1 #ifndef _LINUX_FTRACE_EVENT_H
2 #define _LINUX_FTRACE_EVENT_H
4 #include <linux/ring_buffer.h>
5 #include <linux/trace_seq.h>
6 #include <linux/percpu.h>
7 #include <linux/hardirq.h>
8 #include <linux/perf_event.h>
15 struct trace_print_flags {
20 struct trace_print_flags_u64 {
21 unsigned long long mask;
25 const char *ftrace_print_flags_seq(struct trace_seq *p, const char *delim,
27 const struct trace_print_flags *flag_array);
29 const char *ftrace_print_symbols_seq(struct trace_seq *p, unsigned long val,
30 const struct trace_print_flags *symbol_array);
32 #if BITS_PER_LONG == 32
33 const char *ftrace_print_symbols_seq_u64(struct trace_seq *p,
34 unsigned long long val,
35 const struct trace_print_flags_u64
39 const char *ftrace_print_hex_seq(struct trace_seq *p,
40 const unsigned char *buf, int len);
42 struct trace_iterator;
45 int ftrace_raw_output_prep(struct trace_iterator *iter,
46 struct trace_event *event);
49 * The trace entry - the most basic unit of tracing. This is what
50 * is printed in the end as a single line in the trace output, such as:
52 * bash-15816 [01] 235.197585: idle_cpu <- irq_enter
57 unsigned char preempt_count;
61 #define FTRACE_MAX_EVENT \
62 ((1 << (sizeof(((struct trace_entry *)0)->type) * 8)) - 1)
65 * Trace iterator - used by printout routines who present trace
66 * results to users and which routines might sleep, etc:
68 struct trace_iterator {
69 struct trace_array *tr;
71 struct trace_buffer *trace_buffer;
75 struct ring_buffer_iter **buffer_iter;
76 unsigned long iter_flags;
78 /* trace_seq for __print_flags() and __print_symbolic() etc. */
79 struct trace_seq tmp_seq;
81 cpumask_var_t started;
83 /* it's true when current open file is snapshot */
86 /* The below is zeroed out in pipe_read */
88 struct trace_entry *ent;
89 unsigned long lost_events;
98 /* All new field here will be zeroed out in pipe_read */
101 enum trace_iter_flags {
102 TRACE_FILE_LAT_FMT = 1,
103 TRACE_FILE_ANNOTATE = 2,
104 TRACE_FILE_TIME_IN_NS = 4,
108 typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
109 int flags, struct trace_event *event);
111 struct trace_event_functions {
112 trace_print_func trace;
113 trace_print_func raw;
114 trace_print_func hex;
115 trace_print_func binary;
119 struct hlist_node node;
120 struct list_head list;
122 struct trace_event_functions *funcs;
125 extern int register_ftrace_event(struct trace_event *event);
126 extern int unregister_ftrace_event(struct trace_event *event);
128 /* Return values for print_line callback */
130 TRACE_TYPE_PARTIAL_LINE = 0, /* Retry after flushing the seq */
131 TRACE_TYPE_HANDLED = 1,
132 TRACE_TYPE_UNHANDLED = 2, /* Relay to other output functions */
133 TRACE_TYPE_NO_CONSUME = 3 /* Handled but ask to not consume */
136 void tracing_generic_entry_update(struct trace_entry *entry,
139 struct ftrace_event_file;
141 struct ring_buffer_event *
142 trace_event_buffer_lock_reserve(struct ring_buffer **current_buffer,
143 struct ftrace_event_file *ftrace_file,
144 int type, unsigned long len,
145 unsigned long flags, int pc);
146 struct ring_buffer_event *
147 trace_current_buffer_lock_reserve(struct ring_buffer **current_buffer,
148 int type, unsigned long len,
149 unsigned long flags, int pc);
150 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
151 struct ring_buffer_event *event,
152 unsigned long flags, int pc);
153 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
154 struct ring_buffer_event *event,
155 unsigned long flags, int pc);
156 void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer,
