Loading...
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);
140 *
141 * The __string() macro saves off the string that is passed into
142 * the second parameter, and the __assign_str() will store than
143 * saved string into the "foo" field.
144 *
145 * __vstring: This is similar to __string() but instead of taking a
146 * dynamic length, it takes a variable list va_list 'va' variable.
147 * Some event callers already have a message from parameters saved
148 * in a va_list. Passing in the format and the va_list variable
149 * will save just enough on the ring buffer for that string.
150 * Note, the va variable used is a pointer to a va_list, not
151 * to the va_list directly.
152 *
153 * (va_list *va)
154 *
155 * __vstring(foo, fmt, va) is similar to: vsnprintf(foo, fmt, va)
156 *
157 * To assign the string, use the helper macro __assign_vstr().
158 *
159 * __assign_vstr(foo, fmt, va);
160 *
161 * In most cases, the __assign_vstr() macro will take the same
162 * parameters as the __vstring() macro had to declare the string.
163 * Use __get_str() to retrieve the __vstring() just like it would for
164 * __string().
165 *
166 * __string_len: This is a helper to a __dynamic_array, but it understands
167 * that the array has characters in it, 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().
178 * The length is saved via the __string_len() and is retrieved in
179 * __assign_str().
180 *
181 * __assign_str(foo);
182 *
183 * Then len + 1 is allocated to the ring buffer, and a nul terminating
184 * byte is added. This is similar to:
185 *
186 * memcpy(__get_str(foo), bar, len);
187 * __get_str(foo)[len] = 0;
188 *
189 * The advantage of using this over __dynamic_array, is that it
190 * takes care of allocating the extra byte on the ring buffer
191 * for the '\0' terminating byte, and __get_str(foo) can be used
192 * in the TP_printk().
193 *
194 * __bitmask: This is another kind of __dynamic_array, but it expects
195 * an array of longs, and the number of bits to parse. It takes
196 * two parameters (name, nr_bits), where name is the name of the
197 * bitmask to save, and the nr_bits is the number of bits to record.
198 *
199 * __bitmask(target_cpu, nr_cpumask_bits)
200 *
201 * To assign a bitmask, use the __assign_bitmask() helper macro.
202 *
203 * __assign_bitmask(target_cpus, cpumask_bits(bar), nr_cpumask_bits);
204 *
205 * __cpumask: This is pretty much the same as __bitmask but is specific for
206 * CPU masks. The type displayed to the user via the format files will
207 * be "cpumaks_t" such that user space may deal with them differently
208 * if they choose to do so, and the bits is always set to nr_cpumask_bits.
209 *
210 * __cpumask(target_cpu)
211 *
212 * To assign a cpumask, use the __assign_cpumask() helper macro.
213 *
214 * __assign_cpumask(target_cpus, cpumask_bits(bar));
215 *
216 * fast_assign: This is a C like function that is used to store the items
217 * into the ring buffer. A special variable called "__entry" will be the
218 * structure that points into the ring buffer and has the same fields as
219 * described by the struct part of TRACE_EVENT above.
220 *
221 * printk: This is a way to print out the data in pretty print. This is
222 * useful if the system crashes and you are logging via a serial line,
223 * the data can be printed to the console using this "printk" method.
224 * This is also used to print out the data from the trace files.
225 * Again, the __entry macro is used to access the data from the ring buffer.
226 *
227 * Note, __dynamic_array, __string, __bitmask and __cpumask require special
228 * helpers to access the data.
229 *
230 * For __dynamic_array(int, foo, bar) use __get_dynamic_array(foo)
231 * Use __get_dynamic_array_len(foo) to get the length of the array
232 * saved. Note, __get_dynamic_array_len() returns the total allocated
233 * length of the dynamic array; __print_array() expects the second
234 * parameter to be the number of elements. To get that, the array length
235 * needs to be divided by the element size.
236 *
237 * For __string(foo, bar) use __get_str(foo)
238 *
239 * For __bitmask(target_cpus, nr_cpumask_bits) use __get_bitmask(target_cpus)
240 *
241 * For __cpumask(target_cpus) use __get_cpumask(target_cpus)
242 *
243 *
244 * Note, that for both the assign and the printk, __entry is the handler
245 * to the data structure in the ring buffer, and is defined by the
246 * TP_STRUCT__entry.
247 */
248
249/*
250 * It is OK to have helper functions in the file, but they need to be protected
251 * from being defined more than once. Remember, this file gets included more
252 * than once.
253 */
254#ifndef __TRACE_EVENT_SAMPLE_HELPER_FUNCTIONS
255#define __TRACE_EVENT_SAMPLE_HELPER_FUNCTIONS
256static inline int __length_of(const int *list)
257{
258 int i;
259
260 if (!list)
261 return 0;
262
263 for (i = 0; list[i]; i++)
264 ;
265 return i;
266}
267
268enum {
269 TRACE_SAMPLE_FOO = 2,
270 TRACE_SAMPLE_BAR = 4,
271 TRACE_SAMPLE_ZOO = 8,
272};
273#endif
274
275/*
276 * If enums are used in the TP_printk(), their names will be shown in
277 * format files and not their values. This can cause problems with user
278 * space programs that parse the format files to know how to translate
279 * the raw binary trace output into human readable text.
