2 #ifndef _LINUX_TRACE_EVENT_H
3 #define _LINUX_TRACE_EVENT_H
5 #include <linux/ring_buffer.h>
6 #include <linux/trace_seq.h>
7 #include <linux/percpu.h>
8 #include <linux/hardirq.h>
9 #include <linux/perf_event.h>
10 #include <linux/tracepoint.h>
18 const char *trace_print_flags_seq(struct trace_seq *p, const char *delim,
20 const struct trace_print_flags *flag_array);
22 const char *trace_print_symbols_seq(struct trace_seq *p, unsigned long val,
23 const struct trace_print_flags *symbol_array);
25 #if BITS_PER_LONG == 32
26 const char *trace_print_symbols_seq_u64(struct trace_seq *p,
27 unsigned long long val,
28 const struct trace_print_flags_u64
32 const char *trace_print_bitmask_seq(struct trace_seq *p, void *bitmask_ptr,
33 unsigned int bitmask_size);
35 const char *trace_print_hex_seq(struct trace_seq *p,
36 const unsigned char *buf, int len);
38 const char *trace_print_array_seq(struct trace_seq *p,
39 const void *buf, int count,
42 struct trace_iterator;
45 int trace_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 TRACE_EVENT_TYPE_MAX \
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_trace_event(struct trace_event *event);
126 extern int unregister_trace_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 */
137 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
138 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
139 * simplifies those functions and keeps them in sync.
141 static inline enum print_line_t trace_handle_return(struct trace_seq *s)
143 return trace_seq_has_overflowed(s) ?
144 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
147 void tracing_generic_entry_update(struct trace_entry *entry,
150 struct trace_event_file;
152 struct ring_buffer_event *
153 trace_event_buffer_lock_reserve(struct ring_buffer **current_buffer,
154 struct trace_event_file *trace_file,
155 int type, unsigned long len,
156 unsigned long flags, int pc);
158 void tracing_record_cmdline(struct task_struct *tsk);
160 int trace_output_call(struct trace_iterator *iter, char *name, char *fmt, ...);
166 TRACE_REG_UNREGISTER,
167 #ifdef CONFIG_PERF_EVENTS
168 TRACE_REG_PERF_REGISTER,
169 TRACE_REG_PERF_UNREGISTER,
171 TRACE_REG_PERF_CLOSE,
177 struct trace_event_call;
179 struct trace_event_class {
182 #ifdef CONFIG_PERF_EVENTS
185 int (*reg)(struct trace_event_call *event,
186 enum trace_reg type, void *data);
187 int (*define_fields)(struct trace_event_call *);
188 struct list_head *(*get_fields)(struct trace_event_call *);
189 struct list_head fields;
190 int (*raw_init)(struct trace_event_call *);
193 extern int trace_event_reg(struct trace_event_call *event,
194 enum trace_reg type, void *data);
196 struct trace_event_buffer {
197 struct ring_buffer *buffer;
198 struct ring_buffer_event *event;
199 struct trace_event_file *trace_file;
205 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
206 struct trace_event_file *trace_file,
209 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer);
212 TRACE_EVENT_FL_FILTERED_BIT,
213 TRACE_EVENT_FL_CAP_ANY_BIT,
214 TRACE_EVENT_FL_NO_SET_FILTER_BIT,
215 TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
216 TRACE_EVENT_FL_WAS_ENABLED_BIT,
217 TRACE_EVENT_FL_TRACEPOINT_BIT,
218 TRACE_EVENT_FL_KPROBE_BIT,
219 TRACE_EVENT_FL_UPROBE_BIT,
224 * FILTERED - The event has a filter attached
225 * CAP_ANY - Any user can enable for perf
226 * NO_SET_FILTER - Set when filter has error and is to be ignored
227 * IGNORE_ENABLE - For trace internal events, do not enable with debugfs file
228 * WAS_ENABLED - Set and stays set when an event was ever enabled
229 * (used for module unloading, if a module event is enabled,
230 * it is best to clear the buffers that used it).