157 struct ring_buffer_event *event,
158 unsigned long flags, int pc,
159 struct pt_regs *regs);
160 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
161 struct ring_buffer_event *event);
163 void tracing_record_cmdline(struct task_struct *tsk);
169 TRACE_REG_UNREGISTER,
170 #ifdef CONFIG_PERF_EVENTS
171 TRACE_REG_PERF_REGISTER,
172 TRACE_REG_PERF_UNREGISTER,
174 TRACE_REG_PERF_CLOSE,
180 struct ftrace_event_call;
182 struct ftrace_event_class {
185 #ifdef CONFIG_PERF_EVENTS
188 int (*reg)(struct ftrace_event_call *event,
189 enum trace_reg type, void *data);
190 int (*define_fields)(struct ftrace_event_call *);
191 struct list_head *(*get_fields)(struct ftrace_event_call *);
192 struct list_head fields;
193 int (*raw_init)(struct ftrace_event_call *);
196 extern int ftrace_event_reg(struct ftrace_event_call *event,
197 enum trace_reg type, void *data);
200 TRACE_EVENT_FL_FILTERED_BIT,
201 TRACE_EVENT_FL_CAP_ANY_BIT,
202 TRACE_EVENT_FL_NO_SET_FILTER_BIT,
203 TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
204 TRACE_EVENT_FL_WAS_ENABLED_BIT,
205 TRACE_EVENT_FL_USE_CALL_FILTER_BIT,
210 * FILTERED - The event has a filter attached
211 * CAP_ANY - Any user can enable for perf
212 * NO_SET_FILTER - Set when filter has error and is to be ignored
213 * IGNORE_ENABLE - For ftrace internal events, do not enable with debugfs file
214 * WAS_ENABLED - Set and stays set when an event was ever enabled
215 * (used for module unloading, if a module event is enabled,
216 * it is best to clear the buffers that used it).
217 * USE_CALL_FILTER - For ftrace internal events, don't use file filter
220 TRACE_EVENT_FL_FILTERED = (1 << TRACE_EVENT_FL_FILTERED_BIT),
221 TRACE_EVENT_FL_CAP_ANY = (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
222 TRACE_EVENT_FL_NO_SET_FILTER = (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
223 TRACE_EVENT_FL_IGNORE_ENABLE = (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
224 TRACE_EVENT_FL_WAS_ENABLED = (1 << TRACE_EVENT_FL_WAS_ENABLED_BIT),
225 TRACE_EVENT_FL_USE_CALL_FILTER = (1 << TRACE_EVENT_FL_USE_CALL_FILTER_BIT),
228 struct ftrace_event_call {
229 struct list_head list;
230 struct ftrace_event_class *class;
232 struct trace_event event;
233 const char *print_fmt;
234 struct event_filter *filter;
235 struct list_head *files;
239 * bit 0: filter_active
240 * bit 1: allow trace by non root (cap any)
241 * bit 2: failed to apply filter
242 * bit 3: ftrace internal event (do not enable)
243 * bit 4: Event was enabled by module
244 * bit 5: use call filter rather than file filter
246 int flags; /* static flags of different events */
248 #ifdef CONFIG_PERF_EVENTS
250 struct hlist_head __percpu *perf_events;
252 int (*perf_perm)(struct ftrace_event_call *,
253 struct perf_event *);
258 struct ftrace_subsystem_dir;
261 FTRACE_EVENT_FL_ENABLED_BIT,
262 FTRACE_EVENT_FL_RECORDED_CMD_BIT,
263 FTRACE_EVENT_FL_FILTERED_BIT,
264 FTRACE_EVENT_FL_NO_SET_FILTER_BIT,
265 FTRACE_EVENT_FL_SOFT_MODE_BIT,
266 FTRACE_EVENT_FL_SOFT_DISABLED_BIT,
267 FTRACE_EVENT_FL_TRIGGER_MODE_BIT,
271 * Ftrace event file flags:
272 * ENABLED - The event is enabled
273 * RECORDED_CMD - The comms should be recorded at sched_switch
274 * FILTERED - The event has a filter attached
275 * NO_SET_FILTER - Set when filter has error and is to be ignored
276 * SOFT_MODE - The event is enabled/disabled by SOFT_DISABLED
277 * SOFT_DISABLED - When set, do not trace the event (even though its
278 * tracepoint may be enabled)
279 * TRIGGER_MODE - When set, invoke the triggers associated with the event