280 *
281 * To help out user space programs, any enum that is used in the TP_printk()
282 * should be defined by TRACE_DEFINE_ENUM() macro. All that is needed to
283 * be done is to add this macro with the enum within it in the trace
284 * header file, and it will be converted in the output.
285 */
286
287TRACE_DEFINE_ENUM(TRACE_SAMPLE_FOO);
288TRACE_DEFINE_ENUM(TRACE_SAMPLE_BAR);
289TRACE_DEFINE_ENUM(TRACE_SAMPLE_ZOO);
290
291TRACE_EVENT(foo_bar,
292
293 TP_PROTO(const char *foo, int bar, const int *lst,
294 const char *string, const struct cpumask *mask,
295 const char *fmt, va_list *va),
296
297 TP_ARGS(foo, bar, lst, string, mask, fmt, va),
298
299 TP_STRUCT__entry(
300 __array( char, foo, 10 )
301 __field( int, bar )
302 __dynamic_array(int, list, __length_of(lst))
303 __string( str, string )
304 __bitmask( cpus, num_possible_cpus() )
305 __cpumask( cpum )
306 __vstring( vstr, fmt, va )
307 __string_len( lstr, foo, bar / 2 < strlen(foo) ? bar / 2 : strlen(foo) )
308 ),
309
310 TP_fast_assign(
311 strscpy(__entry->foo, foo, 10);
312 __entry->bar = bar;
313 memcpy(__get_dynamic_array(list), lst,
314 __length_of(lst) * sizeof(int));
315 __assign_str(str);
316 __assign_str(lstr);
317 __assign_vstr(vstr, fmt, va);
318 __assign_bitmask(cpus, cpumask_bits(mask), num_possible_cpus());
319 __assign_cpumask(cpum, cpumask_bits(mask));
320 ),
321
322 TP_printk("foo %s %d %s %s %s %s %s %s (%s) (%s) %s", __entry->foo, __entry->bar,
323
324/*
325 * Notice here the use of some helper functions. This includes:
326 *
327 * __print_symbolic( variable, { value, "string" }, ... ),
328 *
329 * The variable is tested against each value of the { } pair. If
330 * the variable matches one of the values, then it will print the
331 * string in that pair. If non are matched, it returns a string
332 * version of the number (if __entry->bar == 7 then "7" is returned).
333 */
334 __print_symbolic(__entry->bar,
335 { 0, "zero" },
336 { TRACE_SAMPLE_FOO, "TWO" },
337 { TRACE_SAMPLE_BAR, "FOUR" },
338 { TRACE_SAMPLE_ZOO, "EIGHT" },
339 { 10, "TEN" }
340 ),
341
342/*
343 * __print_flags( variable, "delim", { value, "flag" }, ... ),
344 *
345 * This is similar to __print_symbolic, except that it tests the bits
346 * of the value. If ((FLAG & variable) == FLAG) then the string is
347 * printed. If more than one flag matches, then each one that does is
348 * also printed with delim in between them.
349 * If not all bits are accounted for, then the not found bits will be
350 * added in hex format: 0x506 will show BIT2|BIT4|0x500
351 */
352 __print_flags(__entry->bar, "|",
353 { 1, "BIT1" },
354 { 2, "BIT2" },
355 { 4, "BIT3" },
356 { 8, "BIT4" }
357 ),
358/*
359 * __print_array( array, len, element_size )
360 *
361 * This prints out the array that is defined by __array in a nice format.
362 */
363 __print_array(__get_dynamic_array(list),
364 __get_dynamic_array_len(list) / sizeof(int),
365 sizeof(int)),
366
367/* A shortcut is to use __print_dynamic_array for dynamic arrays */
368
369 __print_dynamic_array(list, sizeof(int)),
370
371 __get_str(str), __get_str(lstr),
372 __get_bitmask(cpus), __get_cpumask(cpum),
373 __get_str(vstr))
374);
375
376/*
377 * There may be a case where a tracepoint should only be called if
378 * some condition is set. Otherwise the tracepoint should not be called.
379 * But to do something like:
380 *
381 * if (cond)
382 * trace_foo();
383 *
384 * Would cause a little overhead when tracing is not enabled, and that
385 * overhead, even if small, is not something we want. As tracepoints
386 * use static branch (aka jump_labels), where no branch is taken to
387 * skip the tracepoint when not enabled, and a jmp is placed to jump
388 * to the tracepoint code when it is enabled, having a if statement
389 * nullifies that optimization. It would be nice to place that
390 * condition within the static branch. This is where TRACE_EVENT_CONDITION
391 * comes in.
