1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * If TRACE_SYSTEM is defined, that will be the directory created 4 * in the ftrace directory under /sys/kernel/tracing/events/<system> 5 * 6 * The define_trace.h below will also look for a file name of 7 * TRACE_SYSTEM.h where TRACE_SYSTEM is what is defined here. 8 * In this case, it would look for sample-trace.h 9 * 10 * If the header name will be different than the system name 11 * (as in this case), then you can override the header name that 12 * define_trace.h will look up by defining TRACE_INCLUDE_FILE 13 * 14 * This file is called trace-events-sample.h but we want the system 15 * to be called "sample-trace". Therefore we must define the name of this 16 * file: 17 * 18 * #define TRACE_INCLUDE_FILE trace-events-sample 19 * 20 * As we do an the bottom of this file. 21 * 22 * Notice that TRACE_SYSTEM should be defined outside of #if 23 * protection, just like TRACE_INCLUDE_FILE. 24 */ 25 #undef TRACE_SYSTEM 26 #define TRACE_SYSTEM sample-trace 27 28 /* 29 * TRACE_SYSTEM is expected to be a C valid variable (alpha-numeric 30 * and underscore), although it may start with numbers. If for some 31 * reason it is not, you need to add the following lines: 32 */ 33 #undef TRACE_SYSTEM_VAR 34 #define TRACE_SYSTEM_VAR sample_trace 35 /* 36 * But the above is only needed if TRACE_SYSTEM is not alpha-numeric 37 * and underscored. By default, TRACE_SYSTEM_VAR will be equal to 38 * TRACE_SYSTEM. As TRACE_SYSTEM_VAR must be alpha-numeric, if 39 * TRACE_SYSTEM is not, then TRACE_SYSTEM_VAR must be defined with 40 * only alpha-numeric and underscores. 41 * 42 * The TRACE_SYSTEM_VAR is only used internally and not visible to 43 * user space. 44 */ 45 46 /* 47 * Notice that this file is not protected like a normal header. 48 * We also must allow for rereading of this file. The 49 * 50 * || defined(TRACE_HEADER_MULTI_READ) 51 * 52 * serves this purpose. 53 */ 54 #if !defined(_TRACE_EVENT_SAMPLE_H) || defined(TRACE_HEADER_MULTI_READ) 55 #define _TRACE_EVENT_SAMPLE_H 56 57 /* 58 * All trace headers should include tracepoint.h, until we finally 59 * make it into a standard header. 60 */ 61 #include <linux/tracepoint.h> 62 63 /* 64 * The TRACE_EVENT macro is broken up into 5 parts. 65 * 66 * name: name of the trace point. This is also how to enable the tracepoint. 67 * A function called trace_foo_bar() will be created. 68 * 69 * proto: the prototype of the function trace_foo_bar() 70 * Here it is trace_foo_bar(char *foo, int bar). 71 * 72 * args: must match the arguments in the prototype. 73 * Here it is simply "foo, bar". 74 * 75 * struct: This defines the way the data will be stored in the ring buffer. 76 * The items declared here become part of a special structure 77 * called "__entry", which can be used in the fast_assign part of the 78 * TRACE_EVENT macro. 79 * 80 * Here are the currently defined types you can use: 81 * 82 * __field : Is broken up into type and name. Where type can be any 83 * primitive type (integer, long or pointer). 84 * 85 * __field(int, foo) 86 * 87 * __entry->foo = 5; 88 * 89 * __field_struct : This can be any static complex data type (struct, union 90 * but not an array). Be careful using complex types, as each 91 * event is limited in size, and copying large amounts of data 92 * into the ring buffer can slow things down. 93 * 94 * __field_struct(struct bar, foo) 95 * 96 * __entry->bar.x = y; 97 98 * __array: There are three fields (type, name, size). The type is the 99 * type of elements in the array, the name is the name of the array. 