tracing: Modify soft-mode only if there's no other referrer
[linux-3.10.git] / include / linux / ftrace_event.h
1 #ifndef _LINUX_FTRACE_EVENT_H
2 #define _LINUX_FTRACE_EVENT_H
3
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>
9
10 struct trace_array;
11 struct trace_buffer;
12 struct tracer;
13 struct dentry;
14
15 struct trace_print_flags {
16         unsigned long           mask;
17         const char              *name;
18 };
19
20 struct trace_print_flags_u64 {
21         unsigned long long      mask;
22         const char              *name;
23 };
24
25 const char *ftrace_print_flags_seq(struct trace_seq *p, const char *delim,
26                                    unsigned long flags,
27                                    const struct trace_print_flags *flag_array);
28
29 const char *ftrace_print_symbols_seq(struct trace_seq *p, unsigned long val,
30                                      const struct trace_print_flags *symbol_array);
31
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
36                                                                  *symbol_array);
37 #endif
38
39 const char *ftrace_print_hex_seq(struct trace_seq *p,
40                                  const unsigned char *buf, int len);
41
42 struct trace_iterator;
43 struct trace_event;
44
45 int ftrace_raw_output_prep(struct trace_iterator *iter,
46                            struct trace_event *event);
47
48 /*
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:
51  *
52  *     bash-15816 [01]   235.197585: idle_cpu <- irq_enter
53  */
54 struct trace_entry {
55         unsigned short          type;
56         unsigned char           flags;
57         unsigned char           preempt_count;
58         int                     pid;
59 };
60
61 #define FTRACE_MAX_EVENT                                                \
62         ((1 << (sizeof(((struct trace_entry *)0)->type) * 8)) - 1)
63
64 /*
65  * Trace iterator - used by printout routines who present trace
66  * results to users and which routines might sleep, etc:
67  */
68 struct trace_iterator {
69         struct trace_array      *tr;
70         struct tracer           *trace;
71         struct trace_buffer     *trace_buffer;
72         void                    *private;
73         int                     cpu_file;
74         struct mutex            mutex;
75         struct ring_buffer_iter **buffer_iter;
76         unsigned long           iter_flags;
77
78         /* trace_seq for __print_flags() and __print_symbolic() etc. */
79         struct trace_seq        tmp_seq;
80
81         /* The below is zeroed out in pipe_read */
82         struct trace_seq        seq;
83         struct trace_entry      *ent;
84         unsigned long           lost_events;
85         int                     leftover;
86         int                     ent_size;
87         int                     cpu;
88         u64                     ts;
89
90         loff_t                  pos;
91         long                    idx;
92
93         cpumask_var_t           started;
94
95         /* it's true when current open file is snapshot */
96         bool                    snapshot;
97 };
98
99 enum trace_iter_flags {
100         TRACE_FILE_LAT_FMT      = 1,
101         TRACE_FILE_ANNOTATE     = 2,
102         TRACE_FILE_TIME_IN_NS   = 4,
103 };
104
105
106 typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
107                                       int flags, struct trace_event *event);
108
109 struct trace_event_functions {
110         trace_print_func        trace;
111         trace_print_func        raw;
112         trace_print_func        hex;
113         trace_print_func        binary;
114 };
115
116 struct trace_event {
117         struct hlist_node               node;
118         struct list_head                list;
119         int                             type;
120         struct trace_event_functions    *funcs;
121 };
122
123 extern int register_ftrace_event(struct trace_event *event);
124 extern int unregister_ftrace_event(struct trace_event *event);
125
126 /* Return values for print_line callback */
127 enum print_line_t {
128         TRACE_TYPE_PARTIAL_LINE = 0,    /* Retry after flushing the seq */
129         TRACE_TYPE_HANDLED      = 1,
130         TRACE_TYPE_UNHANDLED    = 2,    /* Relay to other output functions */
131         TRACE_TYPE_NO_CONSUME   = 3     /* Handled but ask to not consume */
132 };
133
134 void tracing_generic_entry_update(struct trace_entry *entry,
135                                   unsigned long flags,
136                                   int pc);
137 struct ftrace_event_file;
138
139 struct ring_buffer_event *
140 trace_event_buffer_lock_reserve(struct ring_buffer **current_buffer,
141                                 struct ftrace_event_file *ftrace_file,
142                                 int type, unsigned long len,
143                                 unsigned long flags, int pc);
144 struct ring_buffer_event *
145 trace_current_buffer_lock_reserve(struct ring_buffer **current_buffer,
146                                   int type, unsigned long len,
147                                   unsigned long flags, int pc);
148 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
149                                         struct ring_buffer_event *event,
