blob: 8965d1bb881194685d1af344cbdfd2868c2aa9fd [file] [log] [blame]
Steven Rostedt9cfe06f2009-04-14 21:37:03 -04001/*
Steven Rostedt9cfe06f2009-04-14 21:37:03 -04002 * If TRACE_SYSTEM is defined, that will be the directory created
Steven Rostedt (Red Hat)4e20e3a2015-02-09 15:27:04 -05003 * in the ftrace directory under /sys/kernel/tracing/events/<system>
Steven Rostedt9cfe06f2009-04-14 21:37:03 -04004 *
Steven Rostedt44ad18e2009-06-16 19:53:07 -04005 * The define_trace.h below will also look for a file name of
Steven Rostedt9cfe06f2009-04-14 21:37:03 -04006 * TRACE_SYSTEM.h where TRACE_SYSTEM is what is defined here.
Steven Rostedt44ad18e2009-06-16 19:53:07 -04007 * In this case, it would look for sample.h
Steven Rostedt9cfe06f2009-04-14 21:37:03 -04008 *
Steven Rostedt44ad18e2009-06-16 19:53:07 -04009 * 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
Steven Rostedt9cfe06f2009-04-14 21:37:03 -040012 *
Steven Rostedt44ad18e2009-06-16 19:53:07 -040013 * This file is called trace-events-sample.h but we want the system
14 * to be called "sample". Therefore we must define the name of this
15 * file:
Steven Rostedt9cfe06f2009-04-14 21:37:03 -040016 *
Steven Rostedt44ad18e2009-06-16 19:53:07 -040017 * #define TRACE_INCLUDE_FILE trace-events-sample
Steven Rostedt9cfe06f2009-04-14 21:37:03 -040018 *
Steven Rostedt44ad18e2009-06-16 19:53:07 -040019 * As we do an the bottom of this file.
Li Zefand0b6e042009-07-13 10:33:21 +080020 *
21 * Notice that TRACE_SYSTEM should be defined outside of #if
22 * protection, just like TRACE_INCLUDE_FILE.
Steven Rostedt9cfe06f2009-04-14 21:37:03 -040023 */
24#undef TRACE_SYSTEM
Steven Rostedt (Red Hat)88920422015-04-01 14:57:27 -040025#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 */
Steven Rostedt9cfe06f2009-04-14 21:37:03 -040044
45/*
Li Zefand0b6e042009-07-13 10:33:21 +080046 * 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/*
Steven Rostedt9cfe06f2009-04-14 21:37:03 -040063 * 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.
Steven Rostedt (Red Hat)4e20e3a2015-02-09 15:27:04 -050075 * 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.
Steven Rostedt9cfe06f2009-04-14 21:37:03 -040078 *
Steven Rostedt (Red Hat)4e20e3a2015-02-09 15:27:04 -050079 * 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 is 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 *
Steven Rostedt9cfe06f2009-04-14 21:37:03 -0400154 *
155 * fast_assign: This is a C like function that is used to store the items
Steven Rostedt (Red Hat)4e20e3a2015-02-09 15:27:04 -0500156 * 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.
Steven Rostedt9cfe06f2009-04-14 21:37:03 -0400159 *
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.
Steven Rostedt (Red Hat)4e20e3a2015-02-09 15:27:04 -0500163 * 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.
172 *
173 * For __string(foo, bar) use __get_str(foo)
174 *
175 * For __bitmask(target_cpus, nr_cpumask_bits) use __get_bitmask(target_cpus)
176 *
Steven Rostedt9cfe06f2009-04-14 21:37:03 -0400177 *
178 * Note, that for both the assign and the printk, __entry is the handler
179 * to the data structure in the ring buffer, and is defined by the
180 * TP_STRUCT__entry.
181 */
Steven Rostedt (Red Hat)4e20e3a2015-02-09 15:27:04 -0500182
183/*
184 * It is OK to have helper functions in the file, but they need to be protected
185 * from being defined more than once. Remember, this file gets included more
186 * than once.
187 */
188#ifndef __TRACE_EVENT_SAMPLE_HELPER_FUNCTIONS
189#define __TRACE_EVENT_SAMPLE_HELPER_FUNCTIONS
190static inline int __length_of(const int *list)
191{
192 int i;
193
194 if (!list)
195 return 0;
196
197 for (i = 0; list[i]; i++)
198 ;
199 return i;
200}
Steven Rostedt (Red Hat)32eb3d02015-04-01 15:55:36 -0400201
202enum {
203 TRACE_SAMPLE_FOO = 2,
204 TRACE_SAMPLE_BAR = 4,
205 TRACE_SAMPLE_ZOO = 8,
206};
Steven Rostedt (Red Hat)4e20e3a2015-02-09 15:27:04 -0500207#endif
208
Steven Rostedt (Red Hat)32eb3d02015-04-01 15:55:36 -0400209/*
210 * If enums are used in the TP_printk(), their names will be shown in
211 * format files and not their values. This can cause problems with user
212 * space programs that parse the format files to know how to translate
213 * the raw binary trace output into human readable text.