100 * size is the number of items in the array (not the total size). 101 * 102 * __array( char, foo, 10) is the same as saying: char foo[10]; 103 * 104 * Assigning arrays can be done like any array: 105 * 106 * __entry->foo[0] = 'a'; 107 * 108 * memcpy(__entry->foo, bar, 10); 109 * 110 * __dynamic_array: This is similar to array, but can vary its size from 111 * instance to instance of the tracepoint being called. 112 * Like __array, this too has three elements (type, name, size); 113 * type is the type of the element, name is the name of the array. 114 * The size is different than __array. It is not a static number, 115 * but the algorithm to figure out the length of the array for the 116 * specific instance of tracepoint. Again, size is the number of 117 * items in the array, not the total length in bytes. 118 * 119 * __dynamic_array( int, foo, bar) is similar to: int foo[bar]; 120 * 121 * Note, unlike arrays, you must use the __get_dynamic_array() macro 122 * to access the array. 123 * 124 * memcpy(__get_dynamic_array(foo), bar, 10); 125 * 126 * Notice, that "__entry" is not needed here. 127 * 128 * __string: This is a special kind of __dynamic_array. It expects to 129 * have a null terminated character array passed to it (it allows 130 * for NULL too, which would be converted into "(null)"). __string 131 * takes two parameter (name, src), where name is the name of 132 * the string saved, and src is the string to copy into the 133 * ring buffer. 134 * 135 * __string(foo, bar) is similar to: strcpy(foo, bar) 136 * 137 * To assign a string, use the helper macro __assign_str(). 138 * 139 * __assign_str(foo, bar); 140 * 141 * In most cases, the __assign_str() macro will take the same 142 * parameters as the __string() macro had to declare the string. 143 * 144 * __vstring: This is similar to __string() but instead of taking a 145 * dynamic length, it takes a variable list va_list 'va' variable. 146 * Some event callers already have a message from parameters saved 147 * in a va_list. Passing in the format and the va_list variable 148 * will save just enough on the ring buffer for that string. 149 * Note, the va variable used is a pointer to a va_list, not 150 * to the va_list directly. 151 * 152 * (va_list *va) 153 * 154 * __vstring(foo, fmt, va) is similar to: vsnprintf(foo, fmt, va) 155 * 156 * To assign the string, use the helper macro __assign_vstr(). 157 * 158 * __assign_vstr(foo, fmt, va); 159 * 160 * In most cases, the __assign_vstr() macro will take the same 161 * parameters as the __vstring() macro had to declare the string. 162 * Use __get_str() to retrieve the __vstring() just like it would for 163 * __string(). 164 * 165 * __string_len: This is a helper to a __dynamic_array, but it understands 166 * that the array has characters in it, and with the combined 167 * use of __assign_str_len(), it will allocate 'len' + 1 bytes 168 * in the ring buffer and add a '\0' to the string. This is 169 * useful if the string being saved has no terminating '\0' byte. 170 * It requires that the length of the string is known as it acts 171 * like a memcpy(). 172 * 173 * Declared with: 174 * 175 * __string_len(foo, bar, len) 176 * 177 * To assign this string, use the helper macro __assign_str_len(). 178 * 179 * __assign_str_len(foo, bar, len); 180 * 181 * Then len + 1 is allocated to the ring buffer, and a nul terminating 182 * byte is added. This is similar to: 183 * 184 * memcpy(__get_str(foo), bar, len); 185 * __get_str(foo)[len] = 0; 186 * 187 * The advantage of using this over __dynamic_array, is that it 188 * takes care of allocating the extra byte on the ring buffer 189 * for the '\0' terminating byte, and __get_str(foo) can be used 190 * in the TP_printk(). 