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 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 struct srcu_struct tracepoint_srcu; 36 37 extern int 38 tracepoint_probe_register(struct tracepoint *tp, void *probe, void *data); 39 extern int 40 tracepoint_probe_register_prio(struct tracepoint *tp, void *probe, void *data, 41 int prio); 42 extern int 43 tracepoint_probe_unregister(struct tracepoint *tp, void *probe, void *data); 44 extern void 45 for_each_kernel_tracepoint(void (*fct)(struct tracepoint *tp, void *priv), 46 void *priv); 47 48 #ifdef CONFIG_MODULES 49 struct tp_module { 50 struct list_head list; 51 struct module *mod; 52 }; 53 54 bool trace_module_has_bad_taint(struct module *mod); 55 extern int register_tracepoint_module_notifier(struct notifier_block *nb); 56 extern int unregister_tracepoint_module_notifier(struct notifier_block *nb); 57 #else 58 static inline bool trace_module_has_bad_taint(struct module *mod) 59 { 60 return false; 61 } 62 static inline 63 int register_tracepoint_module_notifier(struct notifier_block *nb) 64 { 65 return 0; 66 } 67 static inline 68 int unregister_tracepoint_module_notifier(struct notifier_block *nb) 69 { 70 return 0; 71 } 72 #endif /* CONFIG_MODULES */ 73 74 /* 75 * tracepoint_synchronize_unregister must be called between the last tracepoint 76 * probe unregistration and the end of module exit to make sure there is no 77 * caller executing a probe when it is freed. 78 */ 79 #ifdef CONFIG_TRACEPOINTS 80 static inline void tracepoint_synchronize_unregister(void) 81 { 82 synchronize_srcu(&tracepoint_srcu); 83 synchronize_rcu(); 84 } 85 #else 86 static inline void tracepoint_synchronize_unregister(void) 87 { } 88 #endif 89 90 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS 91 extern int syscall_regfunc(void); 92 extern void syscall_unregfunc(void); 93 #endif /* CONFIG_HAVE_SYSCALL_TRACEPOINTS */ 94 95 #define PARAMS(args...) args 96 97 #define TRACE_DEFINE_ENUM(x) 98 #define TRACE_DEFINE_SIZEOF(x) 99 100 #ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS 101 static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p) 102 { 103 return offset_to_ptr(p); 104 } 105 106 #define __TRACEPOINT_ENTRY(name) \ 107 asm(" .section \"__tracepoints_ptrs\", \"a\" \n" \ 108 " .balign 4 \n" \ 109 " .long __tracepoint_" #name " - . \n" \ 110 " .previous \n") 111 #else 112 static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p) 113 { 114 return *p; 115 } 116 117 #define __TRACEPOINT_ENTRY(name) \ 118 static tracepoint_ptr_t __tracepoint_ptr_##name __used \ 119 __section(__tracepoints_ptrs) = &__tracepoint_##name 120 #endif 121 122 #endif /* _LINUX_TRACEPOINT_H */ 123 124 /* 125 * Note: we keep the TRACE_EVENT and DECLARE_TRACE outside the include 126 * file ifdef protection. 127 * This is due to the way trace events work. If a file includes two 128 * trace event headers under one "CREATE_TRACE_POINTS" the first include 129 * will override the TRACE_EVENT and break the second include. 130 */ 131 132 #ifndef DECLARE_TRACE 133 134 #define TP_PROTO(args...) args 135 #define TP_ARGS(args...) args 136 #define TP_CONDITION(args...) args 137 138 /* 139 * Individual subsystem my have a separate configuration to 140 * enable their tracepoints. By default, this file will create 141 * the tracepoints if CONFIG_TRACEPOINT is defined. If a subsystem 142 * wants to be able to disable its tracepoints from being created 143 * it can define NOTRACE before including the tracepoint headers. 