214 *
215 * To help out user space programs, any enum that is used in the TP_printk()
216 * should be defined by TRACE_DEFINE_ENUM() macro. All that is needed to
217 * be done is to add this macro with the enum within it in the trace
218 * header file, and it will be converted in the output.
219 */
220
221TRACE_DEFINE_ENUM(TRACE_SAMPLE_FOO);
222TRACE_DEFINE_ENUM(TRACE_SAMPLE_BAR);
223TRACE_DEFINE_ENUM(TRACE_SAMPLE_ZOO);
224
Steven Rostedt9cfe06f2009-04-14 21:37:03 -0400225TRACE_EVENT(foo_bar,
226
Steven Rostedt (Red Hat)4e20e3a2015-02-09 15:27:04 -0500227 TP_PROTO(const char *foo, int bar, const int *lst,
228 const char *string, const struct cpumask *mask),
Steven Rostedt9cfe06f2009-04-14 21:37:03 -0400229
Steven Rostedt (Red Hat)4e20e3a2015-02-09 15:27:04 -0500230 TP_ARGS(foo, bar, lst, string, mask),
Steven Rostedt9cfe06f2009-04-14 21:37:03 -0400231
232 TP_STRUCT__entry(
233 __array( char, foo, 10 )
234 __field( int, bar )
Steven Rostedt (Red Hat)4e20e3a2015-02-09 15:27:04 -0500235 __dynamic_array(int, list, __length_of(lst))
236 __string( str, string )
237 __bitmask( cpus, num_possible_cpus() )
Steven Rostedt9cfe06f2009-04-14 21:37:03 -0400238 ),
239
240 TP_fast_assign(
Zhao Hongjiangd8fae2f2013-06-20 19:05:40 +0800241 strlcpy(__entry->foo, foo, 10);
Steven Rostedt9cfe06f2009-04-14 21:37:03 -0400242 __entry->bar = bar;
Steven Rostedt (Red Hat)4e20e3a2015-02-09 15:27:04 -0500243 memcpy(__get_dynamic_array(list), lst,
244 __length_of(lst) * sizeof(int));
245 __assign_str(str, string);
246 __assign_bitmask(cpus, cpumask_bits(mask), num_possible_cpus());
Steven Rostedt9cfe06f2009-04-14 21:37:03 -0400247 ),
248
Steven Rostedt (Red Hat)32eb3d02015-04-01 15:55:36 -0400249 TP_printk("foo %s %d %s %s %s %s (%s)", __entry->foo, __entry->bar,
250
251/*
252 * Notice here the use of some helper functions. This includes:
253 *
254 * __print_symbolic( variable, { value, "string" }, ... ),
255 *
256 * The variable is tested against each value of the { } pair. If
257 * the variable matches one of the values, then it will print the
258 * string in that pair. If non are matched, it returns a string
259 * version of the number (if __entry->bar == 7 then "7" is returned).
260 */
261 __print_symbolic(__entry->bar,
262 { 0, "zero" },
263 { TRACE_SAMPLE_FOO, "TWO" },
264 { TRACE_SAMPLE_BAR, "FOUR" },
265 { TRACE_SAMPLE_ZOO, "EIGHT" },
266 { 10, "TEN" }
267 ),
268
269/*
270 * __print_flags( variable, "delim", { value, "flag" }, ... ),
271 *
272 * This is similar to __print_symbolic, except that it tests the bits
273 * of the value. If ((FLAG & variable) == FLAG) then the string is
274 * printed. If more than one flag matches, then each one that does is
275 * also printed with delim in between them.
276 * If not all bits are accounted for, then the not found bits will be
277 * added in hex format: 0x506 will show BIT2|BIT4|0x500
278 */
279 __print_flags(__entry->bar, "|",
280 { 1, "BIT1" },
281 { 2, "BIT2" },
282 { 4, "BIT3" },
283 { 8, "BIT4" }
284 ),
285/*
286 * __print_array( array, len, element_size )
287 *
288 * This prints out the array that is defined by __array in a nice format.
289 */
Steven Rostedt (Red Hat)4e20e3a2015-02-09 15:27:04 -0500290 __print_array(__get_dynamic_array(list),
291 __get_dynamic_array_len(list),
292 sizeof(int)),
293 __get_str(str), __get_bitmask(cpus))
Steven Rostedt9cfe06f2009-04-14 21:37:03 -0400294);
Steven Rostedt (Red Hat)c4c7eb22015-02-09 16:05:55 -0500295
296/*
297 * There may be a case where a tracepoint should only be called if
298 * some condition is set. Otherwise the tracepoint should not be called.