191 * 192 * __bitmask: This is another kind of __dynamic_array, but it expects 193 * an array of longs, and the number of bits to parse. It takes 194 * two parameters (name, nr_bits), where name is the name of the 195 * bitmask to save, and the nr_bits is the number of bits to record. 196 * 197 * __bitmask(target_cpu, nr_cpumask_bits) 198 * 199 * To assign a bitmask, use the __assign_bitmask() helper macro. 200 * 201 * __assign_bitmask(target_cpus, cpumask_bits(bar), nr_cpumask_bits); 202 * 203 * 204 * fast_assign: This is a C like function that is used to store the items 205 * into the ring buffer. A special variable called "__entry" will be the 206 * structure that points into the ring buffer and has the same fields as 207 * described by the struct part of TRACE_EVENT above. 208 * 209 * printk: This is a way to print out the data in pretty print. This is 210 * useful if the system crashes and you are logging via a serial line, 211 * the data can be printed to the console using this "printk" method. 212 * This is also used to print out the data from the trace files. 213 * Again, the __entry macro is used to access the data from the ring buffer. 214 * 215 * Note, __dynamic_array, __string, and __bitmask require special helpers 216 * to access the data. 217 * 218 * For __dynamic_array(int, foo, bar) use __get_dynamic_array(foo) 219 * Use __get_dynamic_array_len(foo) to get the length of the array 220 * saved. Note, __get_dynamic_array_len() returns the total allocated 221 * length of the dynamic array; __print_array() expects the second 222 * parameter to be the number of elements. To get that, the array length 223 * needs to be divided by the element size. 224 * 225 * For __string(foo, bar) use __get_str(foo) 226 * 227 * For __bitmask(target_cpus, nr_cpumask_bits) use __get_bitmask(target_cpus) 228 * 229 * 230 * Note, that for both the assign and the printk, __entry is the handler 231 * to the data structure in the ring buffer, and is defined by the 232 * TP_STRUCT__entry. 233 */ 234 235 /* 236 * It is OK to have helper functions in the file, but they need to be protected 237 * from being defined more than once. Remember, this file gets included more 238 * than once. 239 */ 240 #ifndef __TRACE_EVENT_SAMPLE_HELPER_FUNCTIONS 241 #define __TRACE_EVENT_SAMPLE_HELPER_FUNCTIONS 242 static inline int __length_of(const int *list) 243 { 244 int i; 245 246 if (!list) 247 return 0; 248 249 for (i = 0; list[i]; i++) 250 ; 251 return i; 252 } 253 254 enum { 255 TRACE_SAMPLE_FOO = 2, 256 TRACE_SAMPLE_BAR = 4, 257 TRACE_SAMPLE_ZOO = 8, 258 }; 259 #endif 260 261 /* 262 * If enums are used in the TP_printk(), their names will be shown in 263 * format files and not their values. This can cause problems with user 264 * space programs that parse the format files to know how to translate 265 * the raw binary trace output into human readable text. 266 * 267 * To help out user space programs, any enum that is used in the TP_printk() 268 * should be defined by TRACE_DEFINE_ENUM() macro. All that is needed to 269 * be done is to add this macro with the enum within it in the trace 270 * header file, and it will be converted in the output. 