tracing/events: move declarations from trace directory to core include
Steven Rostedt [Mon, 13 Apr 2009 15:20:49 +0000 (11:20 -0400)]
In preparation to allowing trace events to happen in modules, we need
to move some of the local declarations in the kernel/trace directory
into include/linux.

This patch simply moves the declarations and performs no context changes.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

include/linux/ftrace_event.h [new file with mode: 0644]
kernel/trace/trace.h
kernel/trace/trace_output.h

diff --git a/include/linux/ftrace_event.h b/include/linux/ftrace_event.h
new file mode 100644 (file)
index 0000000..496b76d
--- /dev/null
@@ -0,0 +1,146 @@
+#ifndef _LINUX_FTRACE_EVENT_H
+#define _LINUX_FTRACE_EVENT_H
+
+#include <linux/trace_seq.h>
+#include <linux/ring_buffer.h>
+
+
+struct trace_array;
+struct tracer;
+
+/*
+ * The trace entry - the most basic unit of tracing. This is what
+ * is printed in the end as a single line in the trace output, such as:
+ *
+ *     bash-15816 [01]   235.197585: idle_cpu <- irq_enter
+ */
+struct trace_entry {
+       unsigned char           type;
+       unsigned char           flags;
+       unsigned char           preempt_count;
+       int                     pid;
+       int                     tgid;
+};
+
+/*
+ * Trace iterator - used by printout routines who present trace
+ * results to users and which routines might sleep, etc:
+ */
+struct trace_iterator {
+       struct trace_array      *tr;
+       struct tracer           *trace;
+       void                    *private;
+       int                     cpu_file;
+       struct mutex            mutex;
+       struct ring_buffer_iter *buffer_iter[NR_CPUS];
+
+       /* The below is zeroed out in pipe_read */
+       struct trace_seq        seq;
+       struct trace_entry      *ent;
+       int                     cpu;
+       u64                     ts;
+
+       unsigned long           iter_flags;
+       loff_t                  pos;
+       long                    idx;
+
+       cpumask_var_t           started;
+};
+
+
+typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
+                                             int flags);
+struct trace_event {
+       struct hlist_node       node;
+       int                     type;
+       trace_print_func        trace;
+       trace_print_func        raw;
+       trace_print_func        hex;
+       trace_print_func        binary;
+};
+
+extern int register_ftrace_event(struct trace_event *event);
+extern int unregister_ftrace_event(struct trace_event *event);
+
+/* Return values for print_line callback */
+enum print_line_t {
+       TRACE_TYPE_PARTIAL_LINE = 0,    /* Retry after flushing the seq */
+       TRACE_TYPE_HANDLED      = 1,
+       TRACE_TYPE_UNHANDLED    = 2,    /* Relay to other output functions */
+       TRACE_TYPE_NO_CONSUME   = 3     /* Handled but ask to not consume */
+};
+
+
+struct ring_buffer_event *
+trace_current_buffer_lock_reserve(unsigned char type, unsigned long len,
+                                 unsigned long flags, int pc);
+void trace_current_buffer_unlock_commit(struct ring_buffer_event *event,
+                                       unsigned long flags, int pc);
+void trace_nowake_buffer_unlock_commit(struct ring_buffer_event *event,
+                                       unsigned long flags, int pc);
+void trace_current_buffer_discard_commit(struct ring_buffer_event *event);
+
+void tracing_record_cmdline(struct task_struct *tsk);
+
+struct ftrace_event_call {
+       char                    *name;
+       char                    *system;
+       struct dentry           *dir;
+       int                     enabled;
+       int                     (*regfunc)(void);
+       void                    (*unregfunc)(void);
+       int                     id;
+       int                     (*raw_init)(void);
+       int                     (*show_format)(struct trace_seq *s);
+       int                     (*define_fields)(void);
+       struct list_head        fields;
+       int                     n_preds;
+       struct filter_pred      **preds;
+
+#ifdef CONFIG_EVENT_PROFILE
+       atomic_t        profile_count;
+       int             (*profile_enable)(struct ftrace_event_call *);
+       void            (*profile_disable)(struct ftrace_event_call *);
+#endif
+};
+
+#define MAX_FILTER_PRED                8
+#define MAX_FILTER_STR_VAL     128
+
+extern int init_preds(struct ftrace_event_call *call);
+extern int filter_match_preds(struct ftrace_event_call *call, void *rec);
+extern int filter_current_check_discard(struct ftrace_event_call *call,
+                                       void *rec,
+                                       struct ring_buffer_event *event);
+
+extern int trace_define_field(struct ftrace_event_call *call, char *type,
+                             char *name, int offset, int size);
+
+
+/*
+ * The double __builtin_constant_p is because gcc will give us an error
+ * if we try to allocate the static variable to fmt if it is not a
+ * constant. Even with the outer if statement optimizing out.
+ */
+#define event_trace_printk(ip, fmt, args...)                           \
+do {                                                                   \
+       __trace_printk_check_format(fmt, ##args);                       \
+       tracing_record_cmdline(current);                                \
+       if (__builtin_constant_p(fmt)) {                                \
+               static const char *trace_printk_fmt                     \
+                 __attribute__((section("__trace_printk_fmt"))) =      \
+                       __builtin_constant_p(fmt) ? fmt : NULL;         \
+                                                                       \
+               __trace_bprintk(ip, trace_printk_fmt, ##args);          \
+       } else                                                          \
+               __trace_printk(ip, fmt, ##args);                        \
+} while (0)
+
+#define __common_field(type, item)                                     \
+       ret = trace_define_field(event_call, #type, "common_" #item,    \
+                                offsetof(typeof(field.ent), item),     \
+                                sizeof(field.ent.item));               \
+       if (ret)                                                        \
+               return ret;
+
+#endif /* _LINUX_FTRACE_EVENT_H */
index 1882846..6bcdf4a 100644 (file)
@@ -13,6 +13,7 @@
 #include <trace/power.h>
 
