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, args) \ 156 do { \ 157 struct tracepoint_func *it_func_ptr; \ 158 void *__data; \ 159 it_func_ptr = \ 160 rcu_dereference_raw((&__tracepoint_##name)->funcs); \ 161 if (it_func_ptr) { \ 162 __data = (it_func_ptr)->data; \ 163 static_call(tp_func_##name)(__data, args); \ 164 } \ 165 } while (0) 166 #else 167 #define __DO_TRACE_CALL(name, args) __traceiter_##name(NULL, args) 168 #endif /* CONFIG_HAVE_STATIC_CALL */ 169 170 /* 171 * it_func[0] is never NULL because there is at least one element in the array 172 * when the array itself is non NULL. 173 */ 174 #define __DO_TRACE(name, args, cond, rcuidle) \ 175 do { \ 176 int __maybe_unused __idx = 0; \ 177 \ 178 if (!(cond)) \ 179 return; \ 180 \ 181 /* srcu can't be used from NMI */ \ 182 WARN_ON_ONCE(rcuidle && in_nmi()); \ 183 \ 184 /* keep srcu and sched-rcu usage consistent */ \ 185 preempt_disable_notrace(); \ 186 \ 187 /* \ 188 * For rcuidle callers, use srcu since sched-rcu \ 189 * doesn't work from the idle path. \ 190 */ \ 191 if (rcuidle) { \ 192 __idx = srcu_read_lock_notrace(&tracepoint_srcu);\ 193 rcu_irq_enter_irqson(); \ 194 } \ 195 \ 196 __DO_TRACE_CALL(name, TP_ARGS(args)); \ 197 \ 198 if (rcuidle) { \ 199 rcu_irq_exit_irqson(); \ 200 srcu_read_unlock_notrace(&tracepoint_srcu, __idx);\ 201 } \ 202 \ 203 preempt_enable_notrace(); \ 204 } while (0) 205 206 #ifndef MODULE 207 #define __DECLARE_TRACE_RCU(name, proto, args, cond) \ 208 static inline void trace_##name##_rcuidle(proto) \ 209 { \ 210 if (static_key_false(&__tracepoint_##name.key)) \ 211 __DO_TRACE(name, \ 212 TP_ARGS(args), \ 213 TP_CONDITION(cond), 1); \ 214 } 215 #else 216 #define __DECLARE_TRACE_RCU(name, proto, args, cond) 217 #endif 218 219 /* 220 * Make sure the alignment of the structure in the __tracepoints section will 221 * not add unwanted padding between the beginning of the section and the 222 * structure. Force alignment to the same alignment as the section start. 223 * 224 * When lockdep is enabled, we make sure to always do the RCU portions of 225 * the tracepoint code, regardless of whether tracing is on. However, 226 * don't check if the condition is false, due to interaction with idle 227 * instrumentation. This lets us find RCU issues triggered with tracepoints 228 * even when this tracepoint is off. This code has no purpose other than 229 * poking RCU a bit. 230 */ 231 #define __DECLARE_TRACE(name, proto, args, cond, data_proto) \ 232 extern int __traceiter_##name(data_proto); \ 233 DECLARE_STATIC_CALL(tp_func_##name, __traceiter_##name); \ 234 extern struct tracepoint __tracepoint_##name; \ 235 static inline void trace_##name(proto) \ 236 { \ 237 if (static_key_false(&__tracepoint_##name.key)) \ 238 __DO_TRACE(name, \ 239 TP_ARGS(args), \ 240 TP_CONDITION(cond), 0); \ 241 if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) { \ 242 rcu_read_lock_sched_notrace(); \ 243 rcu_dereference_sched(__tracepoint_##name.funcs);\ 244 rcu_read_unlock_sched_notrace(); \ 245 } \ 246 } \ 247 __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args), \ 248 PARAMS(cond)) \ 249 static inline int \ 250 register_trace_##name(void (*probe)(data_proto), void *data) \ 251 { \ 252 return tracepoint_probe_register(&__tracepoint_##name, \ 253 (void *)probe, data); \ 254 } \ 255 static inline int \ 256 register_trace_prio_##name(void (*probe)(data_proto), void *data,\ 257 int prio) \ 258 { \ 259 return tracepoint_probe_register_prio(&__tracepoint_##name, \ 260 (void *)probe, data, prio); \ 261 } \ 262 static inline int \ 263 unregister_trace_##name(void (*probe)(data_proto), void *data) \ 264 { \ 265 return tracepoint_probe_unregister(&__tracepoint_##name,\ 266 (void *)probe, data); \ 267 } \ 268 static inline void \ 269 check_trace_callback_type_##name(void (*cb)(data_proto)) \ 270 { \ 271 } \ 272 static inline bool \ 273 trace_##name##_enabled(void) \ 274 { \ 275 return static_key_false(&__tracepoint_##name.key); \ 276 } 277 278 /* 279 * We have no guarantee that gcc and the linker won't up-align the tracepoint 280 * structures, so we create an array of pointers that will be used for iteration 281 * on the tracepoints. 