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 * __string_len: This is a helper to a __dynamic_array, but it understands 145 * that the array has characters in it, and with the combined 146 * use of __assign_str_len(), it will allocate 'len' + 1 bytes 147 * in the ring buffer and add a '\0' to the string. This is 148 * useful if the string being saved has no terminating '\0' byte. 149 * It requires that the length of the string is known as it acts 150 * like a memcpy(). 151 * 152 * Declared with: 153 * 154 * __string_len(foo, bar, len) 155 * 156 * To assign this string, use the helper macro __assign_str_len(). 157 * 158 * __assign_str_len(foo, bar, len); 159 * 160 * Then len + 1 is allocated to the ring buffer, and a nul terminating 161 * byte is added. This is similar to: 162 * 163 * memcpy(__get_str(foo), bar, len); 164 * __get_str(foo)[len] = 0; 165 * 166 * The advantage of using this over __dynamic_array, is that it 167 * takes care of allocating the extra byte on the ring buffer 168 * for the '\0' terminating byte, and __get_str(foo) can be used 169 * in the TP_printk(). 170 * 171 * __bitmask: This is another kind of __dynamic_array, but it expects 172 * an array of longs, and the number of bits to parse. It takes 173 * two parameters (name, nr_bits), where name is the name of the 174 * bitmask to save, and the nr_bits is the number of bits to record. 175 * 176 * __bitmask(target_cpu, nr_cpumask_bits) 177 * 178 * To assign a bitmask, use the __assign_bitmask() helper macro. 179 * 180 * __assign_bitmask(target_cpus, cpumask_bits(bar), nr_cpumask_bits); 181 * 182 * 183 * fast_assign: This is a C like function that is used to store the items 184 * into the ring buffer. A special variable called "__entry" will be the 185 * structure that points into the ring buffer and has the same fields as 186 * described by the struct part of TRACE_EVENT above. 187 * 188 * printk: This is a way to print out the data in pretty print. This is 189 * useful if the system crashes and you are logging via a serial line, 190 * the data can be printed to the console using this "printk" method. 191 * This is also used to print out the data from the trace files. 192 * Again, the __entry macro is used to access the data from the ring buffer. 193 * 194 * Note, __dynamic_array, __string, and __bitmask require special helpers 195 * to access the data. 196 * 197 * For __dynamic_array(int, foo, bar) use __get_dynamic_array(foo) 198 * Use __get_dynamic_array_len(foo) to get the length of the array 199 * saved. Note, __get_dynamic_array_len() returns the total allocated 200 * length of the dynamic array; __print_array() expects the second 201 * parameter to be the number of elements. To get that, the array length 202 * needs to be divided by the element size. 203 * 204 * For __string(foo, bar) use __get_str(foo) 205 * 206 * For __bitmask(target_cpus, nr_cpumask_bits) use __get_bitmask(target_cpus) 207 * 208 * 209 * Note, that for both the assign and the printk, __entry is the handler 210 * to the data structure in the ring buffer, and is defined by the 211 * TP_STRUCT__entry. 212 */ 213 214 /* 215 * It is OK to have helper functions in the file, but they need to be protected 216 * from being defined more than once. Remember, this file gets included more 217 * than once. 218 */ 219 #ifndef __TRACE_EVENT_SAMPLE_HELPER_FUNCTIONS 220 #define __TRACE_EVENT_SAMPLE_HELPER_FUNCTIONS 221 static inline int __length_of(const int *list) 222 { 223 int i; 224 225 if (!list) 226 return 0; 227 228 for (i = 0; list[i]; i++) 229 ; 230 return i; 231 } 232 233 enum { 234 TRACE_SAMPLE_FOO = 2, 235 TRACE_SAMPLE_BAR = 4, 236 TRACE_SAMPLE_ZOO = 8, 237 }; 238 #endif 239 240 /* 241 * If enums are used in the TP_printk(), their names will be shown in 242 * format files and not their values. This can cause problems with user 243 * space programs that parse the format files to know how to translate 244 * the raw binary trace output into human readable text. 245 * 246 * To help out user space programs, any enum that is used in the TP_printk() 247 * should be defined by TRACE_DEFINE_ENUM() macro. All that is needed to 248 * be done is to add this macro with the enum within it in the trace 249 * header file, and it will be converted in the output. 