register_trace_##name(void (*probe)(data_proto), \ 348 void *data) \ 349 { \ 350 return -ENOSYS; \ 351 } \ 352 static inline int \ 353 unregister_trace_##name(void (*probe)(data_proto), \ 354 void *data) \ 355 { \ 356 return -ENOSYS; \ 357 } \ 358 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \ 359 { \ 360 } \ 361 static inline bool \ 362 trace_##name##_enabled(void) \ 363 { \ 364 return false; \ 365 } 366 367 #define DEFINE_TRACE_FN(name, reg, unreg, proto, args) 368 #define DEFINE_TRACE(name, proto, args) 369 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) 370 #define EXPORT_TRACEPOINT_SYMBOL(name) 371 372 #endif /* TRACEPOINTS_ENABLED */ 373 374 #ifdef CONFIG_TRACING 375 /** 376 * tracepoint_string - register constant persistent string to trace system 377 * @str - a constant persistent string that will be referenced in tracepoints 378 * 379 * If constant strings are being used in tracepoints, it is faster and 380 * more efficient to just save the pointer to the string and reference 381 * that with a printf "%s" instead of saving the string in the ring buffer 382 * and wasting space and time. 383 * 384 * The problem with the above approach is that userspace tools that read 385 * the binary output of the trace buffers do not have access to the string. 386 * Instead they just show the address of the string which is not very 387 * useful to users. 388 * 389 * With tracepoint_string(), the string will be registered to the tracing 390 * system and exported to userspace via the debugfs/tracing/printk_formats 391 * file that maps the string address to the string text. This way userspace 392 * tools that read the binary buffers have a way to map the pointers to 393 * the ASCII strings they represent. 394 * 395 * The @str used must be a constant string and persistent as it would not 396 * make sense to show a string that no longer exists. But it is still fine 397 * to be used with modules, because when modules are unloaded, if they 398 * had tracepoints, the ring buffers are cleared too. As long as the string 399 * does not change during the life of the module, it is fine to use 400 * tracepoint_string() within a module. 401 */ 402 #define tracepoint_string(str) \ 403 ({ \ 404 static const char *___tp_str __tracepoint_string = str; \ 405 ___tp_str; \ 406 }) 407 #define __tracepoint_string __used __section("__tracepoint_str") 408 #else 409 /* 410 * tracepoint_string() is used to save the string address for userspace 411 * tracing tools. When tracing isn't configured, there's no need to save 412 * anything. 413 */ 414 # define tracepoint_string(str) str 415 # define __tracepoint_string 416 #endif 417 418 #define DECLARE_TRACE(name, proto, args) \ 419 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \ 420 cpu_online(raw_smp_processor_id()), \ 421 PARAMS(void *__data, proto)) 422 423 #define DECLARE_TRACE_CONDITION(name, proto, args, cond) \ 424 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \ 425 cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \ 426 PARAMS(void *__data, proto)) 427 428 #define TRACE_EVENT_FLAGS(event, flag) 429 430 #define TRACE_EVENT_PERF_PERM(event, expr...) 431 432 #endif /* DECLARE_TRACE */ 433 434 #ifndef TRACE_EVENT 435 /* 436 * For use with the TRACE_EVENT macro: 437 * 438 * We define a tracepoint, its arguments, its printk format 439 * and its 'fast binary record' layout. 440 * 441 * Firstly, name your tracepoint via TRACE_EVENT(name : the 442 * 'subsystem_event' notation is fine. 443 * 444 * Think about this whole construct as the 445 * 'trace_sched_switch() function' from now on. 446 * 447 * 448 * TRACE_EVENT(sched_switch, 449 * 450 * * 451 * * A function has a regular function arguments 452 * * prototype, declare it via TP_PROTO(): 453 * * 454 * 455 * TP_PROTO(struct rq *rq, struct task_struct *prev, 456 * struct task_struct *next), 457 * 458 * * 459 * * Define the call signature of the 'function'. 460 * * (Design sidenote: we use this instead of a 461 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.) 462 * * 463 * 464 * TP_ARGS(rq, prev, next), 465 * 466 * * 467 * * Fast binary tracing: define the trace record via 468 * * TP_STRUCT__entry(). You can think about it like a 469 * * regular C structure local variable definition. 