271 */ 272 273 TRACE_DEFINE_ENUM(TRACE_SAMPLE_FOO); 274 TRACE_DEFINE_ENUM(TRACE_SAMPLE_BAR); 275 TRACE_DEFINE_ENUM(TRACE_SAMPLE_ZOO); 276 277 TRACE_EVENT(foo_bar, 278 279 TP_PROTO(const char *foo, int bar, const int *lst, 280 const char *string, const struct cpumask *mask, 281 const char *fmt, va_list *va), 282 283 TP_ARGS(foo, bar, lst, string, mask, fmt, va), 284 285 TP_STRUCT__entry( 286 __array( char, foo, 10 ) 287 __field( int, bar ) 288 __dynamic_array(int, list, __length_of(lst)) 289 __string( str, string ) 290 __bitmask( cpus, num_possible_cpus() ) 291 __vstring( vstr, fmt, va ) 292 ), 293 294 TP_fast_assign( 295 strlcpy(__entry->foo, foo, 10); 296 __entry->bar = bar; 297 memcpy(__get_dynamic_array(list), lst, 298 __length_of(lst) * sizeof(int)); 299 __assign_str(str, string); 300 __assign_vstr(vstr, fmt, va); 301 __assign_bitmask(cpus, cpumask_bits(mask), num_possible_cpus()); 302 ), 303 304 TP_printk("foo %s %d %s %s %s %s (%s) %s", __entry->foo, __entry->bar, 305 306 /* 307 * Notice here the use of some helper functions. This includes: 308 * 309 * __print_symbolic( variable, { value, "string" }, ... ), 310 * 311 * The variable is tested against each value of the { } pair. If 312 * the variable matches one of the values, then it will print the 313 * string in that pair. If non are matched, it returns a string 314 * version of the number (if __entry->bar == 7 then "7" is returned). 315 */ 316 __print_symbolic(__entry->bar, 317 { 0, "zero" }, 318 { TRACE_SAMPLE_FOO, "TWO" }, 319 { TRACE_SAMPLE_BAR, "FOUR" }, 320 { TRACE_SAMPLE_ZOO, "EIGHT" }, 321 { 10, "TEN" } 322 ), 323 324 /* 325 * __print_flags( variable, "delim", { value, "flag" }, ... ), 326 * 327 * This is similar to __print_symbolic, except that it tests the bits 328 * of the value. If ((FLAG & variable) == FLAG) then the string is 329 * printed. If more than one flag matches, then each one that does is 330 * also printed with delim in between them. 331 * If not all bits are accounted for, then the not found bits will be 332 * added in hex format: 0x506 will show BIT2|BIT4|0x500 333 */ 334 __print_flags(__entry->bar, "|", 335 { 1, "BIT1" }, 336 { 2, "BIT2" }, 337 { 4, "BIT3" }, 338 { 8, "BIT4" } 339 ), 340 /* 341 * __print_array( array, len, element_size ) 342 * 343 * This prints out the array that is defined by __array in a nice format. 344 */ 345 __print_array(__get_dynamic_array(list), 346 __get_dynamic_array_len(list) / sizeof(int), 347 sizeof(int)), 348 __get_str(str), __get_bitmask(cpus), __get_str(vstr)) 349 ); 350 351 /* 352 * There may be a case where a tracepoint should only be called if 353 * some condition is set. Otherwise the tracepoint should not be called. 354 * But to do something like: 355 * 356 * if (cond) 357 * trace_foo(); 358 * 359 * Would cause a little overhead when tracing is not enabled, and that 360 * overhead, even if small, is not something we want. As tracepoints 361 * use static branch (aka jump_labels), where no branch is taken to 362 * skip the tracepoint when not enabled, and a jmp is placed to jump 363 * to the tracepoint code when it is enabled, having a if statement 364 * nullifies that optimization. It would be nice to place that 365 * condition within the static branch. This is where TRACE_EVENT_CONDITION 366 * comes in. 367 * 368 * TRACE_EVENT_CONDITION() is just like TRACE_EVENT, except it adds another 369 * parameter just after args. Where TRACE_EVENT has: 370 * 371 * TRACE_EVENT(name, proto, args, struct, assign, printk) 372 * 373 * the CONDITION version has: 374 * 375 * TRACE_EVENT_CONDITION(name, proto, args, cond, struct, assign, printk) 376 * 377 * Everything is the same as TRACE_EVENT except for the new cond. Think 378 * of the cond variable as: 379 * 380 * if (cond) 381 * trace_foo_bar_with_cond(); 382 * 383 * Except that the logic for the if branch is placed after the static branch. 