250 */ 251 252 TRACE_DEFINE_ENUM(TRACE_SAMPLE_FOO); 253 TRACE_DEFINE_ENUM(TRACE_SAMPLE_BAR); 254 TRACE_DEFINE_ENUM(TRACE_SAMPLE_ZOO); 255 256 TRACE_EVENT(foo_bar, 257 258 TP_PROTO(const char *foo, int bar, const int *lst, 259 const char *string, const struct cpumask *mask), 260 261 TP_ARGS(foo, bar, lst, string, mask), 262 263 TP_STRUCT__entry( 264 __array( char, foo, 10 ) 265 __field( int, bar ) 266 __dynamic_array(int, list, __length_of(lst)) 267 __string( str, string ) 268 __bitmask( cpus, num_possible_cpus() ) 269 ), 270 271 TP_fast_assign( 272 strlcpy(__entry->foo, foo, 10); 273 __entry->bar = bar; 274 memcpy(__get_dynamic_array(list), lst, 275 __length_of(lst) * sizeof(int)); 276 __assign_str(str, string); 277 __assign_bitmask(cpus, cpumask_bits(mask), num_possible_cpus()); 278 ), 279 280 TP_printk("foo %s %d %s %s %s %s (%s)", __entry->foo, __entry->bar, 281 282 /* 283 * Notice here the use of some helper functions. This includes: 284 * 285 * __print_symbolic( variable, { value, "string" }, ... ), 286 * 287 * The variable is tested against each value of the { } pair. If 288 * the variable matches one of the values, then it will print the 289 * string in that pair. If non are matched, it returns a string 290 * version of the number (if __entry->bar == 7 then "7" is returned). 291 */ 292 __print_symbolic(__entry->bar, 293 { 0, "zero" }, 294 { TRACE_SAMPLE_FOO, "TWO" }, 295 { TRACE_SAMPLE_BAR, "FOUR" }, 296 { TRACE_SAMPLE_ZOO, "EIGHT" }, 297 { 10, "TEN" } 298 ), 299 300 /* 301 * __print_flags( variable, "delim", { value, "flag" }, ... ), 302 * 303 * This is similar to __print_symbolic, except that it tests the bits 304 * of the value. If ((FLAG & variable) == FLAG) then the string is 305 * printed. If more than one flag matches, then each one that does is 306 * also printed with delim in between them. 307 * If not all bits are accounted for, then the not found bits will be 308 * added in hex format: 0x506 will show BIT2|BIT4|0x500 309 */ 310 __print_flags(__entry->bar, "|", 311 { 1, "BIT1" }, 312 { 2, "BIT2" }, 313 { 4, "BIT3" }, 314 { 8, "BIT4" } 315 ), 316 /* 317 * __print_array( array, len, element_size ) 318 * 319 * This prints out the array that is defined by __array in a nice format. 320 */ 321 __print_array(__get_dynamic_array(list), 322 __get_dynamic_array_len(list) / sizeof(int), 323 sizeof(int)), 324 __get_str(str), __get_bitmask(cpus)) 325 ); 326 327 /* 328 * There may be a case where a tracepoint should only be called if 329 * some condition is set. Otherwise the tracepoint should not be called. 330 * But to do something like: 331 * 332 * if (cond) 333 * trace_foo(); 334 * 335 * Would cause a little overhead when tracing is not enabled, and that 336 * overhead, even if small, is not something we want. As tracepoints 337 * use static branch (aka jump_labels), where no branch is taken to 338 * skip the tracepoint when not enabled, and a jmp is placed to jump 339 * to the tracepoint code when it is enabled, having a if statement 340 * nullifies that optimization. It would be nice to place that 341 * condition within the static branch. This is where TRACE_EVENT_CONDITION 342 * comes in. 343 * 344 * TRACE_EVENT_CONDITION() is just like TRACE_EVENT, except it adds another 345 * parameter just after args. Where TRACE_EVENT has: 346 * 347 * TRACE_EVENT(name, proto, args, struct, assign, printk) 348 * 349 * the CONDITION version has: 350 * 351 * TRACE_EVENT_CONDITION(name, proto, args, cond, struct, assign, printk) 352 * 353 * Everything is the same as TRACE_EVENT except for the new cond. Think 354 * of the cond variable as: 355 * 356 * if (cond) 357 * trace_foo_bar_with_cond(); 358 * 359 * Except that the logic for the if branch is placed after the static branch. 