392 *
393 * TRACE_EVENT_CONDITION() is just like TRACE_EVENT, except it adds another
394 * parameter just after args. Where TRACE_EVENT has:
395 *
396 * TRACE_EVENT(name, proto, args, struct, assign, printk)
397 *
398 * the CONDITION version has:
399 *
400 * TRACE_EVENT_CONDITION(name, proto, args, cond, struct, assign, printk)
401 *
402 * Everything is the same as TRACE_EVENT except for the new cond. Think
403 * of the cond variable as:
404 *
405 * if (cond)
406 * trace_foo_bar_with_cond();
407 *
408 * Except that the logic for the if branch is placed after the static branch.
409 * That is, the if statement that processes the condition will not be
410 * executed unless that traecpoint is enabled. Otherwise it still remains
411 * a nop.
412 */
413TRACE_EVENT_CONDITION(foo_bar_with_cond,
414
415 TP_PROTO(const char *foo, int bar),
416
417 TP_ARGS(foo, bar),
418
419 TP_CONDITION(!(bar % 10)),
420
421 TP_STRUCT__entry(
422 __string( foo, foo )
423 __field( int, bar )
424 ),
425
426 TP_fast_assign(
427 __assign_str(foo);
428 __entry->bar = bar;
429 ),
430
431 TP_printk("foo %s %d", __get_str(foo), __entry->bar)
432);
433
434int foo_bar_reg(void);
435void foo_bar_unreg(void);
436
437/*
438 * Now in the case that some function needs to be called when the
439 * tracepoint is enabled and/or when it is disabled, the
440 * TRACE_EVENT_FN() serves this purpose. This is just like TRACE_EVENT()
441 * but adds two more parameters at the end:
442 *
443 * TRACE_EVENT_FN( name, proto, args, struct, assign, printk, reg, unreg)
444 *
445 * reg and unreg are functions with the prototype of:
446 *
447 * void reg(void)
448 *
449 * The reg function gets called before the tracepoint is enabled, and
450 * the unreg function gets called after the tracepoint is disabled.
451 *
452 * Note, reg and unreg are allowed to be NULL. If you only need to
453 * call a function before enabling, or after disabling, just set one
454 * function and pass in NULL for the other parameter.
455 */
456TRACE_EVENT_FN(foo_bar_with_fn,
457
458 TP_PROTO(const char *foo, int bar),
459
460 TP_ARGS(foo, bar),
461
462 TP_STRUCT__entry(
463 __string( foo, foo )
464 __field( int, bar )
465 ),
466
467 TP_fast_assign(
468 __assign_str(foo);
469 __entry->bar = bar;
470 ),
471
472 TP_printk("foo %s %d", __get_str(foo), __entry->bar),
473
474 foo_bar_reg, foo_bar_unreg
475);
476
477/*
478 * Each TRACE_EVENT macro creates several helper functions to produce
479 * the code to add the tracepoint, create the files in the trace
480 * directory, hook it to perf, assign the values and to print out
481 * the raw data from the ring buffer. To prevent too much bloat,
482 * if there are more than one tracepoint that uses the same format
483 * for the proto, args, struct, assign and printk, and only the name
484 * is different, it is highly recommended to use the DECLARE_EVENT_CLASS
485 *
486 * DECLARE_EVENT_CLASS() macro creates most of the functions for the
487 * tracepoint. Then DEFINE_EVENT() is use to hook a tracepoint to those
488 * functions. This DEFINE_EVENT() is an instance of the class and can
489 * be enabled and disabled separately from other events (either TRACE_EVENT
490 * or other DEFINE_EVENT()s).
491 *
492 * Note, TRACE_EVENT() itself is simply defined as:
493 *
494 * #define TRACE_EVENT(name, proto, args, tstruct, assign, printk) \
495 * DECLARE_EVENT_CLASS(name, proto, args, tstruct, assign, printk); \
496 * DEFINE_EVENT(name, name, proto, args)
497 *
498 * The DEFINE_EVENT() also can be declared with conditions and reg functions:
499 *
500 * DEFINE_EVENT_CONDITION(template, name, proto, args, cond);
501 * DEFINE_EVENT_FN(template, name, proto, args, reg, unreg);
502 */
503DECLARE_EVENT_CLASS(foo_template,
504
505 TP_PROTO(const char *foo, int bar),
506
507 TP_ARGS(foo, bar),
508
509 TP_STRUCT__entry(
510 __string( foo, foo )
511 __field( int, bar )
512 ),
513
514 TP_fast_assign(
515 __assign_str(foo);
516 __entry->bar = bar;
517 ),
518
519 TP_printk("foo %s %d", __get_str(foo), __entry->bar)
520);
521
522/*
523 * Here's a better way for the previous samples (except, the first
524 * example had more fields and could not be used here).
