net: Avoid extra wakeups of threads blocked in wait_for_packet()
[linux-2.6.git] / include / linux / tracepoint.h
1 #ifndef _LINUX_TRACEPOINT_H
2 #define _LINUX_TRACEPOINT_H
3
4 /*
5  * Kernel Tracepoint API.
6  *
7  * See Documentation/tracepoint.txt.
8  *
9  * (C) Copyright 2008 Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
10  *
11  * Heavily inspired from the Linux Kernel Markers.
12  *
13  * This file is released under the GPLv2.
14  * See the file COPYING for more details.
15  */
16
17 #include <linux/types.h>
18 #include <linux/rcupdate.h>
19
20 struct module;
21 struct tracepoint;
22
23 struct tracepoint {
24         const char *name;               /* Tracepoint name */
25         int state;                      /* State. */
26         void **funcs;
27 } __attribute__((aligned(32)));         /*
28                                          * Aligned on 32 bytes because it is
29                                          * globally visible and gcc happily
30                                          * align these on the structure size.
31                                          * Keep in sync with vmlinux.lds.h.
32                                          */
33
34 #define TP_PROTO(args...)       args
35 #define TP_ARGS(args...)                args
36
37 #ifdef CONFIG_TRACEPOINTS
38
39 /*
40  * it_func[0] is never NULL because there is at least one element in the array
41  * when the array itself is non NULL.
42  */
43 #define __DO_TRACE(tp, proto, args)                                     \
44         do {                                                            \
45                 void **it_func;                                         \
46                                                                         \
47                 rcu_read_lock_sched_notrace();                          \
48                 it_func = rcu_dereference((tp)->funcs);                 \
49                 if (it_func) {                                          \
50                         do {                                            \
51                                 ((void(*)(proto))(*it_func))(args);     \
52                         } while (*(++it_func));                         \
53                 }                                                       \
54                 rcu_read_unlock_sched_notrace();                        \
55         } while (0)
56
57 /*
58  * Make sure the alignment of the structure in the __tracepoints section will
59  * not add unwanted padding between the beginning of the section and the
60  * structure. Force alignment to the same alignment as the section start.
61  */
62 #define DECLARE_TRACE(name, proto, args)                                \
63         extern struct tracepoint __tracepoint_##name;                   \
64         static inline void trace_##name(proto)                          \
65         {                                                               \
66                 if (unlikely(__tracepoint_##name.state))                \
67                         __DO_TRACE(&__tracepoint_##name,                \
68                                 TP_PROTO(proto), TP_ARGS(args));        \
69         }                                                               \
70         static inline int register_trace_##name(void (*probe)(proto))   \
71         {                                                               \
72                 return tracepoint_probe_register(#name, (void *)probe); \
73         }                                                               \
74         static inline int unregister_trace_##name(void (*probe)(proto)) \
75         {                                                               \
76                 return tracepoint_probe_unregister(#name, (void *)probe);\
77         }
78
79 #define DEFINE_TRACE(name)                                              \
80         static const char __tpstrtab_##name[]                           \
81         __attribute__((section("__tracepoints_strings"))) = #name;      \
82         struct tracepoint __tracepoint_##name                           \
83         __attribute__((section("__tracepoints"), aligned(32))) =        \
84                 { __tpstrtab_##name, 0, NULL }
85
86 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)                              \
87         EXPORT_SYMBOL_GPL(__tracepoint_##name)
88 #define EXPORT_TRACEPOINT_SYMBOL(name)                                  \
89         EXPORT_SYMBOL(__tracepoint_##name)
90
91 extern void tracepoint_update_probe_range(struct tracepoint *begin,
92         struct tracepoint *end);
93
94 #else /* !CONFIG_TRACEPOINTS */
95 #define DECLARE_TRACE(name, proto, args)                                \
96         static inline void _do_trace_##name(struct tracepoint *tp, proto) \
97         { }                                                             \
98         static inline void trace_##name(proto)                          \
99         { }                                                             \
100         static inline int register_trace_##name(void (*probe)(proto))   \
101         {                                                               \
102                 return -ENOSYS;                                         \
103         }                                                               \
104         static inline int unregister_trace_##name(void (*probe)(proto)) \
105         {                                                               \
106                 return -ENOSYS;                                         \
107         }
108
109 #define DEFINE_TRACE(name)
110 #define EXPORT_TRACEPOINT_SYMBOL_GPL(name)
111 #define EXPORT_TRACEPOINT_SYMBOL(name)
112
113 static inline void tracepoint_update_probe_range(struct tracepoint *begin,
114         struct tracepoint *end)
115 { }
116 #endif /* CONFIG_TRACEPOINTS */
117
118 /*
119  * Connect a probe to a tracepoint.
120  * Internal API, should not be used directly.
121  */
122 extern int tracepoint_probe_register(const char *name, void *probe);
123
124 /*
125  * Disconnect a probe from a tracepoint.
126  * Internal API, should not be used directly.