144 */ 145 #if defined(CONFIG_TRACEPOINTS) && !defined(NOTRACE) 146 #define TRACEPOINTS_ENABLED 147 #endif 148 149 #ifdef TRACEPOINTS_ENABLED 150 151 /* 152 * it_func[0] is never NULL because there is at least one element in the array 153 * when the array itself is non NULL. 154 * 155 * Note, the proto and args passed in includes "__data" as the first parameter. 156 * The reason for this is to handle the "void" prototype. If a tracepoint 157 * has a "void" prototype, then it is invalid to declare a function 158 * as "(void *, void)". 159 */ 160 #define __DO_TRACE(tp, proto, args, cond, rcuidle) \ 161 do { \ 162 struct tracepoint_func *it_func_ptr; \ 163 void *it_func; \ 164 void *__data; \ 165 int __maybe_unused __idx = 0; \ 166 \ 167 if (!(cond)) \ 168 return; \ 169 \ 170 /* srcu can't be used from NMI */ \ 171 WARN_ON_ONCE(rcuidle && in_nmi()); \ 172 \ 173 /* keep srcu and sched-rcu usage consistent */ \ 174 preempt_disable_notrace(); \ 175 \ 176 /* \ 177 * For rcuidle callers, use srcu since sched-rcu \ 178 * doesn't work from the idle path. \ 179 */ \ 180 if (rcuidle) { \ 181 __idx = srcu_read_lock_notrace(&tracepoint_srcu);\ 182 rcu_irq_enter_irqson(); \ 183 } \ 184 \ 185 it_func_ptr = rcu_dereference_raw((tp)->funcs); \ 186 \ 187 if (it_func_ptr) { \ 188 do { \ 189 it_func = (it_func_ptr)->func; \ 190 __data = (it_func_ptr)->data; \ 191 ((void(*)(proto))(it_func))(args); \ 192 } while ((++it_func_ptr)->func); \ 193 } \ 194 \ 195 if (rcuidle) { \ 196 rcu_irq_exit_irqson(); \ 197 srcu_read_unlock_notrace(&tracepoint_srcu, __idx);\ 198 } \ 199 \ 200 preempt_enable_notrace(); \ 201 } while (0) 202 203 #ifndef MODULE 204 #define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args) \ 205 static inline void trace_##name##_rcuidle(proto) \ 206 { \ 207 if (static_key_false(&__tracepoint_##name.key)) \ 208 __DO_TRACE(&__tracepoint_##name, \ 209 TP_PROTO(data_proto), \ 210 TP_ARGS(data_args), \ 211 TP_CONDITION(cond), 1); \ 212 } 213 #else 214 #define __DECLARE_TRACE_RCU(name, proto, args, cond, data_proto, data_args) 215 #endif 216 217 /* 218 * Make sure the alignment of the structure in the __tracepoints section will 219 * not add unwanted padding between the beginning of the section and the 220 * structure. Force alignment to the same alignment as the section start. 221 * 222 * When lockdep is enabled, we make sure to always do the RCU portions of 223 * the tracepoint code, regardless of whether tracing is on. However, 224 * don't check if the condition is false, due to interaction with idle 225 * instrumentation. This lets us find RCU issues triggered with tracepoints 226 * even when this tracepoint is off. This code has no purpose other than 227 * poking RCU a bit. 228 */ 229 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \ 230 extern struct tracepoint __tracepoint_##name; \ 231 static inline void trace_##name(proto) \ 232 { \ 233 if (static_key_false(&__tracepoint_##name.key)) \ 234 __DO_TRACE(&__tracepoint_##name, \ 235 TP_PROTO(data_proto), \ 236 TP_ARGS(data_args), \ 237 TP_CONDITION(cond), 0); \ 238 if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) { \ 239 rcu_read_lock_sched_notrace(); \ 240 rcu_dereference_sched(__tracepoint_##name.funcs);\ 241 rcu_read_unlock_sched_notrace(); \ 242 } \ 243 } \ 244 __DECLARE_TRACE_RCU(name, PARAMS(proto), PARAMS(args), \ 245 PARAMS(cond), PARAMS(data_proto), PARAMS(data_args)) \ 246 static inline int \ 247 register_trace_##name(void (*probe)(data_proto), void *data) \ 248 { \ 249 return