525 */
526DEFINE_EVENT(foo_template, foo_with_template_simple,
527 TP_PROTO(const char *foo, int bar),
528 TP_ARGS(foo, bar));
529
530DEFINE_EVENT_CONDITION(foo_template, foo_with_template_cond,
531 TP_PROTO(const char *foo, int bar),
532 TP_ARGS(foo, bar),
533 TP_CONDITION(!(bar % 8)));
534
535
536DEFINE_EVENT_FN(foo_template, foo_with_template_fn,
537 TP_PROTO(const char *foo, int bar),
538 TP_ARGS(foo, bar),
539 foo_bar_reg, foo_bar_unreg);
540
541/*
542 * Anytime two events share basically the same values and have
543 * the same output, use the DECLARE_EVENT_CLASS() and DEFINE_EVENT()
544 * when ever possible.
545 */
546
547/*
548 * If the event is similar to the DECLARE_EVENT_CLASS, but you need
549 * to have a different output, then use DEFINE_EVENT_PRINT() which
550 * lets you override the TP_printk() of the class.
551 */
552
553DEFINE_EVENT_PRINT(foo_template, foo_with_template_print,
554 TP_PROTO(const char *foo, int bar),
555 TP_ARGS(foo, bar),
556 TP_printk("bar %s %d", __get_str(foo), __entry->bar));
557
558/*
559 * There are yet another __rel_loc dynamic data attribute. If you
560 * use __rel_dynamic_array() and __rel_string() etc. macros, you
561 * can use this attribute. There is no difference from the viewpoint
562 * of functionality with/without 'rel' but the encoding is a bit
563 * different. This is expected to be used with user-space event,
564 * there is no reason that the kernel event use this, but only for
565 * testing.
566 */
567
568TRACE_EVENT(foo_rel_loc,
569
570 TP_PROTO(const char *foo, int bar, unsigned long *mask, const cpumask_t *cpus),
571
572 TP_ARGS(foo, bar, mask, cpus),
573
574 TP_STRUCT__entry(
575 __rel_string( foo, foo )
576 __field( int, bar )
577 __rel_bitmask( bitmask,
578 BITS_PER_BYTE * sizeof(unsigned long) )
579 __rel_cpumask( cpumask )
580 ),
581
582 TP_fast_assign(
583 __assign_rel_str(foo);
584 __entry->bar = bar;
585 __assign_rel_bitmask(bitmask, mask,
586 BITS_PER_BYTE * sizeof(unsigned long));
587 __assign_rel_cpumask(cpumask, cpus);
588 ),
589
590 TP_printk("foo_rel_loc %s, %d, %s, %s", __get_rel_str(foo), __entry->bar,
591 __get_rel_bitmask(bitmask),
592 __get_rel_cpumask(cpumask))
593);
594#endif
595
596/***** NOTICE! The #if protection ends here. *****/
597
598
599/*
600 * There are several ways I could have done this. If I left out the
601 * TRACE_INCLUDE_PATH, then it would default to the kernel source
602 * include/trace/events directory.
603 *
604 * I could specify a path from the define_trace.h file back to this
605 * file.
606 *
607 * #define TRACE_INCLUDE_PATH ../../samples/trace_events
608 *
609 * But the safest and easiest way to simply make it use the directory
610 * that the file is in is to add in the Makefile:
611 *
612 * CFLAGS_trace-events-sample.o := -I$(src)
613 *
614 * This will make sure the current path is part of the include
615 * structure for our file so that define_trace.h can find it.
616 *
617 * I could have made only the top level directory the include:
618 *
619 * CFLAGS_trace-events-sample.o := -I$(PWD)
620 *
621 * And then let the path to this directory be the TRACE_INCLUDE_PATH:
622 *
623 * #define TRACE_INCLUDE_PATH samples/trace_events
624 *
625 * But then if something defines "samples" or "trace_events" as a macro
626 * then we could risk that being converted too, and give us an unexpected
627 * result.
628 */
629#undef TRACE_INCLUDE_PATH
630#undef TRACE_INCLUDE_FILE
631#define TRACE_INCLUDE_PATH .
632/*
633 * TRACE_INCLUDE_FILE is not needed if the filename and TRACE_SYSTEM are equal
634 */
635#define TRACE_INCLUDE_FILE trace-events-sample
636#include <trace/define_trace.h>
1/*
2 * If TRACE_SYSTEM is defined, that will be the directory created
3 * in the ftrace directory under /sys/kernel/tracing/events/<system>
4 *
5 * The define_trace.h below will also look for a file name of
6 * TRACE_SYSTEM.h where TRACE_SYSTEM is what is defined here.
7 * In this case, it would look for sample-trace.h
8 *
9 * If the header name will be different than the system name
10 * (as in this case), then you can override the header name that
11 * define_trace.h will look up by defining TRACE_INCLUDE_FILE
12 *
13 * This file is called trace-events-sample.h but we want the system
14 * to be called "sample-trace". Therefore we must define the name of this
15 * file:
16 *
17 * #define TRACE_INCLUDE_FILE trace-events-sample
18 *
19 * As we do an the bottom of this file.