127  */
128 extern int tracepoint_probe_unregister(const char *name, void *probe);
129
130 extern int tracepoint_probe_register_noupdate(const char *name, void *probe);
131 extern int tracepoint_probe_unregister_noupdate(const char *name, void *probe);
132 extern void tracepoint_probe_update_all(void);
133
134 struct tracepoint_iter {
135         struct module *module;
136         struct tracepoint *tracepoint;
137 };
138
139 extern void tracepoint_iter_start(struct tracepoint_iter *iter);
140 extern void tracepoint_iter_next(struct tracepoint_iter *iter);
141 extern void tracepoint_iter_stop(struct tracepoint_iter *iter);
142 extern void tracepoint_iter_reset(struct tracepoint_iter *iter);
143 extern int tracepoint_get_iter_range(struct tracepoint **tracepoint,
144         struct tracepoint *begin, struct tracepoint *end);
145
146 /*
147  * tracepoint_synchronize_unregister must be called between the last tracepoint
148  * probe unregistration and the end of module exit to make sure there is no
149  * caller executing a probe when it is freed.
150  */
151 static inline void tracepoint_synchronize_unregister(void)
152 {
153         synchronize_sched();
154 }
155
156 #define PARAMS(args...) args
157 #define TRACE_FORMAT(name, proto, args, fmt)            \
158         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
159
160
161 /*
162  * For use with the TRACE_EVENT macro:
163  *
164  * We define a tracepoint, its arguments, its printk format
165  * and its 'fast binay record' layout.
166  *
167  * Firstly, name your tracepoint via TRACE_EVENT(name : the
168  * 'subsystem_event' notation is fine.
169  *
170  * Think about this whole construct as the
171  * 'trace_sched_switch() function' from now on.
172  *
173  *
174  *  TRACE_EVENT(sched_switch,
175  *
176  *      *
177  *      * A function has a regular function arguments
178  *      * prototype, declare it via TP_PROTO():
179  *      *
180  *
181  *      TP_PROTO(struct rq *rq, struct task_struct *prev,
182  *               struct task_struct *next),
183  *
184  *      *
185  *      * Define the call signature of the 'function'.
186  *      * (Design sidenote: we use this instead of a
187  *      *  TP_PROTO1/TP_PROTO2/TP_PROTO3 ugliness.)
188  *      *
189  *
190  *      TP_ARGS(rq, prev, next),
191  *
192  *      *
193  *      * Fast binary tracing: define the trace record via
194  *      * TP_STRUCT__entry(). You can think about it like a
195  *      * regular C structure local variable definition.
196  *      *
197  *      * This is how the trace record is structured and will
198  *      * be saved into the ring buffer. These are the fields
199  *      * that will be exposed to user-space in
200  *      * /debug/tracing/events/<*>/format.
201  *      *
202  *      * The declared 'local variable' is called '__entry'
203  *      *
204  *      * __field(pid_t, prev_prid) is equivalent to a standard declariton:
205  *      *
206  *      *       pid_t   prev_pid;
207  *      *
208  *      * __array(char, prev_comm, TASK_COMM_LEN) is equivalent to:
209  *      *
210  *      *       char    prev_comm[TASK_COMM_LEN];
211  *      *
212  *
213  *      TP_STRUCT__entry(
214  *              __array(        char,   prev_comm,      TASK_COMM_LEN   )
215  *              __field(        pid_t,  prev_pid                        )
216  *              __field(        int,    prev_prio                       )
217  *              __array(        char,   next_comm,      TASK_COMM_LEN   )
218  *              __field(        pid_t,  next_pid                        )
219  *              __field(        int,    next_prio                       )
220  *      ),
221  *
222  *      *
223  *      * Assign the entry into the trace record, by embedding
224  *      * a full C statement block into TP_fast_assign(). You
225  *      * can refer to the trace record as '__entry' -
226  *      * otherwise you can put arbitrary C code in here.
227  *      *
228  *      * Note: this C code will execute every time a trace event
229  *      * happens, on an active tracepoint.
230  *      *
231  *
232  *      TP_fast_assign(
233  *              memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
234  *              __entry->prev_pid       = prev->pid;
235  *              __entry->prev_prio      = prev->prio;
236  *              memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
237  *              __entry->next_pid       = next->pid;
238  *              __entry->next_prio      = next->prio;
239  *      )
240  *
241  *      *
242  *      * Formatted output of a trace record via TP_printk().
243  *      * This is how the tracepoint will appear under ftrace
244  *      * plugins that make use of this tracepoint.
245  *      *
246  *      * (raw-binary tracing wont actually perform this step.)
247  *      *
248  *
249  *      TP_printk("task %s:%d [%d] ==> %s:%d [%d]",
250  *              __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
251  *              __entry->next_comm, __entry->next_pid, __entry->next_prio),
252  *
253  * );
254  *
255  * This macro construct is thus used for the regular printk format
256  * tracing setup, it is used to construct a function pointer based
257  * tracepoint callback (this is used by programmatic plugins and
258  * can also by used by generic instrumentation like SystemTap), and
259  * it is also used to expose a structured trace record in
260  * /debug/tracing/events/.
261  */
262
263 #define TRACE_EVENT(name, proto, args, struct, assign, print)   \
264         DECLARE_TRACE(name, PARAMS(proto), PARAMS(args))
265
266 #endif