299 * But to do something like:
300 *
301 * if (cond)
302 * trace_foo();
303 *
304 * Would cause a little overhead when tracing is not enabled, and that
305 * overhead, even if small, is not something we want. As tracepoints
306 * use static branch (aka jump_labels), where no branch is taken to
307 * skip the tracepoint when not enabled, and a jmp is placed to jump
308 * to the tracepoint code when it is enabled, having a if statement
309 * nullifies that optimization. It would be nice to place that
310 * condition within the static branch. This is where TRACE_EVENT_CONDITION
311 * comes in.
312 *
313 * TRACE_EVENT_CONDITION() is just like TRACE_EVENT, except it adds another
314 * parameter just after args. Where TRACE_EVENT has:
315 *
316 * TRACE_EVENT(name, proto, args, struct, assign, printk)
317 *
318 * the CONDITION version has:
319 *
320 * TRACE_EVENT_CONDITION(name, proto, args, cond, struct, assign, printk)
321 *
322 * Everything is the same as TRACE_EVENT except for the new cond. Think
323 * of the cond variable as:
324 *
325 * if (cond)
326 * trace_foo_bar_with_cond();
327 *
328 * Except that the logic for the if branch is placed after the static branch.
329 * That is, the if statement that processes the condition will not be
330 * executed unless that traecpoint is enabled. Otherwise it still remains
331 * a nop.
332 */
333TRACE_EVENT_CONDITION(foo_bar_with_cond,
334
335 TP_PROTO(const char *foo, int bar),
336
337 TP_ARGS(foo, bar),
338
339 TP_CONDITION(!(bar % 10)),
340
341 TP_STRUCT__entry(
342 __string( foo, foo )
343 __field( int, bar )
344 ),
345
346 TP_fast_assign(
347 __assign_str(foo, foo);
348 __entry->bar = bar;
349 ),
350
351 TP_printk("foo %s %d", __get_str(foo), __entry->bar)
352);
Steven Rostedt (Red Hat)6adc13f82015-02-09 16:32:19 -0500353
354void foo_bar_reg(void);
355void foo_bar_unreg(void);
356
357/*
358 * Now in the case that some function needs to be called when the
359 * tracepoint is enabled and/or when it is disabled, the
360 * TRACE_EVENT_FN() serves this purpose. This is just like TRACE_EVENT()
361 * but adds two more parameters at the end:
362 *
363 * TRACE_EVENT_FN( name, proto, args, struct, assign, printk, reg, unreg)
364 *
365 * reg and unreg are functions with the prototype of:
366 *
367 * void reg(void)
368 *
369 * The reg function gets called before the tracepoint is enabled, and
370 * the unreg function gets called after the tracepoint is disabled.
371 *
372 * Note, reg and unreg are allowed to be NULL. If you only need to
373 * call a function before enabling, or after disabling, just set one
374 * function and pass in NULL for the other parameter.
375 */
376TRACE_EVENT_FN(foo_bar_with_fn,
377
378 TP_PROTO(const char *foo, int bar),
379
380 TP_ARGS(foo, bar),
381
382 TP_STRUCT__entry(
383 __string( foo, foo )
384 __field( int, bar )
385 ),
386
387 TP_fast_assign(
388 __assign_str(foo, foo);
389 __entry->bar = bar;
390 ),
391
392 TP_printk("foo %s %d", __get_str(foo), __entry->bar),
393
394 foo_bar_reg, foo_bar_unreg
395);
396
Steven Rostedt (Red Hat)74969462015-02-09 17:14:04 -0500397/*
398 * Each TRACE_EVENT macro creates several helper functions to produce
399 * the code to add the tracepoint, create the files in the trace
400 * directory, hook it to perf, assign the values and to print out
401 * the raw data from the ring buffer. To prevent too much bloat,
402 * if there are more than one tracepoint that uses the same format
403 * for the proto, args, struct, assign and printk, and only the name
404 * is different, it is highly recommended to use the DECLARE_EVENT_CLASS
405 *
406 * DECLARE_EVENT_CLASS() macro creates most of the functions for the
407 * tracepoint. Then DEFINE_EVENT() is use to hook a tracepoint to those
408 * functions. This DEFINE_EVENT() is an instance of the class and can
409 * be enabled and disabled separately from other events (either TRACE_EVENT
410 * or other DEFINE_EVENT()s).
