1 /* SPDX-License-Identifier: GPL-2.0 */
3 #ifndef _LINUX_TRACE_EVENT_H
4 #define _LINUX_TRACE_EVENT_H
6 #include <linux/ring_buffer.h>
7 #include <linux/trace_seq.h>
8 #include <linux/percpu.h>
9 #include <linux/hardirq.h>
10 #include <linux/perf_event.h>
11 #include <linux/tracepoint.h>
19 const char *trace_print_flags_seq(struct trace_seq *p, const char *delim,
21 const struct trace_print_flags *flag_array);
23 const char *trace_print_symbols_seq(struct trace_seq *p, unsigned long val,
24 const struct trace_print_flags *symbol_array);
26 #if BITS_PER_LONG == 32
27 const char *trace_print_flags_seq_u64(struct trace_seq *p, const char *delim,
28 unsigned long long flags,
29 const struct trace_print_flags_u64 *flag_array);
31 const char *trace_print_symbols_seq_u64(struct trace_seq *p,
32 unsigned long long val,
33 const struct trace_print_flags_u64
37 const char *trace_print_bitmask_seq(struct trace_seq *p, void *bitmask_ptr,
38 unsigned int bitmask_size);
40 const char *trace_print_hex_seq(struct trace_seq *p,
41 const unsigned char *buf, int len,
44 const char *trace_print_array_seq(struct trace_seq *p,
45 const void *buf, int count,
49 trace_print_hex_dump_seq(struct trace_seq *p, const char *prefix_str,
50 int prefix_type, int rowsize, int groupsize,
51 const void *buf, size_t len, bool ascii);
53 struct trace_iterator;
56 int trace_raw_output_prep(struct trace_iterator *iter,
57 struct trace_event *event);
60 * The trace entry - the most basic unit of tracing. This is what
61 * is printed in the end as a single line in the trace output, such as:
63 * bash-15816 [01] 235.197585: idle_cpu <- irq_enter
68 unsigned char preempt_count;
72 #define TRACE_EVENT_TYPE_MAX \
73 ((1 << (sizeof(((struct trace_entry *)0)->type) * 8)) - 1)
76 * Trace iterator - used by printout routines who present trace
77 * results to users and which routines might sleep, etc:
79 struct trace_iterator {
80 struct trace_array *tr;
82 struct array_buffer *array_buffer;
86 struct ring_buffer_iter **buffer_iter;
87 unsigned long iter_flags;
89 /* trace_seq for __print_flags() and __print_symbolic() etc. */
90 struct trace_seq tmp_seq;
92 cpumask_var_t started;
94 /* it's true when current open file is snapshot */
97 /* The below is zeroed out in pipe_read */
99 struct trace_entry *ent;
100 unsigned long lost_events;
109 /* All new field here will be zeroed out in pipe_read */
112 enum trace_iter_flags {
113 TRACE_FILE_LAT_FMT = 1,
114 TRACE_FILE_ANNOTATE = 2,
115 TRACE_FILE_TIME_IN_NS = 4,
119 typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
120 int flags, struct trace_event *event);
122 struct trace_event_functions {
123 trace_print_func trace;
124 trace_print_func raw;
125 trace_print_func hex;
126 trace_print_func binary;
130 struct hlist_node node;
131 struct list_head list;
133 struct trace_event_functions *funcs;
136 extern int register_trace_event(struct trace_event *event);
137 extern int unregister_trace_event(struct trace_event *event);
139 /* Return values for print_line callback */
141 TRACE_TYPE_PARTIAL_LINE = 0, /* Retry after flushing the seq */
142 TRACE_TYPE_HANDLED = 1,
143 TRACE_TYPE_UNHANDLED = 2, /* Relay to other output functions */
144 TRACE_TYPE_NO_CONSUME = 3 /* Handled but ask to not consume */
147 enum print_line_t trace_handle_return(struct trace_seq *s);
149 void tracing_generic_entry_update(struct trace_entry *entry,
153 struct trace_event_file;
155 struct ring_buffer_event *
156 trace_event_buffer_lock_reserve(struct trace_buffer **current_buffer,
157 struct trace_event_file *trace_file,
158 int type, unsigned long len,
159 unsigned long flags, int pc);
161 #define TRACE_RECORD_CMDLINE BIT(0)
162 #define TRACE_RECORD_TGID BIT(1)