tracepoint_probe_register(&__tracepoint_##name, \ 250 (void *)probe, data); \ 251 } \ 252 static inline int \ 253 register_trace_prio_##name(void (*probe)(data_proto), void *data,\ 254 int prio) \ 255 { \ 256 return tracepoint_probe_register_prio(&__tracepoint_##name, \ 257 (void *)probe, data, prio); \ 258 } \ 259 static inline int \ 260 unregister_trace_##name(void (*probe)(data_proto), void *data) \ 261 { \ 262 return tracepoint_probe_unregister(&__tracepoint_##name,\ 263 (void *)probe, data); \ 264 } \ 265 static inline void \ 266 check_trace_callback_type_##name(void (*cb)(data_proto)) \ 267 { \ 268 } \ 269 static inline bool \ 270 trace_##name##_enabled(void) \ 271 { \ 272 return static_key_false(&__tracepoint_##name.key); \ 273 } 274 275 /* 276 * We have no guarantee that gcc and the linker won't up-align the tracepoint 277 * structures, so we create an array of pointers that will be used for iteration 278 * on the tracepoints. 279 */ 280 #define DEFINE_TRACE_FN(name, reg, unreg) \ 281 static const char __tpstrtab_##name[] \ 282 __section(__tracepoints_strings) = #name; \ 283 struct tracepoint __tracepoint_##name __used \ 284 __section(__tracepoints) = \ 285 { __tpstrtab_##name, STATIC_KEY_INIT_FALSE, reg, unreg, NULL };\ 286 __TRACEPOINT_ENTRY(name); 287 288 #define DEFINE_TRACE(name) \ 289 DEFINE_TRACE_FN(name, NULL, NULL); 290 291 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) \ 292 EXPORT_SYMBOL_GPL(__tracepoint_##name) 293 #define EXPORT_TRACEPOINT_SYMBOL(name) \ 294 EXPORT_SYMBOL(__tracepoint_##name) 295 296 #else /* !TRACEPOINTS_ENABLED */ 297 #define __DECLARE_TRACE(name, proto, args, cond, data_proto, data_args) \ 298 static inline void trace_##name(proto) \ 299 { } \ 300 static inline void trace_##name##_rcuidle(proto) \ 301 { } \ 302 static inline int \ 303 register_trace_##name(void (*probe)(data_proto), \ 304 void *data) \ 305 { \ 306 return -ENOSYS; \ 307 } \ 308 static inline int \ 309 unregister_trace_##name(void (*probe)(data_proto), \ 310 void *data) \ 311 { \ 312 return -ENOSYS; \ 313 } \ 314 static inline void check_trace_callback_type_##name(void (*cb)(data_proto)) \ 315 { \ 316 } \ 317 static inline bool \ 318 trace_##name##_enabled(void) \ 319 { \ 320 return false; \ 321 } 322 323 #define DEFINE_TRACE_FN(name, reg, unreg) 324 #define DEFINE_TRACE(name) 325 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name) 326 #define EXPORT_TRACEPOINT_SYMBOL(name) 327 328 #endif /* TRACEPOINTS_ENABLED */ 329 330 #ifdef CONFIG_TRACING 331 /** 332 * tracepoint_string - register constant persistent string to trace system 333 * @str - a constant persistent string that will be referenced in tracepoints 334 * 335 * If constant strings are being used in tracepoints, it is faster and 336 * more efficient to just save the pointer to the string and reference 337 * that with a printf "%s" instead of saving the string in the ring buffer 338 * and wasting space and time. 339 * 340 * The problem with the above approach is that userspace tools that read 341 * the binary output of the trace buffers do not have access to the string. 342 * Instead they just show the address of the string which is not very 343 * useful to users. 344 * 345 * With tracepoint_string(), the string will be registered to the tracing 346 * system and exported to userspace via the debugfs/tracing/printk_formats 347 * file that maps the string address to the string text. This way userspace 348 * tools that read the binary buffers have a way to map the pointers to 349 * the ASCII strings they represent. 