384 * That is, the if statement that processes the condition will not be 385 * executed unless that traecpoint is enabled. Otherwise it still remains 386 * a nop. 387 */ 388 TRACE_EVENT_CONDITION(foo_bar_with_cond, 389 390 TP_PROTO(const char *foo, int bar), 391 392 TP_ARGS(foo, bar), 393 394 TP_CONDITION(!(bar % 10)), 395 396 TP_STRUCT__entry( 397 __string( foo, foo ) 398 __field( int, bar ) 399 ), 400 401 TP_fast_assign( 402 __assign_str(foo, foo); 403 __entry->bar = bar; 404 ), 405 406 TP_printk("foo %s %d", __get_str(foo), __entry->bar) 407 ); 408 409 int foo_bar_reg(void); 410 void foo_bar_unreg(void); 411 412 /* 413 * Now in the case that some function needs to be called when the 414 * tracepoint is enabled and/or when it is disabled, the 415 * TRACE_EVENT_FN() serves this purpose. This is just like TRACE_EVENT() 416 * but adds two more parameters at the end: 417 * 418 * TRACE_EVENT_FN( name, proto, args, struct, assign, printk, reg, unreg) 419 * 420 * reg and unreg are functions with the prototype of: 421 * 422 * void reg(void) 423 * 424 * The reg function gets called before the tracepoint is enabled, and 425 * the unreg function gets called after the tracepoint is disabled. 426 * 427 * Note, reg and unreg are allowed to be NULL. If you only need to 428 * call a function before enabling, or after disabling, just set one 429 * function and pass in NULL for the other parameter. 430 */ 431 TRACE_EVENT_FN(foo_bar_with_fn, 432 433 TP_PROTO(const char *foo, int bar), 434 435 TP_ARGS(foo, bar), 436 437 TP_STRUCT__entry( 438 __string( foo, foo ) 439 __field( int, bar ) 440 ), 441 442 TP_fast_assign( 443 __assign_str(foo, foo); 444 __entry->bar = bar; 445 ), 446 447 TP_printk("foo %s %d", __get_str(foo), __entry->bar), 448 449 foo_bar_reg, foo_bar_unreg 450 ); 451 452 /* 453 * Each TRACE_EVENT macro creates several helper functions to produce 454 * the code to add the tracepoint, create the files in the trace 455 * directory, hook it to perf, assign the values and to print out 456 * the raw data from the ring buffer. To prevent too much bloat, 457 * if there are more than one tracepoint that uses the same format 458 * for the proto, args, struct, assign and printk, and only the name 459 * is different, it is highly recommended to use the DECLARE_EVENT_CLASS 460 * 461 * DECLARE_EVENT_CLASS() macro creates most of the functions for the 462 * tracepoint. Then DEFINE_EVENT() is use to hook a tracepoint to those 463 * functions. This DEFINE_EVENT() is an instance of the class and can 464 * be enabled and disabled separately from other events (either TRACE_EVENT 465 * or other DEFINE_EVENT()s). 466 * 467 * Note, TRACE_EVENT() itself is simply defined as: 468 * 469 * #define TRACE_EVENT(name, proto, args, tstruct, assign, printk) \ 470 * DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, printk); \ 471 * DEFINE_EVENT(name, name, proto, args) 472 * 473 * The DEFINE_EVENT() also can be declared with conditions and reg functions: 474 * 475 * DEFINE_EVENT_CONDITION(template, name, proto, args, cond); 476 * DEFINE_EVENT_FN(template, name, proto, args, reg, unreg); 477 */ 478 DECLARE_EVENT_CLASS(foo_template, 479 480 TP_PROTO(const char *foo, int bar), 481 482 TP_ARGS(foo, bar), 483 484 TP_STRUCT__entry( 485 __string( foo, foo ) 486 __field( int, bar ) 487 ), 488 489 TP_fast_assign( 490 __assign_str(foo, foo); 491 __entry->bar = bar; 492 ), 493 494 TP_printk("foo %s %d", __get_str(foo), __entry->bar) 495 ); 496 497 /* 498 * Here's a better way for the previous samples (except, the first 499 * example had more fields and could not be used here). 