20 *
21 * Notice that TRACE_SYSTEM should be defined outside of #if
22 * protection, just like TRACE_INCLUDE_FILE.
23 */
24#undef TRACE_SYSTEM
25#define TRACE_SYSTEM sample-trace
26
27/*
28 * TRACE_SYSTEM is expected to be a C valid variable (alpha-numeric
29 * and underscore), although it may start with numbers. If for some
30 * reason it is not, you need to add the following lines:
31 */
32#undef TRACE_SYSTEM_VAR
33#define TRACE_SYSTEM_VAR sample_trace
34/*
35 * But the above is only needed if TRACE_SYSTEM is not alpha-numeric
36 * and underscored. By default, TRACE_SYSTEM_VAR will be equal to
37 * TRACE_SYSTEM. As TRACE_SYSTEM_VAR must be alpha-numeric, if
38 * TRACE_SYSTEM is not, then TRACE_SYSTEM_VAR must be defined with
39 * only alpha-numeric and underscores.
40 *
41 * The TRACE_SYSTEM_VAR is only used internally and not visible to
42 * user space.
43 */
44
45/*
46 * Notice that this file is not protected like a normal header.
47 * We also must allow for rereading of this file. The
48 *
49 * || defined(TRACE_HEADER_MULTI_READ)
50 *
51 * serves this purpose.
52 */
53#if !defined(_TRACE_EVENT_SAMPLE_H) || defined(TRACE_HEADER_MULTI_READ)
54#define _TRACE_EVENT_SAMPLE_H
55
56/*
57 * All trace headers should include tracepoint.h, until we finally
58 * make it into a standard header.
59 */
60#include <linux/tracepoint.h>
61
62/*
63 * The TRACE_EVENT macro is broken up into 5 parts.
64 *
65 * name: name of the trace point. This is also how to enable the tracepoint.
66 * A function called trace_foo_bar() will be created.
67 *
68 * proto: the prototype of the function trace_foo_bar()
69 * Here it is trace_foo_bar(char *foo, int bar).
70 *
71 * args: must match the arguments in the prototype.
72 * Here it is simply "foo, bar".
73 *
74 * struct: This defines the way the data will be stored in the ring buffer.
75 * The items declared here become part of a special structure
76 * called "__entry", which can be used in the fast_assign part of the
77 * TRACE_EVENT macro.
78 *
79 * Here are the currently defined types you can use:
80 *
81 * __field : Is broken up into type and name. Where type can be any
82 * primitive type (integer, long or pointer).
83 *
84 * __field(int, foo)
85 *
86 * __entry->foo = 5;
87 *
88 * __field_struct : This can be any static complex data type (struct, union
89 * but not an array). Be careful using complex types, as each
90 * event is limited in size, and copying large amounts of data
91 * into the ring buffer can slow things down.
92 *
93 * __field_struct(struct bar, foo)
94 *
95 * __entry->bar.x = y;
96
97 * __array: There are three fields (type, name, size). The type is the
98 * type of elements in teh array, the name is the name of the array.
99 * size is the number of items in the array (not the total size).
100 *
101 * __array( char, foo, 10) is the same as saying: char foo[10];
102 *
103 * Assigning arrays can be done like any array:
104 *
105 * __entry->foo[0] = 'a';
106 *
107 * memcpy(__entry->foo, bar, 10);
108 *
109 * __dynamic_array: This is similar to array, but can vary its size from
110 * instance to instance of the tracepoint being called.
111 * Like __array, this too has three elements (type, name, size);
112 * type is the type of the element, name is the name of the array.
113 * The size is different than __array. It is not a static number,
114 * but the algorithm to figure out the length of the array for the
115 * specific instance of tracepoint. Again, size is the numebr of
116 * items in the array, not the total length in bytes.
117 *
118 * __dynamic_array( int, foo, bar) is similar to: int foo[bar];
119 *
120 * Note, unlike arrays, you must use the __get_dynamic_array() macro
121 * to access the array.
122 *
123 * memcpy(__get_dynamic_array(foo), bar, 10);
124 *
125 * Notice, that "__entry" is not needed here.
126 *
127 * __string: This is a special kind of __dynamic_array. It expects to
128 * have a nul terminated character array passed to it (it allows
129 * for NULL too, which would be converted into "(null)"). __string
130 * takes two paramenter (name, src), where name is the name of
131 * the string saved, and src is the string to copy into the
132 * ring buffer.
133 *
134 * __string(foo, bar) is similar to: strcpy(foo, bar)
135 *
136 * To assign a string, use the helper macro __assign_str().