282 */ 283 #define DEFINE_TRACE_FN(_name, _reg, _unreg, proto, args) \ 284 static const char __tpstrtab_##_name[] \ 285 __section("__tracepoints_strings") = #_name; \ 286 extern struct static_call_key STATIC_CALL_KEY(tp_func_##_name); \ 287 int __traceiter_##_name(void *__data, proto); \ 288 struct tracepoint __tracepoint_##_name __used \ 289 __section("__tracepoints") = { \ 290 .name = __tpstrtab_##_name, \ 291 .key = STATIC_KEY_INIT_FALSE, \ 292 .static_call_key = &STATIC_CALL_KEY(tp_func_##_name), \ 293 .static_call_tramp = STATIC_CALL_TRAMP_ADDR(tp_func_##_name), \ 294 .iterator = &__traceiter_##_name, \ 295 .regfunc = _reg, \ 296 .unregfunc = _unreg, \ 297 .funcs = NULL }; \ 298 __TRACEPOINT_ENTRY(_name); \ 299 int __traceiter_##_name(void *__data, proto) \ 300 { \ 301 struct tracepoint_func *it_func_ptr; \ 302 void *it_func; \ 303 \ 304 it_func_ptr = \ 305 rcu_dereference_raw((&__tracepoint_##_name)->funcs); \ 306 if (it_func_ptr) { \ 307 do { \ 308 it_func = READ_ONCE((it_func_ptr)->func); \ 309 __data = (it_func_ptr)->data; \ 310 ((void(*)(void *, proto))(it_func))(__data, args); \ 311 } while ((++it_func_ptr)->func); \ 312 } \ 313 return 0; \ 314 } \ 315 DEFINE_STATIC_CALL(tp_func_##_name, __traceiter_##_name); 316 317 #define DEFINE_TRACE(name, proto, args) \ 318 DEFINE_TRACE_FN(name, NULL, NULL, PARAMS(proto), PARAMS(args)); 319 320 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \ 321 EXPORT_SYMBOL_GPL(__tracepoint_##name); \ 322 EXPORT_SYMBOL_GPL(__traceiter_##name); \ 323 EXPORT_STATIC_CALL_GPL(tp_func_##name) 324 #define EXPORT_TRACEPOINT_SYMBOL(name) \ 325 EXPORT_SYMBOL(__tracepoint_##name); \ 326 EXPORT_SYMBOL(__traceiter_##name); \ 327 EXPORT_STATIC_CALL(tp_func_##name) 328 329 330 #else /* !TRACEPOINTS_ENABLED */ 331 #define __DECLARE_TRACE(name, proto, args, cond, data_proto) \ 332 static inline void trace_##name(proto) \ 333 { } \ 334 static inline void trace_##name##_rcuidle(proto) \ 335 { } \ 336 static inline int \ 337 register_trace_##name(void (*probe)(data_proto), \ 338 void *data) \ 339 { \ 340 return -ENOSYS; \ 341 } \ 342 static inline int \ 343 unregister_trace_##name(void (*probe)(data_proto), \ 344 void *data) \ 345 { \ 346 return -ENOSYS; \ 347 } \ 348 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \ 349 { \ 350 } \ 351 static inline bool \ 352 trace_##name##_enabled(void) \ 353 { \ 354 return false; \ 355 } 356 357 #define DEFINE_TRACE_FN(name, reg, unreg, proto, args) 358 #define DEFINE_TRACE(name, proto, args) 359 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) 360 #define EXPORT_TRACEPOINT_SYMBOL(name) 361 362 #endif /* TRACEPOINTS_ENABLED */ 363 364 #ifdef CONFIG_TRACING 365 /** 366 * tracepoint_string - register constant persistent string to trace system 367 * @str - a constant persistent string that will be referenced in tracepoints 368 * 369 * If constant strings are being used in tracepoints, it is faster and 370 * more efficient to just save the pointer to the string and reference 371 * that with a printf "%s" instead of saving the string in the ring buffer 372 * and wasting space and time. 373 * 374 * The problem with the above approach is that userspace tools that read 375 * the binary output of the trace buffers do not have access to the string. 