 #include <linux/trace_seq.h>
+#include <linux/ftrace_event.h>
 
 enum trace_type {
        __TRACE_FIRST_TYPE = 0,
@@ -44,20 +45,6 @@ enum trace_type {
 };
 
 /*
- * The trace entry - the most basic unit of tracing. This is what
- * is printed in the end as a single line in the trace output, such as:
- *
- *     bash-15816 [01]   235.197585: idle_cpu <- irq_enter
- */
-struct trace_entry {
-       unsigned char           type;
-       unsigned char           flags;
-       unsigned char           preempt_count;
-       int                     pid;
-       int                     tgid;
-};
-
-/*
  * Function trace entry - function address and parent function addres:
  */
 struct ftrace_entry {
@@ -265,8 +252,6 @@ struct trace_array_cpu {
        char                    comm[TASK_COMM_LEN];
 };
 
-struct trace_iterator;
-
 /*
  * The trace array - an array of per-CPU trace arrays. This is the
  * highest level data structure that individual tracers deal with.
@@ -341,15 +326,6 @@ extern void __ftrace_bad_type(void);
                __ftrace_bad_type();                                    \
        } while (0)
 
-/* Return values for print_line callback */
-enum print_line_t {
-       TRACE_TYPE_PARTIAL_LINE = 0,    /* Retry after flushing the seq */
-       TRACE_TYPE_HANDLED      = 1,
-       TRACE_TYPE_UNHANDLED    = 2,    /* Relay to other output functions */
-       TRACE_TYPE_NO_CONSUME   = 3     /* Handled but ask to not consume */
-};
-
-
 /*
  * An option specific to a tracer. This is a boolean value.
  * The bit is the bit index that sets its value on the
@@ -428,31 +404,6 @@ struct tracer {
 
 #define TRACE_PIPE_ALL_CPU     -1
 
-/*
- * Trace iterator - used by printout routines who present trace
- * results to users and which routines might sleep, etc:
- */
-struct trace_iterator {
-       struct trace_array      *tr;
-       struct tracer           *trace;
-       void                    *private;
-       int                     cpu_file;
-       struct mutex            mutex;
-       struct ring_buffer_iter *buffer_iter[NR_CPUS];
-
-       /* The below is zeroed out in pipe_read */
-       struct trace_seq        seq;
-       struct trace_entry      *ent;
-       int                     cpu;
-       u64                     ts;
-
-       unsigned long           iter_flags;
-       loff_t                  pos;
-       long                    idx;
-
-       cpumask_var_t           started;
-};
-
 int tracer_init(struct tracer *t, struct trace_array *tr);
 int tracing_is_enabled(void);
 void trace_wake_up(void);
@@ -479,15 +430,6 @@ void trace_buffer_unlock_commit(struct trace_array *tr,
                                struct ring_buffer_event *event,
                                unsigned long flags, int pc);
 
-struct ring_buffer_event *
-trace_current_buffer_lock_reserve(unsigned char type, unsigned long len,
-                                 unsigned long flags, int pc);
-void trace_current_buffer_unlock_commit(struct ring_buffer_event *event,
-                                       unsigned long flags, int pc);
-void trace_nowake_buffer_unlock_commit(struct ring_buffer_event *event,
-                                       unsigned long flags, int pc);
-void trace_current_buffer_discard_commit(struct ring_buffer_event *event);
-
 struct trace_entry *tracing_get_trace_entry(struct trace_array *tr,
                                                struct trace_array_cpu *data);
 
@@ -510,7 +452,6 @@ void tracing_sched_switch_trace(struct trace_array *tr,
                                struct task_struct *prev,
                                struct task_struct *next,
                                unsigned long flags, int pc);
-void tracing_record_cmdline(struct task_struct *tsk);
 
 void tracing_sched_wakeup_trace(struct trace_array *tr,
                                struct task_struct *wakee,
@@ -790,28 +731,6 @@ struct ftrace_event_field {
        int                     size;
 };
 