282 FTRACE_EVENT_FL_ENABLED = (1 << FTRACE_EVENT_FL_ENABLED_BIT),
283 FTRACE_EVENT_FL_RECORDED_CMD = (1 << FTRACE_EVENT_FL_RECORDED_CMD_BIT),
284 FTRACE_EVENT_FL_FILTERED = (1 << FTRACE_EVENT_FL_FILTERED_BIT),
285 FTRACE_EVENT_FL_NO_SET_FILTER = (1 << FTRACE_EVENT_FL_NO_SET_FILTER_BIT),
286 FTRACE_EVENT_FL_SOFT_MODE = (1 << FTRACE_EVENT_FL_SOFT_MODE_BIT),
287 FTRACE_EVENT_FL_SOFT_DISABLED = (1 << FTRACE_EVENT_FL_SOFT_DISABLED_BIT),
288 FTRACE_EVENT_FL_TRIGGER_MODE = (1 << FTRACE_EVENT_FL_TRIGGER_MODE_BIT),
291 struct ftrace_event_file {
292 struct list_head list;
293 struct ftrace_event_call *event_call;
294 struct event_filter *filter;
296 struct trace_array *tr;
297 struct ftrace_subsystem_dir *system;
298 struct list_head triggers;
303 * bit 1: enabled cmd record
304 * bit 2: enable/disable with the soft disable bit
305 * bit 3: soft disabled
306 * bit 4: trigger enabled
308 * Note: The bits must be set atomically to prevent races
309 * from other writers. Reads of flags do not need to be in
310 * sync as they occur in critical sections. But the way flags
311 * is currently used, these changes do not affect the code
312 * except that when a change is made, it may have a slight
313 * delay in propagating the changes to other CPUs due to
314 * caching and such. Which is mostly OK ;-)
317 atomic_t sm_ref; /* soft-mode reference counter */
318 atomic_t tm_ref; /* trigger-mode reference counter */
321 #define __TRACE_EVENT_FLAGS(name, value) \
322 static int __init trace_init_flags_##name(void) \
324 event_##name.flags = value; \
327 early_initcall(trace_init_flags_##name);
329 #define __TRACE_EVENT_PERF_PERM(name, expr...) \
330 static int perf_perm_##name(struct ftrace_event_call *tp_event, \
331 struct perf_event *p_event) \
333 return ({ expr; }); \
335 static int __init trace_init_perf_perm_##name(void) \
337 event_##name.perf_perm = &perf_perm_##name; \
340 early_initcall(trace_init_perf_perm_##name);
342 #define PERF_MAX_TRACE_SIZE 2048
344 #define MAX_FILTER_STR_VAL 256 /* Should handle KSYM_SYMBOL_LEN */
346 enum event_trigger_type {
348 ETT_TRACE_ONOFF = (1 << 0),
349 ETT_SNAPSHOT = (1 << 1),
352 extern void destroy_preds(struct ftrace_event_file *file);
353 extern void destroy_call_preds(struct ftrace_event_call *call);
354 extern int filter_match_preds(struct event_filter *filter, void *rec);
356 extern int filter_check_discard(struct ftrace_event_file *file, void *rec,
357 struct ring_buffer *buffer,
358 struct ring_buffer_event *event);
359 extern int call_filter_check_discard(struct ftrace_event_call *call, void *rec,
360 struct ring_buffer *buffer,
361 struct ring_buffer_event *event);
362 extern void event_triggers_call(struct ftrace_event_file *file);
366 FILTER_STATIC_STRING,
372 #define EVENT_STORAGE_SIZE 128
373 extern struct mutex event_storage_mutex;
374 extern char event_storage[EVENT_STORAGE_SIZE];
376 extern int trace_event_raw_init(struct ftrace_event_call *call);
377 extern int trace_define_field(struct ftrace_event_call *call, const char *type,
378 const char *name, int offset, int size,
379 int is_signed, int filter_type);
380 extern int trace_add_event_call(struct ftrace_event_call *call);
381 extern int trace_remove_event_call(struct ftrace_event_call *call);
383 #define is_signed_type(type) (((type)(-1)) < (type)1)
385 int trace_set_clr_event(const char *system, const char *event, int set);
388 * The double __builtin_constant_p is because gcc will give us an error
389 * if we try to allocate the static variable to fmt if it is not a
390 * constant. Even with the outer if statement optimizing out.