150                                         unsigned long flags, int pc);
151 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
152                                 struct ring_buffer_event *event,
153                                 unsigned long flags, int pc);
154 void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer,
155                                      struct ring_buffer_event *event,
156                                      unsigned long flags, int pc,
157                                      struct pt_regs *regs);
158 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
159                                          struct ring_buffer_event *event);
160
161 void tracing_record_cmdline(struct task_struct *tsk);
162
163 struct event_filter;
164
165 enum trace_reg {
166         TRACE_REG_REGISTER,
167         TRACE_REG_UNREGISTER,
168 #ifdef CONFIG_PERF_EVENTS
169         TRACE_REG_PERF_REGISTER,
170         TRACE_REG_PERF_UNREGISTER,
171         TRACE_REG_PERF_OPEN,
172         TRACE_REG_PERF_CLOSE,
173         TRACE_REG_PERF_ADD,
174         TRACE_REG_PERF_DEL,
175 #endif
176 };
177
178 struct ftrace_event_call;
179
180 struct ftrace_event_class {
181         char                    *system;
182         void                    *probe;
183 #ifdef CONFIG_PERF_EVENTS
184         void                    *perf_probe;
185 #endif
186         int                     (*reg)(struct ftrace_event_call *event,
187                                        enum trace_reg type, void *data);
188         int                     (*define_fields)(struct ftrace_event_call *);
189         struct list_head        *(*get_fields)(struct ftrace_event_call *);
190         struct list_head        fields;
191         int                     (*raw_init)(struct ftrace_event_call *);
192 };
193
194 extern int ftrace_event_reg(struct ftrace_event_call *event,
195                             enum trace_reg type, void *data);
196
197 enum {
198         TRACE_EVENT_FL_FILTERED_BIT,
199         TRACE_EVENT_FL_CAP_ANY_BIT,
200         TRACE_EVENT_FL_NO_SET_FILTER_BIT,
201         TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
202         TRACE_EVENT_FL_WAS_ENABLED_BIT,
203 };
204
205 /*
206  * Event flags:
207  *  FILTERED      - The event has a filter attached
208  *  CAP_ANY       - Any user can enable for perf
209  *  NO_SET_FILTER - Set when filter has error and is to be ignored
210  *  IGNORE_ENABLE - For ftrace internal events, do not enable with debugfs file
211  *  WAS_ENABLED   - Set and stays set when an event was ever enabled
212  *                    (used for module unloading, if a module event is enabled,
213  *                     it is best to clear the buffers that used it).
214  */
215 enum {
216         TRACE_EVENT_FL_FILTERED         = (1 << TRACE_EVENT_FL_FILTERED_BIT),
217         TRACE_EVENT_FL_CAP_ANY          = (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
218         TRACE_EVENT_FL_NO_SET_FILTER    = (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
219         TRACE_EVENT_FL_IGNORE_ENABLE    = (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
220         TRACE_EVENT_FL_WAS_ENABLED      = (1 << TRACE_EVENT_FL_WAS_ENABLED_BIT),
221 };
222
223 struct ftrace_event_call {
224         struct list_head        list;
225         struct ftrace_event_class *class;
226         char                    *name;
227         struct trace_event      event;
228         const char              *print_fmt;
229         struct event_filter     *filter;
230         struct list_head        *files;
231         void                    *mod;
232         void                    *data;
233         /*
234          *   bit 0:             filter_active
235          *   bit 1:             allow trace by non root (cap any)
236          *   bit 2:             failed to apply filter
237          *   bit 3:             ftrace internal event (do not enable)
238          *   bit 4:             Event was enabled by module
239          */
240         int                     flags; /* static flags of different events */
241
242 #ifdef CONFIG_PERF_EVENTS
243         int                             perf_refcount;
244         struct hlist_head __percpu      *perf_events;
245 #endif
246 };
247
248 struct trace_array;
249 struct ftrace_subsystem_dir;
250
251 enum {
252         FTRACE_EVENT_FL_ENABLED_BIT,
253         FTRACE_EVENT_FL_RECORDED_CMD_BIT,
254         FTRACE_EVENT_FL_SOFT_MODE_BIT,
255         FTRACE_EVENT_FL_SOFT_DISABLED_BIT,
256 };
257
258 /*
259  * Ftrace event file flags:
260  *  ENABLED       - The event is enabled
261  *  RECORDED_CMD  - The comms should be recorded at sched_switch
262  *  SOFT_MODE     - The event is enabled/disabled by SOFT_DISABLED
263  *  SOFT_DISABLED - When set, do not trace the event (even though its
264  *                   tracepoint may be enabled)
265  */
266 enum {
267         FTRACE_EVENT_FL_ENABLED         = (1 << FTRACE_EVENT_FL_ENABLED_BIT),
268         FTRACE_EVENT_FL_RECORDED_CMD    = (1 << FTRACE_EVENT_FL_RECORDED_CMD_BIT),
269         FTRACE_EVENT_FL_SOFT_MODE       = (1 << FTRACE_EVENT_FL_SOFT_MODE_BIT),
270         FTRACE_EVENT_FL_SOFT_DISABLED   = (1 << FTRACE_EVENT_FL_SOFT_DISABLED_BIT),
271 };
272
273 struct ftrace_event_file {
274         struct list_head                list;
275         struct ftrace_event_call        *event_call;
