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 do the RCU portions of 246 * the tracepoint code, regardless of whether tracing is on. However, 247 * don't check if the condition is false, due to interaction with idle 248 * instrumentation. This lets us find RCU issues triggered with tracepoints 249 * even when this tracepoint is off. This code has no purpose other than 250 * poking RCU a bit. 251 */ 252 #define __DECLARE_TRACE(name, proto, args, cond, data_proto) \ 253 extern int __traceiter_##name(data_proto); \ 254 DECLARE_STATIC_CALL(tp_func_##name, __traceiter_##name); \ 255 extern struct tracepoint __tracepoint_##name; \ 256 static inline void trace_##name(proto) \ 257 { \ 258 if (static_key_false(&__tracepoint_##name.key)) \ 259 __DO_TRACE(name, \ 260 TP_ARGS(args), \ 261 TP_CONDITION(cond), 0); \ 262 if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) { \ 263 rcu_read_lock_sched_notrace(); \ 264 rcu_dereference_sched(__tracepoint_##name.funcs);\ 265 rcu_read_unlock_sched_notrace(); \ 266 } \ 267 } \ 268 __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args), \ 269 PARAMS(cond)) \ 270 static inline int \ 271 register_trace_##name(void (*probe)(data_proto), void *data) \ 272 { \ 273 return tracepoint_probe_register(&__tracepoint_##name, \ 274 (void *)probe, data); \ 275 } \ 276 static inline int \ 277 register_trace_prio_##name(void (*probe)(data_proto), void *data,\ 278 int prio) \ 279 { \ 280 return tracepoint_probe_register_prio(&__tracepoint_##name, \ 281 (void *)probe, data, prio); \ 282 } \ 283 static inline int \ 284 unregister_trace_##name(void (*probe)(data_proto), void *data) \ 285 { \ 286 return tracepoint_probe_unregister(&__tracepoint_##name,\ 287 (void *)probe, data); \ 288 } \ 289 static inline void \ 290 check_trace_callback_type_##name(void (*cb)(data_proto)) \ 291 { \ 292 } \ 293 static inline bool \ 294 trace_##name##_enabled(void) \ 295 { \ 296 return static_key_false(&__tracepoint_##name.key); \ 297 } 298 299 /* 300 * We have no guarantee that gcc and the linker won't up-align the tracepoint 301 * structures, so we create an array of pointers that will be used for iteration 302 * on the tracepoints. 303 */ 304 #define DEFINE_TRACE_FN(_name, _reg, _unreg, proto, args) \ 305 static const char __tpstrtab_##_name[] \ 306 __section("__tracepoints_strings") = #_name; \ 307 extern struct static_call_key STATIC_CALL_KEY(tp_func_##_name); \ 308 int __traceiter_##_name(void *__data, proto); \ 309 struct tracepoint __tracepoint_##_name __used \ 310 __section("__tracepoints") = { \ 311 .name = __tpstrtab_##_name, \ 312 .key = STATIC_KEY_INIT_FALSE, \ 313 .static_call_key = &STATIC_CALL_KEY(tp_func_##_name), \ 314 .static_call_tramp = STATIC_CALL_TRAMP_ADDR(tp_func_##_name), \ 315 .iterator = &__traceiter_##_name, \ 316 .regfunc = _reg, \ 317 .unregfunc = _unreg, \ 318 .funcs = NULL }; \ 319 __TRACEPOINT_ENTRY(_name); \ 320 int __traceiter_##_name(void *__data, proto) \ 321 { \ 322 struct tracepoint_func *it_func_ptr; \ 323 void *it_func; \ 324 \ 325 it_func_ptr = \ 326 rcu_dereference_raw((&__tracepoint_##_name)->funcs); \ 327 if (it_func_ptr) { \ 328 do { \ 329 it_func = READ_ONCE((it_func_ptr)->func); \ 330 __data = (it_func_ptr)->data; \ 331 ((void(*)(void *, proto))(it_func))(__data, args); \ 332 } while ((++it_func_ptr)->func); \ 333 } \ 334 return 0; \ 335 } \ 336 DEFINE_STATIC_CALL(tp_func_##_name, __traceiter_##_name); 337 338 #define DEFINE_TRACE(name, proto, args) \ 339 DEFINE_TRACE_FN(name, NULL, NULL, PARAMS(proto), PARAMS(args)); 340 341 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \ 342 EXPORT_SYMBOL_GPL(__tracepoint_##name); \ 343 EXPORT_SYMBOL_GPL(__traceiter_##name); \ 344 EXPORT_STATIC_CALL_GPL(tp_func_##name) 345 #define EXPORT_TRACEPOINT_SYMBOL(name) \ 346 EXPORT_SYMBOL(__tracepoint_##name); \ 347 EXPORT_SYMBOL(__traceiter_##name); \ 348 EXPORT_STATIC_CALL(tp_func_##name) 349 350 351 #else /* !TRACEPOINTS_ENABLED */ 352 #define __DECLARE_TRACE(name, proto, args, cond, data_proto) \ 353 static inline void trace_##name(proto) \ 354 { } \ 355 static inline void trace_##name##_rcuidle(proto) \ 356 { } \ 357 static inline int \ 358 register_trace_##name(void (*probe)(data_proto), \ 359 void *data) \ 360 { \ 361 return -ENOSYS; \ 362 } \ 363 static