360 * That is, the if statement that processes the condition will not be 361 * executed unless that traecpoint is enabled. Otherwise it still remains 362 * a nop. 363 */ 364 TRACE_EVENT_CONDITION(foo_bar_with_cond, 365 366 TP_PROTO(const char *foo, int bar), 367 368 TP_ARGS(foo, bar), 369 370 TP_CONDITION(!(bar % 10)), 371 372 TP_STRUCT__entry( 373 __string( foo, foo ) 374 __field( int, bar ) 375 ), 376 377 TP_fast_assign( 378 __assign_str(foo, foo); 379 __entry->bar = bar; 380 ), 381 382 TP_printk("foo %s %d", __get_str(foo), __entry->bar) 383 ); 384 385 int foo_bar_reg(void); 386 void foo_bar_unreg(void); 387 388 /* 389 * Now in the case that some function needs to be called when the 390 * tracepoint is enabled and/or when it is disabled, the 391 * TRACE_EVENT_FN() serves this purpose. This is just like TRACE_EVENT() 392 * but adds two more parameters at the end: 393 * 394 * TRACE_EVENT_FN( name, proto, args, struct, assign, printk, reg, unreg) 395 * 396 * reg and unreg are functions with the prototype of: 397 * 398 * void reg(void) 399 * 400 * The reg function gets called before the tracepoint is enabled, and 401 * the unreg function gets called after the tracepoint is disabled. 402 * 403 * Note, reg and unreg are allowed to be NULL. If you only need to 404 * call a function before enabling, or after disabling, just set one 405 * function and pass in NULL for the other parameter. 406 */ 407 TRACE_EVENT_FN(foo_bar_with_fn, 408 409 TP_PROTO(const char *foo, int bar), 410 411 TP_ARGS(foo, bar), 412 413 TP_STRUCT__entry( 414 __string( foo, foo ) 415 __field( int, bar ) 416 ), 417 418 TP_fast_assign( 419 __assign_str(foo, foo); 420 __entry->bar = bar; 421 ), 422 423 TP_printk("foo %s %d", __get_str(foo), __entry->bar), 424 425 foo_bar_reg, foo_bar_unreg 426 ); 427 428 /* 429 * Each TRACE_EVENT macro creates several helper functions to produce 430 * the code to add the tracepoint, create the files in the trace 431 * directory, hook it to perf, assign the values and to print out 432 * the raw data from the ring buffer. To prevent too much bloat, 433 * if there are more than one tracepoint that uses the same format 434 * for the proto, args, struct, assign and printk, and only the name 435 * is different, it is highly recommended to use the DECLARE_EVENT_CLASS 436 * 437 * DECLARE_EVENT_CLASS() macro creates most of the functions for the 438 * tracepoint. Then DEFINE_EVENT() is use to hook a tracepoint to those 439 * functions. This DEFINE_EVENT() is an instance of the class and can 440 * be enabled and disabled separately from other events (either TRACE_EVENT 441 * or other DEFINE_EVENT()s). 442 * 443 * Note, TRACE_EVENT() itself is simply defined as: 444 * 445 * #define TRACE_EVENT(name, proto, args, tstruct, assign, printk) \ 446 * DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, printk); \ 447 * DEFINE_EVENT(name, name, proto, args) 448 * 449 * The DEFINE_EVENT() also can be declared with conditions and reg functions: 450 * 451 * DEFINE_EVENT_CONDITION(template, name, proto, args, cond); 452 * DEFINE_EVENT_FN(template, name, proto, args, reg, unreg); 453 */ 454 DECLARE_EVENT_CLASS(foo_template, 455 456 TP_PROTO(const char *foo, int bar), 457 458 TP_ARGS(foo, bar), 459 460 TP_STRUCT__entry( 461 __string( foo, foo ) 462 __field( int, bar ) 463 ), 464 465 TP_fast_assign( 466 __assign_str(foo, foo); 467 __entry->bar = bar; 468 ), 469 470 TP_printk("foo %s %d", __get_str(foo), __entry->bar) 471 ); 472 473 /* 474 * Here's a better way for the previous samples (except, the first 475 * example had more fields and could not be used here). 