350 * 351 * The @str used must be a constant string and persistent as it would not 352 * make sense to show a string that no longer exists. But it is still fine 353 * to be used with modules, because when modules are unloaded, if they 354 * had tracepoints, the ring buffers are cleared too. As long as the string 355 * does not change during the life of the module, it is fine to use 356 * tracepoint_string() within a module. 357 */ 358 #define tracepoint_string(str) \ 359 ({ \ 360 static const char *___tp_str __tracepoint_string = str; \ 361 ___tp_str; \ 362 }) 363 #define __tracepoint_string __used __section(__tracepoint_str) 364 #else 365 /* 366 * tracepoint_string() is used to save the string address for userspace 367 * tracing tools. When tracing isn't configured, there's no need to save 368 * anything. 369 */ 370 # define tracepoint_string(str) str 371 # define __tracepoint_string 372 #endif 373 374 #define DECLARE_TRACE(name, proto, args) \ 375 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \ 376 cpu_online(raw_smp_processor_id()), \ 377 PARAMS(void *__data, proto), \ 378 PARAMS(__data, args)) 379 380 #define DECLARE_TRACE_CONDITION(name, proto, args, cond) \ 381 __DECLARE_TRACE(name, PARAMS(proto), PARAMS(args), \ 382 cpu_online(raw_smp_processor_id()) && (PARAMS(cond)), \ 383 PARAMS(void *__data, proto), \ 384 PARAMS(__data, args)) 385 386 #define TRACE_EVENT_FLAGS(event, flag) 387 388 #define TRACE_EVENT_PERF_PERM(event, expr...) 389 390 #endif /* DECLARE_TRACE */ 391 392 #ifndef TRACE_EVENT 393 /* 394 * For use with the TRACE_EVENT macro: 395 * 396 * We define a tracepoint, its arguments, its printk format 397 * and its 'fast binary record' layout. 398 * 399 * Firstly, name your tracepoint via TRACE_EVENT(name : the 400 * 'subsystem_event' notation is fine. 401 * 402 * Think about this whole construct as the 403 * 'trace_sched_switch() function' from now on. 404 * 405 * 406 * TRACE_EVENT(sched_switch, 407 * 408 * * 409 * * A function has a regular function arguments 410 * * prototype, declare it via TP_PROTO(): 411 * * 412 * 413 * TP_PROTO(struct rq *rq, struct task_struct *prev, 414 * struct task_struct *next), 415 * 416 * * 417 * * Define the call signature of the 'function'. 418 * * (Design sidenote: we use this instead of a 419 * * TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.) 420 * * 421 * 422 * TP_ARGS(rq, prev, next), 423 * 424 * * 425 * * Fast binary tracing: define the trace record via 426 * * TP_STRUCT__entry(). You can think about it like a 427 * * regular C structure local variable definition. 428 * * 429 * * This is how the trace record is structured and will 430 * * be saved into the ring buffer. These are the fields 431 * * that will be exposed to user-space in 432 * * /sys/kernel/debug/tracing/events/<*>/format. 433 * * 434 * * The declared 'local variable' is called '__entry' 435 * * 436 * * __field(pid_t, prev_prid) is equivalent to a standard declariton: 437 * * 438 * * pid_t prev_pid; 439 * * 440 * * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to: 441 * * 442 * * char prev_comm[TASK_COMM_LEN]; 443 * * 444 * 445 * TP_STRUCT__entry( 446 * __array( char, prev_comm, TASK_COMM_LEN ) 447 * __field( pid_t, prev_pid ) 448 * __field( int, prev_prio ) 449 * __array( char, next_comm, TASK_COMM_LEN ) 450 * __field( pid_t, next_pid ) 451 * __field( int, next_prio ) 452 * ), 453 * 454 * * 455 * * Assign the entry into the trace record, by embedding 456 * * a full C statement block into TP_fast_assign(). You 457 * * can refer to the trace record as '__entry' - 458 * * otherwise you can put arbitrary C code in here. 459 * * 460 * * Note: this C code will execute every time a trace event 461 * * happens, on an active tracepoint. 