137 *
138 * __assign_str(foo, bar);
139 *
140 * In most cases, the __assign_str() macro will take the same
141 * parameters as the __string() macro had to declare the string.
142 *
143 * __bitmask: This is another kind of __dynamic_array, but it expects
144 * an array of longs, and the number of bits to parse. It takes
145 * two parameters (name, nr_bits), where name is the name of the
146 * bitmask to save, and the nr_bits is the number of bits to record.
147 *
148 * __bitmask(target_cpu, nr_cpumask_bits)
149 *
150 * To assign a bitmask, use the __assign_bitmask() helper macro.
151 *
152 * __assign_bitmask(target_cpus, cpumask_bits(bar), nr_cpumask_bits);
153 *
154 *
155 * fast_assign: This is a C like function that is used to store the items
156 * into the ring buffer. A special variable called "__entry" will be the
157 * structure that points into the ring buffer and has the same fields as
158 * described by the struct part of TRACE_EVENT above.
159 *
160 * printk: This is a way to print out the data in pretty print. This is
161 * useful if the system crashes and you are logging via a serial line,
162 * the data can be printed to the console using this "printk" method.
163 * This is also used to print out the data from the trace files.
164 * Again, the __entry macro is used to access the data from the ring buffer.
165 *
166 * Note, __dynamic_array, __string, and __bitmask require special helpers
167 * to access the data.
168 *
169 * For __dynamic_array(int, foo, bar) use __get_dynamic_array(foo)
170 * Use __get_dynamic_array_len(foo) to get the length of the array
171 * saved. Note, __get_dynamic_array_len() returns the total allocated
172 * length of the dynamic array; __print_array() expects the second
173 * parameter to be the number of elements. To get that, the array length
174 * needs to be divided by the element size.
175 *
176 * For __string(foo, bar) use __get_str(foo)
177 *
178 * For __bitmask(target_cpus, nr_cpumask_bits) use __get_bitmask(target_cpus)
179 *
180 *
181 * Note, that for both the assign and the printk, __entry is the handler
182 * to the data structure in the ring buffer, and is defined by the
183 * TP_STRUCT__entry.
184 */
185
186/*
187 * It is OK to have helper functions in the file, but they need to be protected
188 * from being defined more than once. Remember, this file gets included more
189 * than once.
190 */
191#ifndef __TRACE_EVENT_SAMPLE_HELPER_FUNCTIONS
192#define __TRACE_EVENT_SAMPLE_HELPER_FUNCTIONS
193static inline int __length_of(const int *list)
194{
195 int i;
196
197 if (!list)
198 return 0;
199
200 for (i = 0; list[i]; i++)
201 ;
202 return i;
203}
204
205enum {
206 TRACE_SAMPLE_FOO = 2,
207 TRACE_SAMPLE_BAR = 4,
208 TRACE_SAMPLE_ZOO = 8,
209};
210#endif
211
212/*
213 * If enums are used in the TP_printk(), their names will be shown in
214 * format files and not their values. This can cause problems with user
215 * space programs that parse the format files to know how to translate
216 * the raw binary trace output into human readable text.
217 *
218 * To help out user space programs, any enum that is used in the TP_printk()
219 * should be defined by TRACE_DEFINE_ENUM() macro. All that is needed to
220 * be done is to add this macro with the enum within it in the trace
221 * header file, and it will be converted in the output.
222 */
223
224TRACE_DEFINE_ENUM(TRACE_SAMPLE_FOO);
225TRACE_DEFINE_ENUM(TRACE_SAMPLE_BAR);
226TRACE_DEFINE_ENUM(TRACE_SAMPLE_ZOO);
227
228TRACE_EVENT(foo_bar,
229
230 TP_PROTO(const char *foo, int bar, const int *lst,
231 const char *string, const struct cpumask *mask),
232
233 TP_ARGS(foo, bar, lst, string, mask),
234
235 TP_STRUCT__entry(
236 __array( char, foo, 10 )
237 __field( int, bar )
238 __dynamic_array(int, list, __length_of(lst))
239 __string( str, string )
240 __bitmask( cpus, num_possible_cpus() )
241 ),
242
243 TP_fast_assign(
244 strlcpy(__entry->foo, foo, 10);
245 __entry->bar = bar;
246 memcpy(__get_dynamic_array(list), lst,
247 __length_of(lst) * sizeof(int));
248 __assign_str(str, string);
249 __assign_bitmask(cpus, cpumask_bits(mask), num_possible_cpus());
250 ),
251
252 TP_printk("foo %s %d %s %s %s %s (%s)", __entry->foo, __entry->bar,
253
254/*
255 * Notice here the use of some helper functions. This includes:
256 *
257 * __print_symbolic( variable, { value, "string" }, ... ),
258 *
259 * The variable is tested against each value of the { } pair. If
260 * the variable matches one of the values, then it will print the
261 * string in that pair. If non are matched, it returns a string
262 * version of the number (if __entry->bar == 7 then "7" is returned).