470 * * 471 * * This is how the trace record is structured and will 472 * * be saved into the ring buffer. These are the fields 473 * * that will be exposed to user-space in 474 * * /sys/kernel/debug/tracing/events/<*>/format. 475 * * 476 * * The declared 'local variable' is called '__entry' 477 * * 478 * * __field(pid_t, prev_pid) is equivalent to a standard declaration: 479 * * 480 * * pid_t prev_pid; 481 * * 482 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to: 483 * * 484 * * char prev_comm[TASK_COMM_LEN]; 485 * * 486 * 487 * TP_STRUCT__entry( 488 * __array( char, prev_comm, TASK_COMM_LEN ) 489 * __field( pid_t, prev_pid ) 490 * __field( int, prev_prio ) 491 * __array( char, next_comm, TASK_COMM_LEN ) 492 * __field( pid_t, next_pid ) 493 * __field( int, next_prio ) 494 * ), 495 * 496 * * 497 * * Assign the entry into the trace record, by embedding 498 * * a full C statement block into TP_fast_assign(). You 499 * * can refer to the trace record as '__entry' - 500 * * otherwise you can put arbitrary C code in here. 501 * * 502 * * Note: this C code will execute every time a trace event 503 * * happens, on an active tracepoint. 504 * * 505 * 506 * TP_fast_assign( 507 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN); 508 * __entry->prev_pid = prev->pid; 509 * __entry->prev_prio = prev->prio; 510 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN); 511 * __entry->next_pid = next->pid; 512 * __entry->next_prio = next->prio; 513 * ), 514 * 515 * * 516 * * Formatted output of a trace record via TP_printk(). 517 * * This is how the tracepoint will appear under ftrace 518 * * plugins that make use of this tracepoint. 519 * * 520 * * (raw-binary tracing wont actually perform this step.) 521 * * 522 * 523 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]", 524 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio, 525 * __entry->next_comm, __entry->next_pid, __entry->next_prio), 526 * 527 * ); 528 * 529 * This macro construct is thus used for the regular printk format 530 * tracing setup, it is used to construct a function pointer based 531 * tracepoint callback (this is used by programmatic plugins and 532 * can also by used by generic instrumentation like SystemTap), and 533 * it is also used to expose a structured trace record in 534 * /sys/kernel/debug/tracing/events/. 535 * 536 * A set of (un)registration functions can be passed to the variant 537 * TRACE_EVENT_FN to perform any (un)registration work. 538 */ 539 540 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print) 541 #define DEFINE_EVENT(template, name, proto, args) \ 542 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 543 #define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\ 544 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 545 #define DEFINE_EVENT_PRINT(template, name, proto, args, print) \ 546 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 547 #define DEFINE_EVENT_CONDITION(template, name, proto, \ 548 args, cond) \ 549 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 550 PARAMS(args), PARAMS(cond)) 551 552 #define TRACE_EVENT(name, proto, args, struct, assign, print) \ 553 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 554 #define TRACE_EVENT_FN(name, proto, args, struct, \ 555 assign, print, reg, unreg) \ 556 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 557 #define TRACE_EVENT_FN_COND(name, proto, args, cond, struct, \ 558 assign, print, reg, unreg) \ 559 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 560 PARAMS(args), PARAMS(cond)) 561 #define TRACE_EVENT_CONDITION(name, proto, args, cond, \ 562 struct, assign, print) \ 563 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 564 PARAMS(args), PARAMS(cond)) 565 566 #define TRACE_EVENT_FLAGS(event, flag) 567 568 #define TRACE_EVENT_PERF_PERM(event, expr...) 569 570 #define DECLARE_EVENT_NOP(name, proto, args) \ 571 static inline void trace_##name(proto) \ 572 { } \ 573 static inline bool trace_##name##_enabled(void) \ 574 { \ 575 return false; \ 576 } 577 578 #define TRACE_EVENT_NOP(name, proto, args, struct, assign, print) \ 579 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args)) 580 581 #define DECLARE_EVENT_CLASS_NOP(name, proto, args, tstruct, assign, print) 582 #define DEFINE_EVENT_NOP(template, name, proto, args) \ 583 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args)) 584 585 #endif /* ifdef TRACE_EVENT (see note above) */ 586