476 */ 477 DEFINE_EVENT(foo_template, foo_with_template_simple, 478 TP_PROTO(const char *foo, int bar), 479 TP_ARGS(foo, bar)); 480 481 DEFINE_EVENT_CONDITION(foo_template, foo_with_template_cond, 482 TP_PROTO(const char *foo, int bar), 483 TP_ARGS(foo, bar), 484 TP_CONDITION(!(bar % 8))); 485 486 487 DEFINE_EVENT_FN(foo_template, foo_with_template_fn, 488 TP_PROTO(const char *foo, int bar), 489 TP_ARGS(foo, bar), 490 foo_bar_reg, foo_bar_unreg); 491 492 /* 493 * Anytime two events share basically the same values and have 494 * the same output, use the DECLARE_EVENT_CLASS() and DEFINE_EVENT() 495 * when ever possible. 496 */ 497 498 /* 499 * If the event is similar to the DECLARE_EVENT_CLASS, but you need 500 * to have a different output, then use DEFINE_EVENT_PRINT() which 501 * lets you override the TP_printk() of the class. 502 */ 503 504 DEFINE_EVENT_PRINT(foo_template, foo_with_template_print, 505 TP_PROTO(const char *foo, int bar), 506 TP_ARGS(foo, bar), 507 TP_printk("bar %s %d", __get_str(foo), __entry->bar)); 508 509 /* 510 * There are yet another __rel_loc dynamic data attribute. If you 511 * use __rel_dynamic_array() and __rel_string() etc. macros, you 512 * can use this attribute. There is no difference from the viewpoint 513 * of functionality with/without 'rel' but the encoding is a bit 514 * different. This is expected to be used with user-space event, 515 * there is no reason that the kernel event use this, but only for 516 * testing. 517 */ 518 519 TRACE_EVENT(foo_rel_loc, 520 521 TP_PROTO(const char *foo, int bar, unsigned long *mask), 522 523 TP_ARGS(foo, bar, mask), 524 525 TP_STRUCT__entry( 526 __rel_string( foo, foo ) 527 __field( int, bar ) 528 __rel_bitmask( bitmask, 529 BITS_PER_BYTE * sizeof(unsigned long) ) 530 ), 531 532 TP_fast_assign( 533 __assign_rel_str(foo, foo); 534 __entry->bar = bar; 535 __assign_rel_bitmask(bitmask, mask, 536 BITS_PER_BYTE * sizeof(unsigned long)); 537 ), 538 539 TP_printk("foo_rel_loc %s, %d, %s", __get_rel_str(foo), __entry->bar, 540 __get_rel_bitmask(bitmask)) 541 ); 542 #endif 543 544 /***** NOTICE! The #if protection ends here. *****/ 545 546 547 /* 548 * There are several ways I could have done this. If I left out the 549 * TRACE_INCLUDE_PATH, then it would default to the kernel source 550 * include/trace/events directory. 551 * 552 * I could specify a path from the define_trace.h file back to this 553 * file. 554 * 555 * #define TRACE_INCLUDE_PATH ../../samples/trace_events 556 * 557 * But the safest and easiest way to simply make it use the directory 558 * that the file is in is to add in the Makefile: 559 * 560 * CFLAGS_trace-events-sample.o := -I$(src) 561 * 562 * This will make sure the current path is part of the include 563 * structure for our file so that define_trace.h can find it. 564 * 565 * I could have made only the top level directory the include: 566 * 567 * CFLAGS_trace-events-sample.o := -I$(PWD) 568 * 569 * And then let the path to this directory be the TRACE_INCLUDE_PATH: 570 * 571 * #define TRACE_INCLUDE_PATH samples/trace_events 572 * 573 * But then if something defines "samples" or "trace_events" as a macro 574 * then we could risk that being converted too, and give us an unexpected 575 * result. 576 */ 577 #undef TRACE_INCLUDE_PATH 578 #undef TRACE_INCLUDE_FILE 579 #define TRACE_INCLUDE_PATH . 580 /* 581 * TRACE_INCLUDE_FILE is not needed if the filename and TRACE_SYSTEM are equal 582 */ 583 #define TRACE_INCLUDE_FILE trace-events-sample 584 #include <trace/define_trace.h> 585