164 void tracing_record_taskinfo(struct task_struct *task, int flags);
165 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
166 struct task_struct *next, int flags);
168 void tracing_record_cmdline(struct task_struct *task);
169 void tracing_record_tgid(struct task_struct *task);
171 int trace_output_call(struct trace_iterator *iter, char *name, char *fmt, ...);
177 TRACE_REG_UNREGISTER,
178 #ifdef CONFIG_PERF_EVENTS
179 TRACE_REG_PERF_REGISTER,
180 TRACE_REG_PERF_UNREGISTER,
182 TRACE_REG_PERF_CLOSE,
184 * These (ADD/DEL) use a 'boolean' return value, where 1 (true) means a
185 * custom action was taken and the default action is not to be
193 struct trace_event_call;
195 struct trace_event_class {
198 #ifdef CONFIG_PERF_EVENTS
201 int (*reg)(struct trace_event_call *event,
202 enum trace_reg type, void *data);
203 int (*define_fields)(struct trace_event_call *);
204 struct list_head *(*get_fields)(struct trace_event_call *);
205 struct list_head fields;
206 int (*raw_init)(struct trace_event_call *);
209 extern int trace_event_reg(struct trace_event_call *event,
210 enum trace_reg type, void *data);
212 struct trace_event_buffer {
213 struct trace_buffer *buffer;
214 struct ring_buffer_event *event;
215 struct trace_event_file *trace_file;
219 struct pt_regs *regs;
222 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
223 struct trace_event_file *trace_file,
226 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer);
229 TRACE_EVENT_FL_FILTERED_BIT,
230 TRACE_EVENT_FL_CAP_ANY_BIT,
231 TRACE_EVENT_FL_NO_SET_FILTER_BIT,
232 TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
233 TRACE_EVENT_FL_TRACEPOINT_BIT,
234 TRACE_EVENT_FL_KPROBE_BIT,
235 TRACE_EVENT_FL_UPROBE_BIT,
240 * FILTERED - The event has a filter attached
241 * CAP_ANY - Any user can enable for perf
242 * NO_SET_FILTER - Set when filter has error and is to be ignored
243 * IGNORE_ENABLE - For trace internal events, do not enable with debugfs file
244 * TRACEPOINT - Event is a tracepoint
245 * KPROBE - Event is a kprobe
246 * UPROBE - Event is a uprobe
249 TRACE_EVENT_FL_FILTERED = (1 << TRACE_EVENT_FL_FILTERED_BIT),
250 TRACE_EVENT_FL_CAP_ANY = (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
251 TRACE_EVENT_FL_NO_SET_FILTER = (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
252 TRACE_EVENT_FL_IGNORE_ENABLE = (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
253 TRACE_EVENT_FL_TRACEPOINT = (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
254 TRACE_EVENT_FL_KPROBE = (1 << TRACE_EVENT_FL_KPROBE_BIT),
255 TRACE_EVENT_FL_UPROBE = (1 << TRACE_EVENT_FL_UPROBE_BIT),
258 #define TRACE_EVENT_FL_UKPROBE (TRACE_EVENT_FL_KPROBE | TRACE_EVENT_FL_UPROBE)
260 struct trace_event_call {
261 struct list_head list;
262 struct trace_event_class *class;
265 /* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
266 struct tracepoint *tp;
268 struct trace_event event;
270 struct event_filter *filter;
274 * bit 0: filter_active
275 * bit 1: allow trace by non root (cap any)
276 * bit 2: failed to apply filter
277 * bit 3: trace internal event (do not enable)
278 * bit 4: Event was enabled by module
279 * bit 5: use call filter rather than file filter
280 * bit 6: Event is a tracepoint
282 int flags; /* static flags of different events */
284 #ifdef CONFIG_PERF_EVENTS
286 struct hlist_head __percpu *perf_events;
287 struct bpf_prog_array __rcu *prog_array;
289 int (*perf_perm)(struct trace_event_call *,
290 struct perf_event *);
294 #ifdef CONFIG_PERF_EVENTS
295 static inline bool bpf_prog_array_valid(struct trace_event_call *call)
298 * This inline function checks whether call->prog_array
299 * is valid or not. The function is called in various places,
300 * outside rcu_read_lock/unlock, as a heuristic to speed up execution.