376 * Instead they just show the address of the string which is not very 377 * useful to users. 378 * 379 * With tracepoint_string(), the string will be registered to the tracing 380 * system and exported to userspace via the debugfs/tracing/printk_formats 381 * file that maps the string address to the string text. This way userspace 382 * tools that read the binary buffers have a way to map the pointers to 383 * the ASCII strings they represent. 384 * 385 * The @str used must be a constant string and persistent as it would not 386 * make sense to show a string that no longer exists. But it is still fine 387 * to be used with modules, because when modules are unloaded, if they 388 * had tracepoints, the ring buffers are cleared too. As long as the string 389 * does not change during the life of the module, it is fine to use 390 * tracepoint_string() within a module. 391 */ 392 #define tracepoint_string(str) \ 393 ({ \ 394 static const char *___tp_str __tracepoint_string = str; \ 395 ___tp_str; \ 396 }) 397 #define __tracepoint_string __used __section("__tracepoint_str") 398 #else 399 /* 400 * tracepoint_string() is used to save the string address for userspace 401 * tracing tools. When tracing isn't configured, there's no need to save 402 * anything. 403 */ 404 # define tracepoint_string(str) str 405 # define __tracepoint_string 406 #endif 407 408 #define DECLARE_TRACE(name, proto, args) \ 409 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \ 410 cpu_online(raw_smp_processor_id()), \ 411 PARAMS(void *__data, proto)) 412 413 #define DECLARE_TRACE_CONDITION(name, proto, args, cond) \ 414 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \ 415 cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \ 416 PARAMS(void *__data, proto)) 417 418 #define TRACE_EVENT_FLAGS(event, flag) 419 420 #define TRACE_EVENT_PERF_PERM(event, expr...) 421 422 #endif /* DECLARE_TRACE */ 423 424 #ifndef TRACE_EVENT 425 /* 426 * For use with the TRACE_EVENT macro: 427 * 428 * We define a tracepoint, its arguments, its printk format 429 * and its 'fast binary record' layout. 430 * 431 * Firstly, name your tracepoint via TRACE_EVENT(name : the 432 * 'subsystem_event' notation is fine. 433 * 434 * Think about this whole construct as the 435 * 'trace_sched_switch() function' from now on. 436 * 437 * 438 * TRACE_EVENT(sched_switch, 439 * 440 * * 441 * * A function has a regular function arguments 442 * * prototype, declare it via TP_PROTO(): 443 * * 444 * 445 * TP_PROTO(struct rq *rq, struct task_struct *prev, 446 * struct task_struct *next), 447 * 448 * * 449 * * Define the call signature of the 'function'. 450 * * (Design sidenote: we use this instead of a 451 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.) 452 * * 453 * 454 * TP_ARGS(rq, prev, next), 455 * 456 * * 457 * * Fast binary tracing: define the trace record via 458 * * TP_STRUCT__entry(). You can think about it like a 459 * * regular C structure local variable definition. 460 * * 461 * * This is how the trace record is structured and will 462 * * be saved into the ring buffer. These are the fields 463 * * that will be exposed to user-space in 464 * * /sys/kernel/debug/tracing/events/<*>/format. 465 * * 466 * * The declared 'local variable' is called '__entry' 467 * * 468 * * __field(pid_t, prev_prid) is equivalent to a standard declaration: 469 * * 470 * * pid_t prev_pid; 471 * * 472 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to: 473 * * 474 * * char prev_comm[TASK_COMM_LEN]; 475 * * 476 * 477 * TP_STRUCT__entry( 478 * __array( char, prev_comm, TASK_COMM_LEN ) 479 * __field( pid_t, prev_pid ) 480 * __field( int, prev_prio ) 481 * __array( char, next_comm, TASK_COMM_LEN ) 482 * __field( pid_t, next_pid ) 483 * __field( int, next_prio ) 484 * ), 485 * 486 * * 487 * * Assign the entry into the trace record, by embedding 488 * * a full C statement block into TP_fast_assign(). You 489 * * can refer to the trace record as '__entry' - 490 * * otherwise you can put arbitrary C code in here. 491 * * 492 * * Note: this C code will execute every time a trace event 493 * * happens, on an active tracepoint. 