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