276         struct dentry                   *dir;
277         struct trace_array              *tr;
278         struct ftrace_subsystem_dir     *system;
279
280         /*
281          * 32 bit flags:
282          *   bit 0:             enabled
283          *   bit 1:             enabled cmd record
284          *   bit 2:             enable/disable with the soft disable bit
285          *   bit 3:             soft disabled
286          *
287          * Note: The bits must be set atomically to prevent races
288          * from other writers. Reads of flags do not need to be in
289          * sync as they occur in critical sections. But the way flags
290          * is currently used, these changes do not affect the code
291          * except that when a change is made, it may have a slight
292          * delay in propagating the changes to other CPUs due to
293          * caching and such. Which is mostly OK ;-)
294          */
295         unsigned long           flags;
296         atomic_t                sm_ref; /* soft-mode reference counter */
297 };
298
299 #define __TRACE_EVENT_FLAGS(name, value)                                \
300         static int __init trace_init_flags_##name(void)                 \
301         {                                                               \
302                 event_##name.flags = value;                             \
303                 return 0;                                               \
304         }                                                               \
305         early_initcall(trace_init_flags_##name);
306
307 #define PERF_MAX_TRACE_SIZE     2048
308
309 #define MAX_FILTER_STR_VAL      256     /* Should handle KSYM_SYMBOL_LEN */
310
311 extern void destroy_preds(struct ftrace_event_call *call);
312 extern int filter_match_preds(struct event_filter *filter, void *rec);
313 extern int filter_current_check_discard(struct ring_buffer *buffer,
314                                         struct ftrace_event_call *call,
315                                         void *rec,
316                                         struct ring_buffer_event *event);
317
318 enum {
319         FILTER_OTHER = 0,
320         FILTER_STATIC_STRING,
321         FILTER_DYN_STRING,
322         FILTER_PTR_STRING,
323         FILTER_TRACE_FN,
324 };
325
326 #define EVENT_STORAGE_SIZE 128
327 extern struct mutex event_storage_mutex;
328 extern char event_storage[EVENT_STORAGE_SIZE];
329
330 extern int trace_event_raw_init(struct ftrace_event_call *call);
331 extern int trace_define_field(struct ftrace_event_call *call, const char *type,
332                               const char *name, int offset, int size,
333                               int is_signed, int filter_type);
334 extern int trace_add_event_call(struct ftrace_event_call *call);
335 extern void trace_remove_event_call(struct ftrace_event_call *call);
336
337 #define is_signed_type(type)    (((type)(-1)) < (type)1)
338
339 int trace_set_clr_event(const char *system, const char *event, int set);
340
341 /*
342  * The double __builtin_constant_p is because gcc will give us an error
343  * if we try to allocate the static variable to fmt if it is not a
344  * constant. Even with the outer if statement optimizing out.
345  */
346 #define event_trace_printk(ip, fmt, args...)                            \
347 do {                                                                    \
348         __trace_printk_check_format(fmt, ##args);                       \
349         tracing_record_cmdline(current);                                \
350         if (__builtin_constant_p(fmt)) {                                \
351                 static const char *trace_printk_fmt                     \
352                   __attribute__((section("__trace_printk_fmt"))) =      \
353                         __builtin_constant_p(fmt) ? fmt : NULL;         \
354                                                                         \
355                 __trace_bprintk(ip, trace_printk_fmt, ##args);          \
356         } else                                                          \
357                 __trace_printk(ip, fmt, ##args);                        \
358 } while (0)
359
360 #ifdef CONFIG_PERF_EVENTS
361 struct perf_event;
362
363 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
364
365 extern int  perf_trace_init(struct perf_event *event);
366 extern void perf_trace_destroy(struct perf_event *event);
367 extern int  perf_trace_add(struct perf_event *event, int flags);
368 extern void perf_trace_del(struct perf_event *event, int flags);
369 extern int  ftrace_profile_set_filter(struct perf_event *event, int event_id,
370                                      char *filter_str);
371 extern void ftrace_profile_free_filter(struct perf_event *event);
372 extern void *perf_trace_buf_prepare(int size, unsigned short type,
373                                     struct pt_regs *regs, int *rctxp);
374
375 static inline void
376 perf_trace_buf_submit(void *raw_data, int size, int rctx, u64 addr,
377                        u64 count, struct pt_regs *regs, void *head,
378                        struct task_struct *task)
379 {
380         perf_tp_event(addr, count, raw_data, size, regs, head, rctx, task);
381 }
382 #endif
383
384 #endif /* _LINUX_FTRACE_EVENT_H */