263 */
264 __print_symbolic(__entry->bar,
265 { 0, "zero" },
266 { TRACE_SAMPLE_FOO, "TWO" },
267 { TRACE_SAMPLE_BAR, "FOUR" },
268 { TRACE_SAMPLE_ZOO, "EIGHT" },
269 { 10, "TEN" }
270 ),
271
272/*
273 * __print_flags( variable, "delim", { value, "flag" }, ... ),
274 *
275 * This is similar to __print_symbolic, except that it tests the bits
276 * of the value. If ((FLAG & variable) == FLAG) then the string is
277 * printed. If more than one flag matches, then each one that does is
278 * also printed with delim in between them.
279 * If not all bits are accounted for, then the not found bits will be
280 * added in hex format: 0x506 will show BIT2|BIT4|0x500
281 */
282 __print_flags(__entry->bar, "|",
283 { 1, "BIT1" },
284 { 2, "BIT2" },
285 { 4, "BIT3" },
286 { 8, "BIT4" }
287 ),
288/*
289 * __print_array( array, len, element_size )
290 *
291 * This prints out the array that is defined by __array in a nice format.
292 */
293 __print_array(__get_dynamic_array(list),
294 __get_dynamic_array_len(list) / sizeof(int),
295 sizeof(int)),
296 __get_str(str), __get_bitmask(cpus))
297);
298
299/*
300 * There may be a case where a tracepoint should only be called if
301 * some condition is set. Otherwise the tracepoint should not be called.
302 * But to do something like:
303 *
304 * if (cond)
305 * trace_foo();
306 *
307 * Would cause a little overhead when tracing is not enabled, and that
308 * overhead, even if small, is not something we want. As tracepoints
309 * use static branch (aka jump_labels), where no branch is taken to
310 * skip the tracepoint when not enabled, and a jmp is placed to jump
311 * to the tracepoint code when it is enabled, having a if statement
312 * nullifies that optimization. It would be nice to place that
313 * condition within the static branch. This is where TRACE_EVENT_CONDITION
314 * comes in.
315 *
316 * TRACE_EVENT_CONDITION() is just like TRACE_EVENT, except it adds another
317 * parameter just after args. Where TRACE_EVENT has:
318 *
319 * TRACE_EVENT(name, proto, args, struct, assign, printk)
320 *
321 * the CONDITION version has:
322 *
323 * TRACE_EVENT_CONDITION(name, proto, args, cond, struct, assign, printk)
324 *
325 * Everything is the same as TRACE_EVENT except for the new cond. Think
326 * of the cond variable as:
327 *
328 * if (cond)
329 * trace_foo_bar_with_cond();
330 *
331 * Except that the logic for the if branch is placed after the static branch.
332 * That is, the if statement that processes the condition will not be
333 * executed unless that traecpoint is enabled. Otherwise it still remains
334 * a nop.
335 */
336TRACE_EVENT_CONDITION(foo_bar_with_cond,
337
338 TP_PROTO(const char *foo, int bar),
339
340 TP_ARGS(foo, bar),
341
342 TP_CONDITION(!(bar % 10)),
343
344 TP_STRUCT__entry(
345 __string( foo, foo )
346 __field( int, bar )
347 ),
348
349 TP_fast_assign(
350 __assign_str(foo, foo);
351 __entry->bar = bar;
352 ),
353
354 TP_printk("foo %s %d", __get_str(foo), __entry->bar)
355);
356
357int foo_bar_reg(void);
358void foo_bar_unreg(void);
359
360/*
361 * Now in the case that some function needs to be called when the
362 * tracepoint is enabled and/or when it is disabled, the
363 * TRACE_EVENT_FN() serves this purpose. This is just like TRACE_EVENT()
364 * but adds two more parameters at the end:
365 *
366 * TRACE_EVENT_FN( name, proto, args, struct, assign, printk, reg, unreg)
367 *
368 * reg and unreg are functions with the prototype of:
369 *
370 * void reg(void)
371 *
372 * The reg function gets called before the tracepoint is enabled, and
373 * the unreg function gets called after the tracepoint is disabled.
374 *
375 * Note, reg and unreg are allowed to be NULL. If you only need to
376 * call a function before enabling, or after disabling, just set one
377 * function and pass in NULL for the other parameter.