inline int \ 364 unregister_trace_##name(void (*probe)(data_proto), \ 365 void *data) \ 366 { \ 367 return -ENOSYS; \ 368 } \ 369 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \ 370 { \ 371 } \ 372 static inline bool \ 373 trace_##name##_enabled(void) \ 374 { \ 375 return false; \ 376 } 377 378 #define DEFINE_TRACE_FN(name, reg, unreg, proto, args) 379 #define DEFINE_TRACE(name, proto, args) 380 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) 381 #define EXPORT_TRACEPOINT_SYMBOL(name) 382 383 #endif /* TRACEPOINTS_ENABLED */ 384 385 #ifdef CONFIG_TRACING 386 /** 387 * tracepoint_string - register constant persistent string to trace system 388 * @str - a constant persistent string that will be referenced in tracepoints 389 * 390 * If constant strings are being used in tracepoints, it is faster and 391 * more efficient to just save the pointer to the string and reference 392 * that with a printf "%s" instead of saving the string in the ring buffer 393 * and wasting space and time. 394 * 395 * The problem with the above approach is that userspace tools that read 396 * the binary output of the trace buffers do not have access to the string. 397 * Instead they just show the address of the string which is not very 398 * useful to users. 399 * 400 * With tracepoint_string(), the string will be registered to the tracing 401 * system and exported to userspace via the debugfs/tracing/printk_formats 402 * file that maps the string address to the string text. This way userspace 403 * tools that read the binary buffers have a way to map the pointers to 404 * the ASCII strings they represent. 405 * 406 * The @str used must be a constant string and persistent as it would not 407 * make sense to show a string that no longer exists. But it is still fine 408 * to be used with modules, because when modules are unloaded, if they 409 * had tracepoints, the ring buffers are cleared too. As long as the string 410 * does not change during the life of the module, it is fine to use 411 * tracepoint_string() within a module. 412 */ 413 #define tracepoint_string(str) \ 414 ({ \ 415 static const char *___tp_str __tracepoint_string = str; \ 416 ___tp_str; \ 417 }) 418 #define __tracepoint_string __used __section("__tracepoint_str") 419 #else 420 /* 421 * tracepoint_string() is used to save the string address for userspace 422 * tracing tools. When tracing isn't configured, there's no need to save 423 * anything. 424 */ 425 # define tracepoint_string(str) str 426 # define __tracepoint_string 427 #endif 428 429 #define DECLARE_TRACE(name, proto, args) \ 430 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \ 431 cpu_online(raw_smp_processor_id()), \ 432 PARAMS(void *__data, proto)) 433 434 #define DECLARE_TRACE_CONDITION(name, proto, args, cond) \ 435 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \ 436 cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \ 437 PARAMS(void *__data, proto)) 438 439 #define TRACE_EVENT_FLAGS(event, flag) 440 441 #define TRACE_EVENT_PERF_PERM(event, expr...) 442 443 #endif /* DECLARE_TRACE */ 444 445 #ifndef TRACE_EVENT 446 /* 447 * For use with the TRACE_EVENT macro: 448 * 449 * We define a tracepoint, its arguments, its printk format 450 * and its 'fast binary record' layout. 451 * 452 * Firstly, name your tracepoint via TRACE_EVENT(name : the 453 * 'subsystem_event' notation is fine. 454 * 455 * Think about this whole construct as the 456 * 'trace_sched_switch() function' from now on. 457 * 458 * 459 * TRACE_EVENT(sched_switch, 460 * 461 * * 462 * * A function has a regular function arguments 463 * * prototype, declare it via TP_PROTO(): 464 * * 465 * 466 * TP_PROTO(struct rq *rq, struct task_struct *prev, 467 * struct task_struct *next), 468 * 469 * * 470 * * Define the call signature of the 'function'. 471 * * (Design sidenote: we use this instead of a 472 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.) 473 * * 474 * 475 * TP_ARGS(rq, prev, next), 476 * 477 * * 478 * * Fast binary tracing: define the trace record via 479 * * TP_STRUCT__entry(). You can think about it like a 480 * * regular C structure local variable definition. 481 * * 482 * * This is how the trace record is structured and will 483 * * be saved into the ring buffer. These are the fields 484 * * that will be exposed to user-space in 485 * * /sys/kernel/tracing/events/<*>/format. 