392 #define event_trace_printk(ip, fmt, args...) \
394 __trace_printk_check_format(fmt, ##args); \
395 tracing_record_cmdline(current); \
396 if (__builtin_constant_p(fmt)) { \
397 static const char *trace_printk_fmt \
398 __attribute__((section("__trace_printk_fmt"))) = \
399 __builtin_constant_p(fmt) ? fmt : NULL; \
401 __trace_bprintk(ip, trace_printk_fmt, ##args); \
403 __trace_printk(ip, fmt, ##args); \
407 * tracepoint_string - register constant persistent string to trace system
408 * @str - a constant persistent string that will be referenced in tracepoints
410 * If constant strings are being used in tracepoints, it is faster and
411 * more efficient to just save the pointer to the string and reference
412 * that with a printf "%s" instead of saving the string in the ring buffer
413 * and wasting space and time.
415 * The problem with the above approach is that userspace tools that read
416 * the binary output of the trace buffers do not have access to the string.
417 * Instead they just show the address of the string which is not very
420 * With tracepoint_string(), the string will be registered to the tracing
421 * system and exported to userspace via the debugfs/tracing/printk_formats
422 * file that maps the string address to the string text. This way userspace
423 * tools that read the binary buffers have a way to map the pointers to
424 * the ASCII strings they represent.
426 * The @str used must be a constant string and persistent as it would not
427 * make sense to show a string that no longer exists. But it is still fine
428 * to be used with modules, because when modules are unloaded, if they
429 * had tracepoints, the ring buffers are cleared too. As long as the string
430 * does not change during the life of the module, it is fine to use
431 * tracepoint_string() within a module.
433 #define tracepoint_string(str) \
435 static const char *___tp_str __tracepoint_string = str; \
438 #define __tracepoint_string __attribute__((section("__tracepoint_str")))
440 #ifdef CONFIG_PERF_EVENTS
443 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
445 extern int perf_trace_init(struct perf_event *event);
446 extern void perf_trace_destroy(struct perf_event *event);
447 extern int perf_trace_add(struct perf_event *event, int flags);
448 extern void perf_trace_del(struct perf_event *event, int flags);
449 extern int ftrace_profile_set_filter(struct perf_event *event, int event_id,
451 extern void ftrace_profile_free_filter(struct perf_event *event);
452 extern void *perf_trace_buf_prepare(int size, unsigned short type,
453 struct pt_regs *regs, int *rctxp);
456 perf_trace_buf_submit(void *raw_data, int size, int rctx, u64 addr,
457 u64 count, struct pt_regs *regs, void *head,
458 struct task_struct *task)
460 perf_tp_event(addr, count, raw_data, size, regs, head, rctx, task);
464 #endif /* _LINUX_FTRACE_EVENT_H */