231 * TRACEPOINT - Event is a tracepoint
232 * KPROBE - Event is a kprobe
233 * UPROBE - Event is a uprobe
236 TRACE_EVENT_FL_FILTERED = (1 << TRACE_EVENT_FL_FILTERED_BIT),
237 TRACE_EVENT_FL_CAP_ANY = (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
238 TRACE_EVENT_FL_NO_SET_FILTER = (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
239 TRACE_EVENT_FL_IGNORE_ENABLE = (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
240 TRACE_EVENT_FL_WAS_ENABLED = (1 << TRACE_EVENT_FL_WAS_ENABLED_BIT),
241 TRACE_EVENT_FL_TRACEPOINT = (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
242 TRACE_EVENT_FL_KPROBE = (1 << TRACE_EVENT_FL_KPROBE_BIT),
243 TRACE_EVENT_FL_UPROBE = (1 << TRACE_EVENT_FL_UPROBE_BIT),
246 #define TRACE_EVENT_FL_UKPROBE (TRACE_EVENT_FL_KPROBE | TRACE_EVENT_FL_UPROBE)
248 struct trace_event_call {
249 struct list_head list;
250 struct trace_event_class *class;
253 /* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
254 struct tracepoint *tp;
256 struct trace_event event;
258 struct event_filter *filter;
262 * bit 0: filter_active
263 * bit 1: allow trace by non root (cap any)
264 * bit 2: failed to apply filter
265 * bit 3: trace internal event (do not enable)
266 * bit 4: Event was enabled by module
267 * bit 5: use call filter rather than file filter
268 * bit 6: Event is a tracepoint
270 int flags; /* static flags of different events */
272 #ifdef CONFIG_PERF_EVENTS
274 struct hlist_head __percpu *perf_events;
275 struct bpf_prog *prog;
276 struct perf_event *bpf_prog_owner;
278 int (*perf_perm)(struct trace_event_call *,
279 struct perf_event *);
283 static inline const char *
284 trace_event_name(struct trace_event_call *call)
286 if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
287 return call->tp ? call->tp->name : NULL;
293 struct trace_subsystem_dir;
296 EVENT_FILE_FL_ENABLED_BIT,
297 EVENT_FILE_FL_RECORDED_CMD_BIT,
298 EVENT_FILE_FL_FILTERED_BIT,
299 EVENT_FILE_FL_NO_SET_FILTER_BIT,
300 EVENT_FILE_FL_SOFT_MODE_BIT,
301 EVENT_FILE_FL_SOFT_DISABLED_BIT,
302 EVENT_FILE_FL_TRIGGER_MODE_BIT,
303 EVENT_FILE_FL_TRIGGER_COND_BIT,
304 EVENT_FILE_FL_PID_FILTER_BIT,
309 * ENABLED - The event is enabled
310 * RECORDED_CMD - The comms should be recorded at sched_switch
311 * FILTERED - The event has a filter attached
312 * NO_SET_FILTER - Set when filter has error and is to be ignored
313 * SOFT_MODE - The event is enabled/disabled by SOFT_DISABLED
314 * SOFT_DISABLED - When set, do not trace the event (even though its
315 * tracepoint may be enabled)
316 * TRIGGER_MODE - When set, invoke the triggers associated with the event
317 * TRIGGER_COND - When set, one or more triggers has an associated filter
318 * PID_FILTER - When set, the event is filtered based on pid
321 EVENT_FILE_FL_ENABLED = (1 << EVENT_FILE_FL_ENABLED_BIT),
322 EVENT_FILE_FL_RECORDED_CMD = (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
323 EVENT_FILE_FL_FILTERED = (1 << EVENT_FILE_FL_FILTERED_BIT),
324 EVENT_FILE_FL_NO_SET_FILTER = (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
325 EVENT_FILE_FL_SOFT_MODE = (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
326 EVENT_FILE_FL_SOFT_DISABLED = (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
327 EVENT_FILE_FL_TRIGGER_MODE = (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
328 EVENT_FILE_FL_TRIGGER_COND = (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
329 EVENT_FILE_FL_PID_FILTER = (1 << EVENT_FILE_FL_PID_FILTER_BIT),
332 struct trace_event_file {
333 struct list_head list;
334 struct trace_event_call *event_call;
335 struct event_filter *filter;
337 struct trace_array *tr;
338 struct trace_subsystem_dir *system;
339 struct list_head triggers;
344 * bit 1: enabled cmd record
345 * bit 2: enable/disable with the soft disable bit
346 * bit 3: soft disabled
347 * bit 4: trigger enabled
349 * Note: The bits must be set atomically to prevent races
350 * from other writers. Reads of flags do not need to be in
351 * sync as they occur in critical sections. But the way flags
352 * is currently used, these changes do not affect the code
353 * except that when a change is made, it may have a slight
354 * delay in propagating the changes to other CPUs due to
355 * caching and such. Which is mostly OK ;-)
358 atomic_t sm_ref; /* soft-mode reference counter */
359 atomic_t tm_ref; /* trigger-mode reference counter */