302 * If this function returns true, and later call->prog_array
303 * becomes false inside rcu_read_lock/unlock region,
304 * we bail out then. If this function return false,
305 * there is a risk that we might miss a few events if the checking
306 * were delayed until inside rcu_read_lock/unlock region and
307 * call->prog_array happened to become non-NULL then.
309 * Here, READ_ONCE() is used instead of rcu_access_pointer().
310 * rcu_access_pointer() requires the actual definition of
311 * "struct bpf_prog_array" while READ_ONCE() only needs
312 * a declaration of the same type.
314 return !!READ_ONCE(call->prog_array);
318 static inline const char *
319 trace_event_name(struct trace_event_call *call)
321 if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
322 return call->tp ? call->tp->name : NULL;
327 static inline struct list_head *
328 trace_get_fields(struct trace_event_call *event_call)
330 if (!event_call->class->get_fields)
331 return &event_call->class->fields;
332 return event_call->class->get_fields(event_call);
336 struct trace_subsystem_dir;
339 EVENT_FILE_FL_ENABLED_BIT,
340 EVENT_FILE_FL_RECORDED_CMD_BIT,
341 EVENT_FILE_FL_RECORDED_TGID_BIT,
342 EVENT_FILE_FL_FILTERED_BIT,
343 EVENT_FILE_FL_NO_SET_FILTER_BIT,
344 EVENT_FILE_FL_SOFT_MODE_BIT,
345 EVENT_FILE_FL_SOFT_DISABLED_BIT,
346 EVENT_FILE_FL_TRIGGER_MODE_BIT,
347 EVENT_FILE_FL_TRIGGER_COND_BIT,
348 EVENT_FILE_FL_PID_FILTER_BIT,
349 EVENT_FILE_FL_WAS_ENABLED_BIT,
354 * ENABLED - The event is enabled
355 * RECORDED_CMD - The comms should be recorded at sched_switch
356 * RECORDED_TGID - The tgids should be recorded at sched_switch
357 * FILTERED - The event has a filter attached
358 * NO_SET_FILTER - Set when filter has error and is to be ignored
359 * SOFT_MODE - The event is enabled/disabled by SOFT_DISABLED
360 * SOFT_DISABLED - When set, do not trace the event (even though its
361 * tracepoint may be enabled)
362 * TRIGGER_MODE - When set, invoke the triggers associated with the event
363 * TRIGGER_COND - When set, one or more triggers has an associated filter
364 * PID_FILTER - When set, the event is filtered based on pid
365 * WAS_ENABLED - Set when enabled to know to clear trace on module removal
368 EVENT_FILE_FL_ENABLED = (1 << EVENT_FILE_FL_ENABLED_BIT),
369 EVENT_FILE_FL_RECORDED_CMD = (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
370 EVENT_FILE_FL_RECORDED_TGID = (1 << EVENT_FILE_FL_RECORDED_TGID_BIT),
371 EVENT_FILE_FL_FILTERED = (1 << EVENT_FILE_FL_FILTERED_BIT),
372 EVENT_FILE_FL_NO_SET_FILTER = (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
373 EVENT_FILE_FL_SOFT_MODE = (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
374 EVENT_FILE_FL_SOFT_DISABLED = (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
375 EVENT_FILE_FL_TRIGGER_MODE = (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
376 EVENT_FILE_FL_TRIGGER_COND = (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
377 EVENT_FILE_FL_PID_FILTER = (1 << EVENT_FILE_FL_PID_FILTER_BIT),
378 EVENT_FILE_FL_WAS_ENABLED = (1 << EVENT_FILE_FL_WAS_ENABLED_BIT),
381 struct trace_event_file {
382 struct list_head list;
383 struct trace_event_call *event_call;
384 struct event_filter __rcu *filter;
386 struct trace_array *tr;
387 struct trace_subsystem_dir *system;
388 struct list_head triggers;
393 * bit 1: enabled cmd record
394 * bit 2: enable/disable with the soft disable bit
395 * bit 3: soft disabled
396 * bit 4: trigger enabled
398 * Note: The bits must be set atomically to prevent races
399 * from other writers. Reads of flags do not need to be in
400 * sync as they occur in critical sections. But the way flags
401 * is currently used, these changes do not affect the code
402 * except that when a change is made, it may have a slight
403 * delay in propagating the changes to other CPUs due to
404 * caching and such. Which is mostly OK ;-)
407 atomic_t sm_ref; /* soft-mode reference counter */
408 atomic_t tm_ref; /* trigger-mode reference counter */
411 #define __TRACE_EVENT_FLAGS(name, value) \
412 static int __init trace_init_flags_##name(void) \
414 event_##name.flags |= value; \
417 early_initcall(trace_init_flags_##name);
