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_unregister(struct tracepoint *tp, void *probe, void *data); 45 extern void 46 for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv), 47 void *priv); 48 49 #ifdef CONFIG_MODULES 50 struct tp_module { 51 struct list_head list; 52 struct module *mod; 53 }; 54 55 bool trace_module_has_bad_taint(struct module *mod); 56 extern int register_tracepoint_module_notifier(struct notifier_block *nb); 57 extern int unregister_tracepoint_module_notifier(struct notifier_block *nb); 58 #else 59 static inline bool trace_module_has_bad_taint(struct module *mod) 60 { 61 return false; 62 } 63 static inline 64 int register_tracepoint_module_notifier(struct notifier_block *nb) 65 { 66 return 0; 67 } 68 static inline 69 int 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 81 static inline void tracepoint_synchronize_unregister(void) 82 { 83 synchronize_srcu(&tracepoint_srcu); 84 synchronize_rcu(); 85 } 86 #else 87 static inline void tracepoint_synchronize_unregister(void) 88 { } 89 #endif 90 91 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS 92 extern int syscall_regfunc(void); 93 extern 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 104 static 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 115 static 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 do { \ 311 it_func = (it_func_ptr)->func; \ 312 __data = (it_func_ptr)->data; \ 313 ((void(*)(void *, proto))(it_func))(__data, args); \ 314 } while ((++it_func_ptr)->func); \ 315 return 0; \ 316 } \ 317 DEFINE_STATIC_CALL(tp_func_##_name, __traceiter_##_name); 318 319 #define DEFINE_TRACE(name, proto, args) \ 320 DEFINE_TRACE_FN(name, NULL, NULL, PARAMS(proto), PARAMS(args)); 321 322 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \ 323 EXPORT_SYMBOL_GPL(__tracepoint_##name); \ 324 EXPORT_SYMBOL_GPL(__traceiter_##name); \ 325 EXPORT_STATIC_CALL_GPL(tp_func_##name) 326 #define EXPORT_TRACEPOINT_SYMBOL(name) \ 327 EXPORT_SYMBOL(__tracepoint_##name); \ 328 EXPORT_SYMBOL(__traceiter_##name); \ 329 EXPORT_STATIC_CALL(tp_func_##name) 330 331 332 #else /* !TRACEPOINTS_ENABLED */ 333 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \ 334 static inline void trace_##name(proto) \ 335 { } \ 336 static inline void trace_##name##_rcuidle(proto) \ 337 { } \ 338 static inline int \ 339 register_trace_##name(void (*probe)(data_proto), \ 340 void *data) \ 341 { \ 342 return -ENOSYS; \ 343 } \ 344 static inline int \ 345 unregister_trace_##name(void (*probe)(data_proto), \ 346 void *data) \ 347 { \ 348 return -ENOSYS; \ 349 } \ 350 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \ 351 { \ 352 } \ 353 static inline bool \ 354 trace_##name##_enabled(void) \ 355 { \ 356 return false; \ 357 } 358 359 #define DEFINE_TRACE_FN(name, reg, unreg, proto, args) 360 #define DEFINE_TRACE(name, proto, args) 361 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) 362 #define EXPORT_TRACEPOINT_SYMBOL(name) 363 364 #endif /* TRACEPOINTS_ENABLED */ 365 366 #ifdef CONFIG_TRACING 367 /** 368 * tracepoint_string - register constant persistent string to trace system 369 * @str - a constant persistent string that will be referenced in tracepoints 370 * 371 * If constant strings are being used in tracepoints, it is faster and 372 * more efficient to just save the pointer to the string and reference 373 * that with a printf "%s" instead of saving the string in the ring buffer 374 * and wasting space and time. 375 * 376 * The problem with the above approach is that userspace tools that read 377 * the binary output of the trace buffers do not have access to the string. 378 * Instead they just show the address of the string which is not very 379 * useful to users. 380 * 381 * With tracepoint_string(), the string will be registered to the tracing 382 * system and exported to userspace via the debugfs/tracing/printk_formats 383 * file that maps the string address to the string text. This way userspace 384 * tools that read the binary buffers have a way to map the pointers to 385 * the ASCII strings they represent. 386 * 387 * The @str used must be a constant string and persistent as it would not 388 * make sense to show a string that no longer exists. But it is still fine 389 * to be used with modules, because when modules are unloaded, if they 390 * had tracepoints, the ring buffers are cleared too. As long as the string 391 * does not change during the life of the module, it is fine to use 392 * tracepoint_string() within a module. 393 */ 394 #define tracepoint_string(str) \ 395 ({ \ 396 static const char *___tp_str __tracepoint_string = str; \ 397 ___tp_str; \ 398 }) 399 #define __tracepoint_string __used __section(__tracepoint_str) 400 #else 401 /* 402 * tracepoint_string() is used to save the string address for userspace 403 * tracing tools. When tracing isn't configured, there's no need to save 404 * anything. 405 */ 406 # define tracepoint_string(str) str 407 # define __tracepoint_string 408 #endif 409 410 #define DECLARE_TRACE(name, proto, args) \ 411 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \ 412 cpu_online(raw_smp_processor_id()), \ 413 PARAMS(void *__data, proto), \ 414 PARAMS(__data, args)) 415 416 #define DECLARE_TRACE_CONDITION(name, proto, args, cond) \ 417 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \ 418 cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \ 419 PARAMS(void *__data, proto), \ 420 PARAMS(__data, args)) 421 422 #define TRACE_EVENT_FLAGS(event, flag) 423 424 #define TRACE_EVENT_PERF_PERM(event, expr...) 425 426 #endif /* DECLARE_TRACE */ 427 428 #ifndef TRACE_EVENT 429 /* 430 * For use with the TRACE_EVENT macro: 431 * 432 * We define a tracepoint, its arguments, its printk format 433 * and its 'fast binary record' layout. 