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,
209 * FILTERED - The event has a filter attached
210 * CAP_ANY - Any user can enable for perf
211 * NO_SET_FILTER - Set when filter has error and is to be ignored
212 * IGNORE_ENABLE - For ftrace internal events, do not enable with debugfs file
213 * WAS_ENABLED - Set and stays set when an event was ever enabled
214 * (used for module unloading, if a module event is enabled,
215 * it is best to clear the buffers that used it).
218 TRACE_EVENT_FL_FILTERED = (1 << TRACE_EVENT_FL_FILTERED_BIT),
219 TRACE_EVENT_FL_CAP_ANY = (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
220 TRACE_EVENT_FL_NO_SET_FILTER = (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
221 TRACE_EVENT_FL_IGNORE_ENABLE = (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
222 TRACE_EVENT_FL_WAS_ENABLED = (1 << TRACE_EVENT_FL_WAS_ENABLED_BIT),
225 struct ftrace_event_call {
226 struct list_head list;
227 struct ftrace_event_class *class;
229 struct trace_event event;
230 const char *print_fmt;
231 struct event_filter *filter;
232 struct list_head *files;
236 * bit 0: filter_active
237 * bit 1: allow trace by non root (cap any)
238 * bit 2: failed to apply filter
239 * bit 3: ftrace internal event (do not enable)
240 * bit 4: Event was enabled by module
242 int flags; /* static flags of different events */
244 #ifdef CONFIG_PERF_EVENTS
246 struct hlist_head __percpu *perf_events;
251 struct ftrace_subsystem_dir;
254 FTRACE_EVENT_FL_ENABLED_BIT,
255 FTRACE_EVENT_FL_RECORDED_CMD_BIT,
256 FTRACE_EVENT_FL_SOFT_MODE_BIT,
257 FTRACE_EVENT_FL_SOFT_DISABLED_BIT,
261 * Ftrace event file flags:
262 * ENABLED - The event is enabled
263 * RECORDED_CMD - The comms should be recorded at sched_switch
264 * SOFT_MODE - The event is enabled/disabled by SOFT_DISABLED
265 * SOFT_DISABLED - When set, do not trace the event (even though its
266 * tracepoint may be enabled)
269 FTRACE_EVENT_FL_ENABLED = (1 << FTRACE_EVENT_FL_ENABLED_BIT),
270 FTRACE_EVENT_FL_RECORDED_CMD = (1 << FTRACE_EVENT_FL_RECORDED_CMD_BIT),
271 FTRACE_EVENT_FL_SOFT_MODE = (1 << FTRACE_EVENT_FL_SOFT_MODE_BIT),
272 FTRACE_EVENT_FL_SOFT_DISABLED = (1 << FTRACE_EVENT_FL_SOFT_DISABLED_BIT),
275 struct ftrace_event_file {
276 struct list_head list;
277 struct ftrace_event_call *event_call;
279 struct trace_array *tr;
280 struct ftrace_subsystem_dir *system;
285 * bit 1: enabled cmd record
286 * bit 2: enable/disable with the soft disable bit
287 * bit 3: soft disabled
289 * Note: The bits must be set atomically to prevent races
290 * from other writers. Reads of flags do not need to be in
291 * sync as they occur in critical sections. But the way flags
292 * is currently used, these changes do not affect the code
293 * except that when a change is made, it may have a slight
294 * delay in propagating the changes to other CPUs due to
295 * caching and such. Which is mostly OK ;-)
298 atomic_t sm_ref; /* soft-mode reference counter */
301 #define __TRACE_EVENT_FLAGS(name, value) \
302 static int __init trace_init_flags_##name(void) \
304 event_##name.flags = value; \
307 early_initcall(trace_init_flags_##name);
309 #define PERF_MAX_TRACE_SIZE 2048
311 #define MAX_FILTER_STR_VAL 256 /* Should handle KSYM_SYMBOL_LEN */
313 extern void destroy_preds(struct ftrace_event_call *call);
314 extern int filter_match_preds(struct event_filter *filter, void *rec);
315 extern int filter_current_check_discard(struct ring_buffer *buffer,
316 struct ftrace_event_call *call,
318 struct ring_buffer_event *event);
322 FILTER_STATIC_STRING,
328 #define EVENT_STORAGE_SIZE 128
329 extern struct mutex event_storage_mutex;
330 extern char event_storage[EVENT_STORAGE_SIZE];
332 extern int trace_event_raw_init(struct ftrace_event_call *call);
333 extern int trace_define_field(struct ftrace_event_call *call, const char *type,
334 const char *name, int offset, int size,
335 int is_signed, int filter_type);
336 extern int trace_add_event_call(struct ftrace_event_call *call);
337 extern int trace_remove_event_call(struct ftrace_event_call *call);
339 #define is_signed_type(type) (((type)(-1)) < (type)1)
341 int trace_set_clr_event(const char *system, const char *event, int set);
344 * The double __builtin_constant_p is because gcc will give us an error
345 * if we try to allocate the static variable to fmt if it is not a
346 * constant. Even with the outer if statement optimizing out.