419 #define __TRACE_EVENT_PERF_PERM(name, expr...) \
420 static int perf_perm_##name(struct trace_event_call *tp_event, \
421 struct perf_event *p_event) \
423 return ({ expr; }); \
425 static int __init trace_init_perf_perm_##name(void) \
427 event_##name.perf_perm = &perf_perm_##name; \
430 early_initcall(trace_init_perf_perm_##name);
432 #define PERF_MAX_TRACE_SIZE 2048
434 #define MAX_FILTER_STR_VAL 256 /* Should handle KSYM_SYMBOL_LEN */
436 enum event_trigger_type {
438 ETT_TRACE_ONOFF = (1 << 0),
439 ETT_SNAPSHOT = (1 << 1),
440 ETT_STACKTRACE = (1 << 2),
441 ETT_EVENT_ENABLE = (1 << 3),
442 ETT_EVENT_HIST = (1 << 4),
443 ETT_HIST_ENABLE = (1 << 5),
446 extern int filter_match_preds(struct event_filter *filter, void *rec);
448 extern enum event_trigger_type
449 event_triggers_call(struct trace_event_file *file, void *rec,
450 struct ring_buffer_event *event);
452 event_triggers_post_call(struct trace_event_file *file,
453 enum event_trigger_type tt);
455 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file);
458 * trace_trigger_soft_disabled - do triggers and test if soft disabled
459 * @file: The file pointer of the event to test
461 * If any triggers without filters are attached to this event, they
462 * will be called here. If the event is soft disabled and has no
463 * triggers that require testing the fields, it will return true,
467 trace_trigger_soft_disabled(struct trace_event_file *file)
469 unsigned long eflags = file->flags;
471 if (!(eflags & EVENT_FILE_FL_TRIGGER_COND)) {
472 if (eflags & EVENT_FILE_FL_TRIGGER_MODE)
473 event_triggers_call(file, NULL, NULL);
474 if (eflags & EVENT_FILE_FL_SOFT_DISABLED)
476 if (eflags & EVENT_FILE_FL_PID_FILTER)
477 return trace_event_ignore_this_pid(file);
482 #ifdef CONFIG_BPF_EVENTS
483 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx);
484 int perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog);
485 void perf_event_detach_bpf_prog(struct perf_event *event);
486 int perf_event_query_prog_array(struct perf_event *event, void __user *info);
487 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
488 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
489 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name);
490 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp);
491 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
492 u32 *fd_type, const char **buf,
493 u64 *probe_offset, u64 *probe_addr);
495 static inline unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
501 perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog)
506 static inline void perf_event_detach_bpf_prog(struct perf_event *event) { }
509 perf_event_query_prog_array(struct perf_event *event, void __user *info)
513 static inline int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *p)
517 static inline int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *p)
521 static inline struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
525 static inline void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
528 static inline int bpf_get_perf_event_info(const struct perf_event *event,
529 u32 *prog_id, u32 *fd_type,
530 const char **buf, u64 *probe_offset,
539 FILTER_STATIC_STRING,
547 extern int trace_event_raw_init(struct trace_event_call *call);
548 extern int trace_define_field(struct trace_event_call *call, const char *type,
549 const char *name, int offset, int size,
550 int is_signed, int filter_type);
551 extern int trace_add_event_call(struct trace_event_call *call);
552 extern int trace_remove_event_call(struct trace_event_call *call);
553 extern int trace_event_get_offsets(struct trace_event_call *call);
555 #define is_signed_type(type) (((type)(-1)) < (type)1)
557 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set);
558 int trace_set_clr_event(const char *system, const char *event, int set);
559 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
560 const char *event, bool enable);
562 * The double __builtin_constant_p is because gcc will give us an error
563 * if we try to allocate the static variable to fmt if it is not a
564 * constant. Even with the outer if statement optimizing out.