500 */ 501 DEFINE_EVENT(foo_template, foo_with_template_simple, 502 TP_PROTO(const char *foo, int bar), 503 TP_ARGS(foo, bar)); 504 505 DEFINE_EVENT_CONDITION(foo_template, foo_with_template_cond, 506 TP_PROTO(const char *foo, int bar), 507 TP_ARGS(foo, bar), 508 TP_CONDITION(!(bar % 8))); 509 510 511 DEFINE_EVENT_FN(foo_template, foo_with_template_fn, 512 TP_PROTO(const char *foo, int bar), 513 TP_ARGS(foo, bar), 514 foo_bar_reg, foo_bar_unreg); 515 516 /* 517 * Anytime two events share basically the same values and have 518 * the same output, use the DECLARE_EVENT_CLASS() and DEFINE_EVENT() 519 * when ever possible. 520 */ 521 522 /* 523 * If the event is similar to the DECLARE_EVENT_CLASS, but you need 524 * to have a different output, then use DEFINE_EVENT_PRINT() which 525 * lets you override the TP_printk() of the class. 526 */ 527 528 DEFINE_EVENT_PRINT(foo_template, foo_with_template_print, 529 TP_PROTO(const char *foo, int bar), 530 TP_ARGS(foo, bar), 531 TP_printk("bar %s %d", __get_str(foo), __entry->bar)); 532 533 /* 534 * There are yet another __rel_loc dynamic data attribute. If you 535 * use __rel_dynamic_array() and __rel_string() etc. macros, you 536 * can use this attribute. There is no difference from the viewpoint 537 * of functionality with/without 'rel' but the encoding is a bit 538 * different. This is expected to be used with user-space event, 539 * there is no reason that the kernel event use this, but only for 540 * testing. 541 */ 542 543 TRACE_EVENT(foo_rel_loc, 544 545 TP_PROTO(const char *foo, int bar, unsigned long *mask), 546 547 TP_ARGS(foo, bar, mask), 548 549 TP_STRUCT__entry( 550 __rel_string( foo, foo ) 551 __field( int, bar ) 552 __rel_bitmask( bitmask, 553 BITS_PER_BYTE * sizeof(unsigned long) ) 554 ), 555 556 TP_fast_assign( 557 __assign_rel_str(foo, foo); 558 __entry->bar = bar; 559 __assign_rel_bitmask(bitmask, mask, 560 BITS_PER_BYTE * sizeof(unsigned long)); 561 ), 562 563 TP_printk("foo_rel_loc %s, %d, %s", __get_rel_str(foo), __entry->bar, 564 __get_rel_bitmask(bitmask)) 565 ); 566 #endif 567 568 /***** NOTICE! The #if protection ends here. *****/ 569 570 571 /* 572 * There are several ways I could have done this. If I left out the 573 * TRACE_INCLUDE_PATH, then it would default to the kernel source 574 * include/trace/events directory. 575 * 576 * I could specify a path from the define_trace.h file back to this 577 * file. 578 * 579 * #define TRACE_INCLUDE_PATH ../../samples/trace_events 580 * 581 * But the safest and easiest way to simply make it use the directory 582 * that the file is in is to add in the Makefile: 583 * 584 * CFLAGS_trace-events-sample.o := -I$(src) 585 * 586 * This will make sure the current path is part of the include 587 * structure for our file so that define_trace.h can find it. 588 * 589 * I could have made only the top level directory the include: 590 * 591 * CFLAGS_trace-events-sample.o := -I$(PWD) 592 * 593 * And then let the path to this directory be the TRACE_INCLUDE_PATH: 594 * 595 * #define TRACE_INCLUDE_PATH samples/trace_events 596 * 597 * But then if something defines "samples" or "trace_events" as a macro 598 * then we could risk that being converted too, and give us an unexpected 599 * result. 600 */ 601 #undef TRACE_INCLUDE_PATH 602 #undef TRACE_INCLUDE_FILE 603 #define TRACE_INCLUDE_PATH . 604 /* 605 * TRACE_INCLUDE_FILE is not needed if the filename and TRACE_SYSTEM are equal 606 */ 607 #define TRACE_INCLUDE_FILE trace-events-sample 608 #include <trace/define_trace.h> 609