462 * * 463 * 464 * TP_fast_assign( 465 * memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN); 466 * __entry->prev_pid = prev->pid; 467 * __entry->prev_prio = prev->prio; 468 * memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN); 469 * __entry->next_pid = next->pid; 470 * __entry->next_prio = next->prio; 471 * ), 472 * 473 * * 474 * * Formatted output of a trace record via TP_printk(). 475 * * This is how the tracepoint will appear under ftrace 476 * * plugins that make use of this tracepoint. 477 * * 478 * * (raw-binary tracing wont actually perform this step.) 479 * * 480 * 481 * TP_printk("task %s:%d [%d] ==> %s:%d [%d]", 482 * __entry->prev_comm, __entry->prev_pid, __entry->prev_prio, 483 * __entry->next_comm, __entry->next_pid, __entry->next_prio), 484 * 485 * ); 486 * 487 * This macro construct is thus used for the regular printk format 488 * tracing setup, it is used to construct a function pointer based 489 * tracepoint callback (this is used by programmatic plugins and 490 * can also by used by generic instrumentation like SystemTap), and 491 * it is also used to expose a structured trace record in 492 * /sys/kernel/debug/tracing/events/. 493 * 494 * A set of (un)registration functions can be passed to the variant 495 * TRACE_EVENT_FN to perform any (un)registration work. 496 */ 497 498 #define DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, print) 499 #define DEFINE_EVENT(template, name, proto, args) \ 500 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 501 #define DEFINE_EVENT_FN(template, name, proto, args, reg, unreg)\ 502 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 503 #define DEFINE_EVENT_PRINT(template, name, proto, args, print) \ 504 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 505 #define DEFINE_EVENT_CONDITION(template, name, proto, \ 506 args, cond) \ 507 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 508 PARAMS(args), PARAMS(cond)) 509 510 #define TRACE_EVENT(name, proto, args, struct, assign, print) \ 511 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 512 #define TRACE_EVENT_FN(name, proto, args, struct, \ 513 assign, print, reg, unreg) \ 514 DECLARE_TRACE(name, PARAMS(proto), PARAMS(args)) 515 #define TRACE_EVENT_FN_COND(name, proto, args, cond, struct, \ 516 assign, print, reg, unreg) \ 517 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 518 PARAMS(args), PARAMS(cond)) 519 #define TRACE_EVENT_CONDITION(name, proto, args, cond, \ 520 struct, assign, print) \ 521 DECLARE_TRACE_CONDITION(name, PARAMS(proto), \ 522 PARAMS(args), PARAMS(cond)) 523 524 #define TRACE_EVENT_FLAGS(event, flag) 525 526 #define TRACE_EVENT_PERF_PERM(event, expr...) 527 528 #define DECLARE_EVENT_NOP(name, proto, args) \ 529 static inline void trace_##name(proto) \ 530 { } \ 531 static inline bool trace_##name##_enabled(void) \ 532 { \ 533 return false; \ 534 } 535 536 #define TRACE_EVENT_NOP(name, proto, args, struct, assign, print) \ 537 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args)) 538 539 #define DECLARE_EVENT_CLASS_NOP(name, proto, args, tstruct, assign, print) 540 #define DEFINE_EVENT_NOP(template, name, proto, args) \ 541 DECLARE_EVENT_NOP(name, PARAMS(proto), PARAMS(args)) 542 543 #endif /* ifdef TRACE_EVENT (see note above) */ 544