411 *
412 * Note, TRACE_EVENT() itself is simply defined as:
413 *
414 * #define TRACE_EVENT(name, proto, args, tstruct, assign, printk) \
415 * DEFINE_EVENT_CLASS(name, proto, args, tstruct, assign, printk); \
416 * DEFINE_EVENT(name, name, proto, args)
417 *
418 * The DEFINE_EVENT() also can be declared with conditions and reg functions:
419 *
420 * DEFINE_EVENT_CONDITION(template, name, proto, args, cond);
421 * DEFINE_EVENT_FN(template, name, proto, args, reg, unreg);
422 */
423DECLARE_EVENT_CLASS(foo_template,
424
425 TP_PROTO(const char *foo, int bar),
426
427 TP_ARGS(foo, bar),
428
429 TP_STRUCT__entry(
430 __string( foo, foo )
431 __field( int, bar )
432 ),
433
434 TP_fast_assign(
435 __assign_str(foo, foo);
436 __entry->bar = bar;
437 ),
438
439 TP_printk("foo %s %d", __get_str(foo), __entry->bar)
440);
441
442/*
443 * Here's a better way for the previous samples (except, the first
444 * exmaple had more fields and could not be used here).
445 */
446DEFINE_EVENT(foo_template, foo_with_template_simple,
447 TP_PROTO(const char *foo, int bar),
448 TP_ARGS(foo, bar));
449
450DEFINE_EVENT_CONDITION(foo_template, foo_with_template_cond,
451 TP_PROTO(const char *foo, int bar),
452 TP_ARGS(foo, bar),
453 TP_CONDITION(!(bar % 8)));
454
455
456DEFINE_EVENT_FN(foo_template, foo_with_template_fn,
457 TP_PROTO(const char *foo, int bar),
458 TP_ARGS(foo, bar),
459 foo_bar_reg, foo_bar_unreg);
460
461/*
462 * Anytime two events share basically the same values and have
463 * the same output, use the DECLARE_EVENT_CLASS() and DEFINE_EVENT()
464 * when ever possible.
465 */
466
467/*
468 * If the event is similar to the DECLARE_EVENT_CLASS, but you need
469 * to have a different output, then use DEFINE_EVENT_PRINT() which
470 * lets you override the TP_printk() of the class.
471 */
472
473DEFINE_EVENT_PRINT(foo_template, foo_with_template_print,
474 TP_PROTO(const char *foo, int bar),
475 TP_ARGS(foo, bar),
476 TP_printk("bar %s %d", __get_str(foo), __entry->bar));
477
Steven Rostedt9cfe06f2009-04-14 21:37:03 -0400478#endif
479
480/***** NOTICE! The #if protection ends here. *****/
481
482
483/*
484 * There are several ways I could have done this. If I left out the
485 * TRACE_INCLUDE_PATH, then it would default to the kernel source
486 * include/trace/events directory.
487 *
488 * I could specify a path from the define_trace.h file back to this
489 * file.
490 *
491 * #define TRACE_INCLUDE_PATH ../../samples/trace_events
492 *
Steven Rostedt44ad18e2009-06-16 19:53:07 -0400493 * But the safest and easiest way to simply make it use the directory
494 * that the file is in is to add in the Makefile:
Steven Rostedt9cfe06f2009-04-14 21:37:03 -0400495 *
Steven Rostedt44ad18e2009-06-16 19:53:07 -0400496 * CFLAGS_trace-events-sample.o := -I$(src)
Steven Rostedt9cfe06f2009-04-14 21:37:03 -0400497 *
498 * This will make sure the current path is part of the include
Steven Rostedt44ad18e2009-06-16 19:53:07 -0400499 * structure for our file so that define_trace.h can find it.
Steven Rostedt9cfe06f2009-04-14 21:37:03 -0400500 *
501 * I could have made only the top level directory the include:
502 *
503 * CFLAGS_trace-events-sample.o := -I$(PWD)
504 *
505 * And then let the path to this directory be the TRACE_INCLUDE_PATH:
506 *
507 * #define TRACE_INCLUDE_PATH samples/trace_events
508 *
Steven Rostedt44ad18e2009-06-16 19:53:07 -0400509 * But then if something defines "samples" or "trace_events" as a macro
510 * then we could risk that being converted too, and give us an unexpected
Steven Rostedt9cfe06f2009-04-14 21:37:03 -0400511 * result.
512 */
513#undef TRACE_INCLUDE_PATH
Steven Rostedt71e1c8a2009-05-06 21:20:39 -0400514#undef TRACE_INCLUDE_FILE
Steven Rostedt9cfe06f2009-04-14 21:37:03 -0400515#define TRACE_INCLUDE_PATH .
Steven Rostedt71e1c8a2009-05-06 21:20:39 -0400516/*
517 * TRACE_INCLUDE_FILE is not needed if the filename and TRACE_SYSTEM are equal
518 */
519#define TRACE_INCLUDE_FILE trace-events-sample
Steven Rostedt9cfe06f2009-04-14 21:37:03 -0400520#include <trace/define_trace.h>