-struct ftrace_event_call {
-       char                    *name;
-       char                    *system;
-       struct dentry           *dir;
-       int                     enabled;
-       int                     (*regfunc)(void);
-       void                    (*unregfunc)(void);
-       int                     id;
-       int                     (*raw_init)(void);
-       int                     (*show_format)(struct trace_seq *s);
-       int                     (*define_fields)(void);
-       struct list_head        fields;
-       int                     n_preds;
-       struct filter_pred      **preds;
-
-#ifdef CONFIG_EVENT_PROFILE
-       atomic_t        profile_count;
-       int             (*profile_enable)(struct ftrace_event_call *);
-       void            (*profile_disable)(struct ftrace_event_call *);
-#endif
-};
-
 struct event_subsystem {
        struct list_head        list;
        const char              *name;
@@ -825,9 +744,6 @@ struct event_subsystem {
             (unsigned long)event < (unsigned long)__stop_ftrace_events; \
             event++)
 
-#define MAX_FILTER_PRED                8
-#define MAX_FILTER_STR_VAL     128
-
 struct filter_pred;
 
 typedef int (*filter_pred_fn_t) (struct filter_pred *pred, void *event);
@@ -845,9 +761,6 @@ struct filter_pred {
        int clear;
 };
 
-int trace_define_field(struct ftrace_event_call *call, char *type,
-                      char *name, int offset, int size);
-extern int init_preds(struct ftrace_event_call *call);
 extern void filter_free_pred(struct filter_pred *pred);
 extern void filter_print_preds(struct filter_pred **preds, int n_preds,
                               struct trace_seq *s);
@@ -855,13 +768,9 @@ extern int filter_parse(char **pbuf, struct filter_pred *pred);
 extern int filter_add_pred(struct ftrace_event_call *call,
                           struct filter_pred *pred);
 extern void filter_disable_preds(struct ftrace_event_call *call);
-extern int filter_match_preds(struct ftrace_event_call *call, void *rec);
 extern void filter_free_subsystem_preds(struct event_subsystem *system);
 extern int filter_add_subsystem_pred(struct event_subsystem *system,
                                     struct filter_pred *pred);
-extern int filter_current_check_discard(struct ftrace_event_call *call,
-                                       void *rec,
-                                       struct ring_buffer_event *event);
 
 static inline int
 filter_check_discard(struct ftrace_event_call *call, void *rec,
@@ -876,14 +785,6 @@ filter_check_discard(struct ftrace_event_call *call, void *rec,
        return 0;
 }
 
-#define __common_field(type, item)                                     \
-       ret = trace_define_field(event_call, #type, "common_" #item,    \
-                                offsetof(typeof(field.ent), item),     \
-                                sizeof(field.ent.item));               \
-       if (ret)                                                        \
-               return ret;
-
-void event_trace_printk(unsigned long ip, const char *fmt, ...);
 extern struct ftrace_event_call __start_ftrace_events[];
 extern struct ftrace_event_call __stop_ftrace_events[];
 
@@ -895,25 +796,6 @@ extern struct ftrace_event_call __stop_ftrace_events[];
 extern const char *__start___trace_bprintk_fmt[];
 extern const char *__stop___trace_bprintk_fmt[];
 
-/*
- * The double __builtin_constant_p is because gcc will give us an error
- * if we try to allocate the static variable to fmt if it is not a
- * constant. Even with the outer if statement optimizing out.
- */
-#define event_trace_printk(ip, fmt, args...)                           \
-do {                                                                   \
-       __trace_printk_check_format(fmt, ##args);                       \
-       tracing_record_cmdline(current);                                \
-       if (__builtin_constant_p(fmt)) {                                \
-               static const char *trace_printk_fmt                     \
-                 __attribute__((section("__trace_printk_fmt"))) =      \
-                       __builtin_constant_p(fmt) ? fmt : NULL;         \
-                                                                       \
-               __trace_bprintk(ip, trace_printk_fmt, ##args);          \
-       } else                                                          \
-               __trace_printk(ip, fmt, ##args);                        \
-} while (0)
-
 #undef TRACE_EVENT_FORMAT
 #define TRACE_EVENT_FORMAT(call, proto, args, fmt, tstruct, tpfmt)     \
        extern struct ftrace_event_call event_##call;
index 5c7cbfb..6e220a8 100644 (file)
@@ -4,18 +4,6 @@
 #include <linux/trace_seq.h>
 #include "trace.h"
 
-typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
-                                             int flags);
-
-struct trace_event {
-       struct hlist_node       node;
-       int                     type;
-       trace_print_func        trace;
-       trace_print_func        raw;
-       trace_print_func        hex;
-       trace_print_func        binary;
-};
-
 extern enum print_line_t
 trace_print_bprintk_msg_only(struct trace_iterator *iter);
 extern enum print_line_t
@@ -33,8 +21,6 @@ extern int trace_print_context(struct trace_iterator *iter);
 extern int trace_print_lat_context(struct trace_iterator *iter);
 
 extern struct trace_event *ftrace_find_event(int type);
-extern int register_ftrace_event(struct trace_event *event);
-extern int unregister_ftrace_event(struct trace_event *event);
 
 extern enum print_line_t trace_nop_print(struct trace_iterator *iter,
                                         int flags);