566 #define event_trace_printk(ip, fmt, args...) \
568 __trace_printk_check_format(fmt, ##args); \
569 tracing_record_cmdline(current); \
570 if (__builtin_constant_p(fmt)) { \
571 static const char *trace_printk_fmt \
572 __attribute__((section("__trace_printk_fmt"))) = \
573 __builtin_constant_p(fmt) ? fmt : NULL; \
575 __trace_bprintk(ip, trace_printk_fmt, ##args); \
577 __trace_printk(ip, fmt, ##args); \
580 #ifdef CONFIG_PERF_EVENTS
583 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
584 DECLARE_PER_CPU(int, bpf_kprobe_override);
586 extern int perf_trace_init(struct perf_event *event);
587 extern void perf_trace_destroy(struct perf_event *event);
588 extern int perf_trace_add(struct perf_event *event, int flags);
589 extern void perf_trace_del(struct perf_event *event, int flags);
590 #ifdef CONFIG_KPROBE_EVENTS
591 extern int perf_kprobe_init(struct perf_event *event, bool is_retprobe);
592 extern void perf_kprobe_destroy(struct perf_event *event);
593 extern int bpf_get_kprobe_info(const struct perf_event *event,
594 u32 *fd_type, const char **symbol,
595 u64 *probe_offset, u64 *probe_addr,
596 bool perf_type_tracepoint);
598 #ifdef CONFIG_UPROBE_EVENTS
599 extern int perf_uprobe_init(struct perf_event *event,
600 unsigned long ref_ctr_offset, bool is_retprobe);
601 extern void perf_uprobe_destroy(struct perf_event *event);
602 extern int bpf_get_uprobe_info(const struct perf_event *event,
603 u32 *fd_type, const char **filename,
604 u64 *probe_offset, bool perf_type_tracepoint);
606 extern int ftrace_profile_set_filter(struct perf_event *event, int event_id,
608 extern void ftrace_profile_free_filter(struct perf_event *event);
609 void perf_trace_buf_update(void *record, u16 type);
610 void *perf_trace_buf_alloc(int size, struct pt_regs **regs, int *rctxp);
612 void bpf_trace_run1(struct bpf_prog *prog, u64 arg1);
613 void bpf_trace_run2(struct bpf_prog *prog, u64 arg1, u64 arg2);
614 void bpf_trace_run3(struct bpf_prog *prog, u64 arg1, u64 arg2,
616 void bpf_trace_run4(struct bpf_prog *prog, u64 arg1, u64 arg2,
618 void bpf_trace_run5(struct bpf_prog *prog, u64 arg1, u64 arg2,
619 u64 arg3, u64 arg4, u64 arg5);
620 void bpf_trace_run6(struct bpf_prog *prog, u64 arg1, u64 arg2,
621 u64 arg3, u64 arg4, u64 arg5, u64 arg6);
622 void bpf_trace_run7(struct bpf_prog *prog, u64 arg1, u64 arg2,
623 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7);
624 void bpf_trace_run8(struct bpf_prog *prog, u64 arg1, u64 arg2,
625 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
627 void bpf_trace_run9(struct bpf_prog *prog, u64 arg1, u64 arg2,
628 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
630 void bpf_trace_run10(struct bpf_prog *prog, u64 arg1, u64 arg2,
631 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
632 u64 arg8, u64 arg9, u64 arg10);
633 void bpf_trace_run11(struct bpf_prog *prog, u64 arg1, u64 arg2,
634 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
635 u64 arg8, u64 arg9, u64 arg10, u64 arg11);
636 void bpf_trace_run12(struct bpf_prog *prog, u64 arg1, u64 arg2,
637 u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
638 u64 arg8, u64 arg9, u64 arg10, u64 arg11, u64 arg12);
639 void perf_trace_run_bpf_submit(void *raw_data, int size, int rctx,
640 struct trace_event_call *call, u64 count,
641 struct pt_regs *regs, struct hlist_head *head,
642 struct task_struct *task);
645 perf_trace_buf_submit(void *raw_data, int size, int rctx, u16 type,
646 u64 count, struct pt_regs *regs, void *head,
647 struct task_struct *task)
649 perf_tp_event(type, count, raw_data, size, regs, head, rctx, task);
654 #endif /* _LINUX_TRACE_EVENT_H */