486 * * 487 * * The declared 'local variable' is called '__entry' 488 * * 489 * * __field(pid_t, prev_pid) is equivalent to a standard declaration: 490 * * 491 * * pid_t prev_pid; 492 * * 493 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to: 494 * * 495 * * char prev_comm[TASK_COMM_LEN]; 496 * * 497 * 498 * TP_STRUCT__entry( 499 * __array( char, prev_comm, TASK_COMM_LEN ) 500 * __field( pid_t, prev_pid ) 501 * __field( int, prev_prio ) 502 * __array( char, next_comm, TASK_COMM_LEN ) 503 * __field( pid_t, next_pid ) 504 * __field( int, next_prio ) 505 * ), 506 * 507 * * 508 * * Assign the entry into the trace record, by embedding 509 * * a full C statement block into TP_fast_assign(). You 510 * * can refer to the trace record as '__entry' - 511 * * otherwise you can put arbitrary C code in here. 512 * * 513 * * Note: this C code will execute every time a trace event 514 * * happens, on an active tracepoint. 515 * * 516 * 517 * TP_fast_assign( 518 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN); 519 * __entry->prev_pid = prev->pid; 520 * __entry->prev_prio = prev->prio; 521 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN); 522 * __entry->next_pid = next->pid; 523 * __entry->next_prio = next->prio; 524 * ), 525 * 526 * * 527 * * Formatted output of a trace record via TP_printk(). 528 * * This is how the tracepoint will appear under ftrace 529 * * plugins that make use of this tracepoint. 530 * * 531 * * (raw-binary tracing wont actually perform this step.) 532 * * 533 * 534 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]", 535 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio, 536 * __entry->next_comm, __entry->next_pid, __entry->next_prio), 537 * 538 * ); 539 * 540 * This macro construct is thus used for the regular printk format 541 * tracing setup, it is used to construct a function pointer based 542 * tracepoint callback (this is used by programmatic plugins and 543 * can also by used by generic instrumentation like SystemTap), and 544 * it is also used to expose a structured trace record in 545 * /sys/kernel/tracing/events/. 546 * 547 * A set of (un)registration functions can be passed to the variant 548 * TRACE_EVENT_FN to perform any (un)registration work. 549 */ 550 551 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print) 552 #define DEFINE_EVENT(template, name, proto, args) \ 553 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 554 #define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\ 555 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 556 #define DEFINE_EVENT_PRINT(template, name, proto, args, print) \ 557 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 558 #define DEFINE_EVENT_CONDITION(template, name, proto, \ 559 args, cond) \ 560 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 561 PARAMS(args), PARAMS(cond)) 562 563 #define TRACE_EVENT(name, proto, args, struct, assign, print) \ 564 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 565 #define TRACE_EVENT_FN(name, proto, args, struct, \ 566 assign, print, reg, unreg) \ 567 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 568 #define TRACE_EVENT_FN_COND(name, proto, args, cond, struct, \ 569 assign, print, reg, unreg) \ 570 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 571 PARAMS(args), PARAMS(cond)) 572 #define TRACE_EVENT_CONDITION(name, proto, args, cond, \ 573 struct, assign, print) \ 574 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 575 PARAMS(args), PARAMS(cond)) 576 577 #define TRACE_EVENT_FLAGS(event, flag) 578 579 #define TRACE_EVENT_PERF_PERM(event, expr...) 580 581 #define DECLARE_EVENT_NOP(name, proto, args) \ 582 static inline void trace_##name(proto) \ 583 { } \ 584 static inline bool trace_##name##_enabled(void) \ 585 { \ 586 return false; \ 587 } 588 589 #define TRACE_EVENT_NOP(name, proto, args, struct, assign, print) \ 590 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args)) 591 592 #define DECLARE_EVENT_CLASS_NOP(name, proto, args, tstruct, assign, print) 593 #define DEFINE_EVENT_NOP(template, name, proto, args) \ 594 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args)) 595 596 #endif /* ifdef TRACE_EVENT (see note above) */ 597