348 #define event_trace_printk(ip, fmt, args...) \
350 __trace_printk_check_format(fmt, ##args); \
351 tracing_record_cmdline(current); \
352 if (__builtin_constant_p(fmt)) { \
353 static const char *trace_printk_fmt \
354 __attribute__((section("__trace_printk_fmt"))) = \
355 __builtin_constant_p(fmt) ? fmt : NULL; \
357 __trace_bprintk(ip, trace_printk_fmt, ##args); \
359 __trace_printk(ip, fmt, ##args); \
363 * tracepoint_string - register constant persistent string to trace system
364 * @str - a constant persistent string that will be referenced in tracepoints
366 * If constant strings are being used in tracepoints, it is faster and
367 * more efficient to just save the pointer to the string and reference
368 * that with a printf "%s" instead of saving the string in the ring buffer
369 * and wasting space and time.
371 * The problem with the above approach is that userspace tools that read
372 * the binary output of the trace buffers do not have access to the string.
373 * Instead they just show the address of the string which is not very
376 * With tracepoint_string(), the string will be registered to the tracing
377 * system and exported to userspace via the debugfs/tracing/printk_formats
378 * file that maps the string address to the string text. This way userspace
379 * tools that read the binary buffers have a way to map the pointers to
380 * the ASCII strings they represent.
382 * The @str used must be a constant string and persistent as it would not
383 * make sense to show a string that no longer exists. But it is still fine
384 * to be used with modules, because when modules are unloaded, if they
385 * had tracepoints, the ring buffers are cleared too. As long as the string
386 * does not change during the life of the module, it is fine to use
387 * tracepoint_string() within a module.
389 #define tracepoint_string(str) \
391 static const char *___tp_str __tracepoint_string = str; \
394 #define __tracepoint_string __attribute__((section("__tracepoint_str")))
396 #ifdef CONFIG_PERF_EVENTS
399 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
401 extern int perf_trace_init(struct perf_event *event);
402 extern void perf_trace_destroy(struct perf_event *event);
403 extern int perf_trace_add(struct perf_event *event, int flags);
404 extern void perf_trace_del(struct perf_event *event, int flags);
405 extern int ftrace_profile_set_filter(struct perf_event *event, int event_id,
407 extern void ftrace_profile_free_filter(struct perf_event *event);
408 extern void *perf_trace_buf_prepare(int size, unsigned short type,
409 struct pt_regs *regs, int *rctxp);
412 perf_trace_buf_submit(void *raw_data, int size, int rctx, u64 addr,
413 u64 count, struct pt_regs *regs, void *head,
414 struct task_struct *task)
416 perf_tp_event(addr, count, raw_data, size, regs, head, rctx, task);
420 #endif /* _LINUX_FTRACE_EVENT_H */