434 * 435 * Firstly, name your tracepoint via TRACE_EVENT(name : the 436 * 'subsystem_event' notation is fine. 437 * 438 * Think about this whole construct as the 439 * 'trace_sched_switch() function' from now on. 440 * 441 * 442 * TRACE_EVENT(sched_switch, 443 * 444 * * 445 * * A function has a regular function arguments 446 * * prototype, declare it via TP_PROTO(): 447 * * 448 * 449 * TP_PROTO(struct rq *rq, struct task_struct *prev, 450 * struct task_struct *next), 451 * 452 * * 453 * * Define the call signature of the 'function'. 454 * * (Design sidenote: we use this instead of a 455 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.) 456 * * 457 * 458 * TP_ARGS(rq, prev, next), 459 * 460 * * 461 * * Fast binary tracing: define the trace record via 462 * * TP_STRUCT__entry(). You can think about it like a 463 * * regular C structure local variable definition. 464 * * 465 * * This is how the trace record is structured and will 466 * * be saved into the ring buffer. These are the fields 467 * * that will be exposed to user-space in 468 * * /sys/kernel/debug/tracing/events/<*>/format. 469 * * 470 * * The declared 'local variable' is called '__entry' 471 * * 472 * * __field(pid_t, prev_prid) is equivalent to a standard declariton: 473 * * 474 * * pid_t prev_pid; 475 * * 476 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to: 477 * * 478 * * char prev_comm[TASK_COMM_LEN]; 479 * * 480 * 481 * TP_STRUCT__entry( 482 * __array( char, prev_comm, TASK_COMM_LEN ) 483 * __field( pid_t, prev_pid ) 484 * __field( int, prev_prio ) 485 * __array( char, next_comm, TASK_COMM_LEN ) 486 * __field( pid_t, next_pid ) 487 * __field( int, next_prio ) 488 * ), 489 * 490 * * 491 * * Assign the entry into the trace record, by embedding 492 * * a full C statement block into TP_fast_assign(). You 493 * * can refer to the trace record as '__entry' - 494 * * otherwise you can put arbitrary C code in here. 495 * * 496 * * Note: this C code will execute every time a trace event 497 * * happens, on an active tracepoint. 498 * * 499 * 500 * TP_fast_assign( 501 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN); 502 * __entry->prev_pid = prev->pid; 503 * __entry->prev_prio = prev->prio; 504 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN); 505 * __entry->next_pid = next->pid; 506 * __entry->next_prio = next->prio; 507 * ), 508 * 509 * * 510 * * Formatted output of a trace record via TP_printk(). 511 * * This is how the tracepoint will appear under ftrace 512 * * plugins that make use of this tracepoint. 513 * * 514 * * (raw-binary tracing wont actually perform this step.) 515 * * 516 * 517 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]", 518 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio, 519 * __entry->next_comm, __entry->next_pid, __entry->next_prio), 520 * 521 * ); 522 * 523 * This macro construct is thus used for the regular printk format 524 * tracing setup, it is used to construct a function pointer based 525 * tracepoint callback (this is used by programmatic plugins and 526 * can also by used by generic instrumentation like SystemTap), and 527 * it is also used to expose a structured trace record in 528 * /sys/kernel/debug/tracing/events/. 529 * 530 * A set of (un)registration functions can be passed to the variant 531 * TRACE_EVENT_FN to perform any (un)registration work. 532 */ 533 534 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print) 535 #define DEFINE_EVENT(template, name, proto, args) \ 536 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 537 #define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\ 538 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 539 #define DEFINE_EVENT_PRINT(template, name, proto, args, print) \ 540 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 541 #define DEFINE_EVENT_CONDITION(template, name, proto, \ 542 args, cond) \ 543 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 544 PARAMS(args), PARAMS(cond)) 545 546 #define TRACE_EVENT(name, proto, args, struct, assign, print) \ 547 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 548 #define TRACE_EVENT_FN(name, proto, args, struct, \ 549 assign, print, reg, unreg) \ 550 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 551 #define TRACE_EVENT_FN_COND(name, proto, args, cond, struct, \ 552 assign, print, reg, unreg) \ 553 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 554 PARAMS(args), PARAMS(cond)) 555 #define TRACE_EVENT_CONDITION(name, proto, args, cond, \ 556 struct, assign, print) \ 557 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 558 PARAMS(args), PARAMS(cond)) 559 560 #define TRACE_EVENT_FLAGS(event, flag) 561 562 #define TRACE_EVENT_PERF_PERM(event, expr...) 563 564 #define DECLARE_EVENT_NOP(name, proto, args) \ 565 static inline void trace_##name(proto) \ 566 { } \ 567 static inline bool trace_##name##_enabled(void) \ 568 { \ 569 return false; \ 570 } 571 572 #define TRACE_EVENT_NOP(name, proto, args, struct, assign, print) \ 573 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args)) 574 575 #define DECLARE_EVENT_CLASS_NOP(name, proto, args, tstruct, assign, print) 576 #define DEFINE_EVENT_NOP(template, name, proto, args) \ 577 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args)) 578 579 #endif /* ifdef TRACE_EVENT (see note above) */ 580