362 #define __TRACE_EVENT_FLAGS(name, value) \
363 static int __init trace_init_flags_##name(void) \
365 event_##name.flags |= value; \
368 early_initcall(trace_init_flags_##name);
370 #define __TRACE_EVENT_PERF_PERM(name, expr...) \
371 static int perf_perm_##name(struct trace_event_call *tp_event, \
372 struct perf_event *p_event) \
374 return ({ expr; }); \
376 static int __init trace_init_perf_perm_##name(void) \
378 event_##name.perf_perm = &perf_perm_##name; \
381 early_initcall(trace_init_perf_perm_##name);
383 #define PERF_MAX_TRACE_SIZE 2048
385 #define MAX_FILTER_STR_VAL 256 /* Should handle KSYM_SYMBOL_LEN */
387 enum event_trigger_type {
389 ETT_TRACE_ONOFF = (1 << 0),
390 ETT_SNAPSHOT = (1 << 1),
391 ETT_STACKTRACE = (1 << 2),
392 ETT_EVENT_ENABLE = (1 << 3),
393 ETT_EVENT_HIST = (1 << 4),
394 ETT_HIST_ENABLE = (1 << 5),
397 extern int filter_match_preds(struct event_filter *filter, void *rec);
399 extern enum event_trigger_type event_triggers_call(struct trace_event_file *file,
401 extern void event_triggers_post_call(struct trace_event_file *file,
402 enum event_trigger_type tt,
405 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file);
408 * trace_trigger_soft_disabled - do triggers and test if soft disabled
409 * @file: The file pointer of the event to test
411 * If any triggers without filters are attached to this event, they
412 * will be called here. If the event is soft disabled and has no
413 * triggers that require testing the fields, it will return true,
417 trace_trigger_soft_disabled(struct trace_event_file *file)
419 unsigned long eflags = file->flags;
421 if (!(eflags & EVENT_FILE_FL_TRIGGER_COND)) {
422 if (eflags & EVENT_FILE_FL_TRIGGER_MODE)
423 event_triggers_call(file, NULL);
424 if (eflags & EVENT_FILE_FL_SOFT_DISABLED)
426 if (eflags & EVENT_FILE_FL_PID_FILTER)
427 return trace_event_ignore_this_pid(file);
432 #ifdef CONFIG_BPF_EVENTS
433 unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx);
435 static inline unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx)
443 FILTER_STATIC_STRING,
451 extern int trace_event_raw_init(struct trace_event_call *call);
452 extern int trace_define_field(struct trace_event_call *call, const char *type,
453 const char *name, int offset, int size,
454 int is_signed, int filter_type);
455 extern int trace_add_event_call(struct trace_event_call *call);
456 extern int trace_remove_event_call(struct trace_event_call *call);
457 extern int trace_event_get_offsets(struct trace_event_call *call);
459 #define is_signed_type(type) (((type)(-1)) < (type)1)
461 int trace_set_clr_event(const char *system, const char *event, int set);
464 * The double __builtin_constant_p is because gcc will give us an error
465 * if we try to allocate the static variable to fmt if it is not a
466 * constant. Even with the outer if statement optimizing out.
468 #define event_trace_printk(ip, fmt, args...) \
470 __trace_printk_check_format(fmt, ##args); \
471 tracing_record_cmdline(current); \
472 if (__builtin_constant_p(fmt)) { \
473 static const char *trace_printk_fmt \
474 __attribute__((section("__trace_printk_fmt"))) = \
475 __builtin_constant_p(fmt) ? fmt : NULL; \
477 __trace_bprintk(ip, trace_printk_fmt, ##args); \
479 __trace_printk(ip, fmt, ##args); \
482 #ifdef CONFIG_PERF_EVENTS
485 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
487 extern int perf_trace_init(struct perf_event *event);
488 extern void perf_trace_destroy(struct perf_event *event);
489 extern int perf_trace_add(struct perf_event *event, int flags);
490 extern void perf_trace_del(struct perf_event *event, int flags);
491 extern int ftrace_profile_set_filter(struct perf_event *event, int event_id,
493 extern void ftrace_profile_free_filter(struct perf_event *event);
494 void perf_trace_buf_update(void *record, u16 type);
495 void *perf_trace_buf_alloc(int size, struct pt_regs **regs, int *rctxp);
497 void perf_trace_run_bpf_submit(void *raw_data, int size, int rctx,
498 struct trace_event_call *call, u64 count,
499 struct pt_regs *regs, struct hlist_head *head,
500 struct task_struct *task);
503 perf_trace_buf_submit(void *raw_data, int size, int rctx, u16 type,
504 u64 count, struct pt_regs *regs, void *head,
505 struct task_struct *task)
507 perf_tp_event(type, count, raw_data, size, regs, head, rctx, task);
511 #endif /* _LINUX_TRACE_EVENT_H */