Documentation/trace/tracepoints.txt

   1                      Using the Linux Kernel Tracepoints
   2
   3                             Mathieu Desnoyers
   4
   5
   6 This document introduces Linux Kernel Tracepoints and their use. It
   7 provides examples of how to insert tracepoints in the kernel and
   8 connect probe functions to them and provides some examples of probe
   9 functions.
  10
  11
  12 * Purpose of tracepoints
  13
  14 A tracepoint placed in code provides a hook to call a function (probe)
  15 that you can provide at runtime. A tracepoint can be "on" (a probe is
  16 connected to it) or "off" (no probe is attached). When a tracepoint is
  17 "off" it has no effect, except for adding a tiny time penalty
  18 (checking a condition for a branch) and space penalty (adding a few
  19 bytes for the function call at the end of the instrumented function
  20 and adds a data structure in a separate section).  When a tracepoint
  21 is "on", the function you provide is called each time the tracepoint
  22 is executed, in the execution context of the caller. When the function
  23 provided ends its execution, it returns to the caller (continuing from
  24 the tracepoint site).
  25
  26 You can put tracepoints at important locations in the code. They are
  27 lightweight hooks that can pass an arbitrary number of parameters,
  28 which prototypes are described in a tracepoint declaration placed in a
  29 header file.
  30
  31 They can be used for tracing and performance accounting.
  32
  33
  34 * Usage
  35
  36 Two elements are required for tracepoints :
  37
  38 - A tracepoint definition, placed in a header file.
  39 - The tracepoint statement, in C code.
  40
  41 In order to use tracepoints, you should include linux/tracepoint.h.
  42
  43 In include/trace/events/subsys.h :
  44
  45 #undef TRACE_SYSTEM
  46 #define TRACE_SYSTEM subsys
  47
  48 #if !defined(_TRACE_SUBSYS_H) || defined(TRACE_HEADER_MULTI_READ)
  49 #define _TRACE_SUBSYS_H
  50
  51 #include <linux/tracepoint.h>
  52
  53 DECLARE_TRACE(subsys_eventname,
  54         TP_PROTO(int firstarg, struct task_struct *p),
  55         TP_ARGS(firstarg, p));
  56
  57 #endif /* _TRACE_SUBSYS_H */
  58
  59 /* This part must be outside protection */
  60 #include <trace/define_trace.h>
  61
  62 In subsys/file.c (where the tracing statement must be added) :
  63
  64 #include <trace/events/subsys.h>
  65
  66 #define CREATE_TRACE_POINTS
  67 DEFINE_TRACE(subsys_eventname);
  68
  69 void somefct(void)
  70 {
  71         ...
  72         trace_subsys_eventname(arg, task);
  73         ...
  74 }
  75
  76 Where :
  77 - subsys_eventname is an identifier unique to your event
  78     - subsys is the name of your subsystem.
  79     - eventname is the name of the event to trace.
  80
  81 - TP_PROTO(int firstarg, struct task_struct *p) is the prototype of the
  82   function called by this tracepoint.
  83
  84 - TP_ARGS(firstarg, p) are the parameters names, same as found in the
  85   prototype.
  86
  87 - if you use the header in multiple source files, #define CREATE_TRACE_POINTS
  88   should appear only in one source file.
  89
  90 Connecting a function (probe) to a tracepoint is done by providing a
  91 probe (function to call) for the specific tracepoint through
  92 register_trace_subsys_eventname().  Removing a probe is done through
  93 unregister_trace_subsys_eventname(); it will remove the probe.
  94
  95 tracepoint_synchronize_unregister() must be called before the end of
  96 the module exit function to make sure there is no caller left using
  97 the probe. This, and the fact that preemption is disabled around the
  98 probe call, make sure that probe removal and module unload are safe.
  99
 100 The tracepoint mechanism supports inserting multiple instances of the
 101 same tracepoint, but a single definition must be made of a given
 102 tracepoint name over all the kernel to make sure no type conflict will
 103 occur. Name mangling of the tracepoints is done using the prototypes
 104 to make sure typing is correct. Verification of probe type correctness
 105 is done at the registration site by the compiler. Tracepoints can be
 106 put in inline functions, inlined static functions, and unrolled loops
 107 as well as regular functions.
 108
 109 The naming scheme "subsys_event" is suggested here as a convention
 110 intended to limit collisions. Tracepoint names are global to the
 111 kernel: they are considered as being the same whether they are in the
 112 core kernel image or in modules.
 113
 114 If the tracepoint has to be used in kernel modules, an
 115 EXPORT_TRACEPOINT_SYMBOL_GPL() or EXPORT_TRACEPOINT_SYMBOL() can be
 116 used to export the defined tracepoints.
 117
 118 If you need to do a bit of work for a tracepoint parameter, and
 119 that work is only used for the tracepoint, that work can be encapsulated
 120 within an if statement with the following:
 121
 122         if (trace_foo_bar_enabled()) {
 123                 int i;
 124                 int tot = 0;
 125
 126                 for (i = 0; i < count; i++)
 127                         tot += calculate_nuggets();
 128
 129                 trace_foo_bar(tot);
 130         }
 131
 132 All trace_<tracepoint>() calls have a matching trace_<tracepoint>_enabled()
 133 function defined that returns true if the tracepoint is enabled and
 134 false otherwise. The trace_<tracepoint>() should always be within the
 135 block of the if (trace_<tracepoint>_enabled()) to prevent races between
 136 the tracepoint being enabled and the check being seen.
 137
 138 The advantage of using the trace_<tracepoint>_enabled() is that it uses
 139 the static_key of the tracepoint to allow the if statement to be implemented
 140 with jump labels and avoid conditional branches.
 141
 142 Note: The convenience macro TRACE_EVENT provides an alternative way to
 143       define tracepoints. Check http://lwn.net/Articles/379903,
 144       http://lwn.net/Articles/381064 and http://lwn.net/Articles/383362
 145       for a series of articles with more details.