378 */
379TRACE_EVENT_FN(foo_bar_with_fn,
380
381 TP_PROTO(const char *foo, int bar),
382
383 TP_ARGS(foo, bar),
384
385 TP_STRUCT__entry(
386 __string( foo, foo )
387 __field( int, bar )
388 ),
389
390 TP_fast_assign(
391 __assign_str(foo, foo);
392 __entry->bar = bar;
393 ),
394
395 TP_printk("foo %s %d", __get_str(foo), __entry->bar),
396
397 foo_bar_reg, foo_bar_unreg
398);
399
400/*
401 * Each TRACE_EVENT macro creates several helper functions to produce
402 * the code to add the tracepoint, create the files in the trace
403 * directory, hook it to perf, assign the values and to print out
404 * the raw data from the ring buffer. To prevent too much bloat,
405 * if there are more than one tracepoint that uses the same format
406 * for the proto, args, struct, assign and printk, and only the name
407 * is different, it is highly recommended to use the DECLARE_EVENT_CLASS
408 *
409 * DECLARE_EVENT_CLASS() macro creates most of the functions for the
410 * tracepoint. Then DEFINE_EVENT() is use to hook a tracepoint to those
411 * functions. This DEFINE_EVENT() is an instance of the class and can
412 * be enabled and disabled separately from other events (either TRACE_EVENT
413 * or other DEFINE_EVENT()s).
414 *
415 * Note, TRACE_EVENT() itself is simply defined as:
416 *
417 * #define TRACE_EVENT(name, proto, args, tstruct, assign, printk) \
418 * DEFINE_EVENT_CLASS(name, proto, args, tstruct, assign, printk); \
419 * DEFINE_EVENT(name, name, proto, args)
420 *
421 * The DEFINE_EVENT() also can be declared with conditions and reg functions:
422 *
423 * DEFINE_EVENT_CONDITION(template, name, proto, args, cond);
424 * DEFINE_EVENT_FN(template, name, proto, args, reg, unreg);
425 */
426DECLARE_EVENT_CLASS(foo_template,
427
428 TP_PROTO(const char *foo, int bar),
429
430 TP_ARGS(foo, bar),
431
432 TP_STRUCT__entry(
433 __string( foo, foo )
434 __field( int, bar )
435 ),
436
437 TP_fast_assign(
438 __assign_str(foo, foo);
439 __entry->bar = bar;
440 ),
441
442 TP_printk("foo %s %d", __get_str(foo), __entry->bar)
443);
444
445/*
446 * Here's a better way for the previous samples (except, the first
447 * exmaple had more fields and could not be used here).
448 */
449DEFINE_EVENT(foo_template, foo_with_template_simple,
450 TP_PROTO(const char *foo, int bar),
451 TP_ARGS(foo, bar));
452
453DEFINE_EVENT_CONDITION(foo_template, foo_with_template_cond,
454 TP_PROTO(const char *foo, int bar),
455 TP_ARGS(foo, bar),
456 TP_CONDITION(!(bar % 8)));
457
458
459DEFINE_EVENT_FN(foo_template, foo_with_template_fn,
460 TP_PROTO(const char *foo, int bar),
461 TP_ARGS(foo, bar),
462 foo_bar_reg, foo_bar_unreg);
463
464/*
465 * Anytime two events share basically the same values and have
466 * the same output, use the DECLARE_EVENT_CLASS() and DEFINE_EVENT()
467 * when ever possible.
468 */
469
470/*
471 * If the event is similar to the DECLARE_EVENT_CLASS, but you need
472 * to have a different output, then use DEFINE_EVENT_PRINT() which
473 * lets you override the TP_printk() of the class.
474 */
475
476DEFINE_EVENT_PRINT(foo_template, foo_with_template_print,
477 TP_PROTO(const char *foo, int bar),
478 TP_ARGS(foo, bar),
479 TP_printk("bar %s %d", __get_str(foo), __entry->bar));
480
481#endif
482
483/***** NOTICE! The #if protection ends here. *****/
484
485
486/*
487 * There are several ways I could have done this. If I left out the
488 * TRACE_INCLUDE_PATH, then it would default to the kernel source
489 * include/trace/events directory.
490 *
491 * I could specify a path from the define_trace.h file back to this
492 * file.
493 *
494 * #define TRACE_INCLUDE_PATH ../../samples/trace_events
495 *
496 * But the safest and easiest way to simply make it use the directory
497 * that the file is in is to add in the Makefile:
498 *
499 * CFLAGS_trace-events-sample.o := -I$(src)
500 *
501 * This will make sure the current path is part of the include
502 * structure for our file so that define_trace.h can find it.
503 *
504 * I could have made only the top level directory the include:
505 *
506 * CFLAGS_trace-events-sample.o := -I$(PWD)
507 *
508 * And then let the path to this directory be the TRACE_INCLUDE_PATH:
509 *
510 * #define TRACE_INCLUDE_PATH samples/trace_events
511 *
512 * But then if something defines "samples" or "trace_events" as a macro
513 * then we could risk that being converted too, and give us an unexpected
514 * result.
515 */
516#undef TRACE_INCLUDE_PATH
517#undef TRACE_INCLUDE_FILE
518#define TRACE_INCLUDE_PATH .
519/*
520 * TRACE_INCLUDE_FILE is not needed if the filename and TRACE_SYSTEM are equal
521 */
522#define TRACE_INCLUDE_FILE trace-events-sample
523#include <trace/define_trace.h>