494 * * 495 * 496 * TP_fast_assign( 497 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN); 498 * __entry->prev_pid = prev->pid; 499 * __entry->prev_prio = prev->prio; 500 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN); 501 * __entry->next_pid = next->pid; 502 * __entry->next_prio = next->prio; 503 * ), 504 * 505 * * 506 * * Formatted output of a trace record via TP_printk(). 507 * * This is how the tracepoint will appear under ftrace 508 * * plugins that make use of this tracepoint. 509 * * 510 * * (raw-binary tracing wont actually perform this step.) 511 * * 512 * 513 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]", 514 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio, 515 * __entry->next_comm, __entry->next_pid, __entry->next_prio), 516 * 517 * ); 518 * 519 * This macro construct is thus used for the regular printk format 520 * tracing setup, it is used to construct a function pointer based 521 * tracepoint callback (this is used by programmatic plugins and 522 * can also by used by generic instrumentation like SystemTap), and 523 * it is also used to expose a structured trace record in 524 * /sys/kernel/debug/tracing/events/. 525 * 526 * A set of (un)registration functions can be passed to the variant 527 * TRACE_EVENT_FN to perform any (un)registration work. 528 */ 529 530 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print) 531 #define DEFINE_EVENT(template, name, proto, args) \ 532 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 533 #define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\ 534 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 535 #define DEFINE_EVENT_PRINT(template, name, proto, args, print) \ 536 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 537 #define DEFINE_EVENT_CONDITION(template, name, proto, \ 538 args, cond) \ 539 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 540 PARAMS(args), PARAMS(cond)) 541 542 #define TRACE_EVENT(name, proto, args, struct, assign, print) \ 543 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 544 #define TRACE_EVENT_FN(name, proto, args, struct, \ 545 assign, print, reg, unreg) \ 546 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 547 #define TRACE_EVENT_FN_COND(name, proto, args, cond, struct, \ 548 assign, print, reg, unreg) \ 549 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 550 PARAMS(args), PARAMS(cond)) 551 #define TRACE_EVENT_CONDITION(name, proto, args, cond, \ 552 struct, assign, print) \ 553 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 554 PARAMS(args), PARAMS(cond)) 555 556 #define TRACE_EVENT_FLAGS(event, flag) 557 558 #define TRACE_EVENT_PERF_PERM(event, expr...) 559 560 #define DECLARE_EVENT_NOP(name, proto, args) \ 561 static inline void trace_##name(proto) \ 562 { } \ 563 static inline bool trace_##name##_enabled(void) \ 564 { \ 565 return false; \ 566 } 567 568 #define TRACE_EVENT_NOP(name, proto, args, struct, assign, print) \ 569 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args)) 570 571 #define DECLARE_EVENT_CLASS_NOP(name, proto, args, tstruct, assign, print) 572 #define DEFINE_EVENT_NOP(template, name, proto, args) \ 573 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args)) 574 575 #endif /* ifdef TRACE_EVENT (see note above) */ 576