docs/tracing.txt

   1 = Tracing =
   2
   3 == Introduction ==
   4
   5 This document describes the tracing infrastructure in QEMU and how to use it
   6 for debugging, profiling, and observing execution.
   7
   8 == Quickstart ==
   9
  10 1. Build with the 'simple' trace backend:
  11
  12     ./configure --enable-trace-backend=simple
  13     make
  14
  15 2. Create a file with the events you want to trace:
  16
  17    echo bdrv_aio_readv   > /tmp/events
  18    echo bdrv_aio_writev >> /tmp/events
  19
  20 3. Run the virtual machine to produce a trace file:
  21
  22     qemu -trace events=/tmp/events ... # your normal QEMU invocation
  23
  24 4. Pretty-print the binary trace file:
  25
  26     ./simpletrace.py trace-events trace-*
  27
  28 == Trace events ==
  29
  30 There is a set of static trace events declared in the "trace-events" source
  31 file.  Each trace event declaration names the event, its arguments, and the
  32 format string which can be used for pretty-printing:
  33
  34     qemu_vmalloc(size_t size, void *ptr) "size %zu ptr %p"
  35     qemu_vfree(void *ptr) "ptr %p"
  36
  37 The "trace-events" file is processed by the "tracetool" script during build to
  38 generate code for the trace events.  Trace events are invoked directly from
  39 source code like this:
  40
  41     #include "trace.h"  /* needed for trace event prototype */
  42
  43     void *qemu_vmalloc(size_t size)
  44     {
  45         void *ptr;
  46         size_t align = QEMU_VMALLOC_ALIGN;
  47
  48         if (size < align) {
  49             align = getpagesize();
  50         }
  51         ptr = qemu_memalign(align, size);
  52         trace_qemu_vmalloc(size, ptr);
  53         return ptr;
  54     }
  55
  56 === Declaring trace events ===
  57
  58 The "tracetool" script produces the trace.h header file which is included by
  59 every source file that uses trace events.  Since many source files include
  60 trace.h, it uses a minimum of types and other header files included to keep the
  61 namespace clean and compile times and dependencies down.
  62
  63 Trace events should use types as follows:
  64
  65  * Use stdint.h types for fixed-size types.  Most offsets and guest memory
  66    addresses are best represented with uint32_t or uint64_t.  Use fixed-size
  67    types over primitive types whose size may change depending on the host
  68    (32-bit versus 64-bit) so trace events don't truncate values or break
  69    the build.
  70
  71  * Use void * for pointers to structs or for arrays.  The trace.h header
  72    cannot include all user-defined struct declarations and it is therefore
  73    necessary to use void * for pointers to structs.
  74
  75  * For everything else, use primitive scalar types (char, int, long) with the
  76    appropriate signedness.
  77
  78 Format strings should reflect the types defined in the trace event.  Take
  79 special care to use PRId64 and PRIu64 for int64_t and uint64_t types,
  80 respectively.  This ensures portability between 32- and 64-bit platforms.
  81
  82 === Hints for adding new trace events ===
  83
  84 1. Trace state changes in the code.  Interesting points in the code usually
  85    involve a state change like starting, stopping, allocating, freeing.  State
  86    changes are good trace events because they can be used to understand the
  87    execution of the system.
  88
  89 2. Trace guest operations.  Guest I/O accesses like reading device registers
  90    are good trace events because they can be used to understand guest
  91    interactions.
  92
  93 3. Use correlator fields so the context of an individual line of trace output
  94    can be understood.  For example, trace the pointer returned by malloc and
  95    used as an argument to free.  This way mallocs and frees can be matched up.
  96    Trace events with no context are not very useful.
  97
  98 4. Name trace events after their function.  If there are multiple trace events
  99    in one function, append a unique distinguisher at the end of the name.
 100
 101 == Generic interface and monitor commands ==
 102
 103 You can programmatically query and control the dynamic state of trace events
 104 through a backend-agnostic interface:
 105
 106 * trace_print_events
 107
 108 * trace_event_set_state
 109   Enables or disables trace events at runtime inside QEMU.
 110   The function returns "true" if the state of the event has been successfully
 111   changed, or "false" otherwise:
 112
 113     #include "trace/control.h"
 114
 115     trace_event_set_state("virtio_irq", true); /* enable */
 116     [...]
 117     trace_event_set_state("virtio_irq", false); /* disable */
 118
 119 Note that some of the backends do not provide an implementation for this
 120 interface, in which case QEMU will just print a warning.
 121
 122 This functionality is also provided through monitor commands:
 123
 124 * info trace-events
 125   View available trace events and their state.  State 1 means enabled, state 0
 126   means disabled.
 127
 128 * trace-event NAME on|off
 129   Enable/disable a given trace event or a group of events having common prefix
 130   through wildcard.
 131
 132 The "-trace events=<file>" command line argument can be used to enable the
 133 events listed in <file> from the very beginning of the program. This file must
 134 contain one event name per line.
 135
 136 A basic wildcard matching is supported in both the monitor command "trace
 137 -event" and the events list file. That means you can enable/disable the events
 138 having a common prefix in a batch. For example, virtio-blk trace events could
 139 be enabled using:
 140   trace-event virtio_blk_* on
 141
 142 If a line in the "-trace events=<file>" file begins with a '-', the trace event
 143 will be disabled instead of enabled.  This is useful when a wildcard was used
 144 to enable an entire family of events but one noisy event needs to be disabled.
 145
 146 == Trace backends ==
 147
 148 The "tracetool" script automates tedious trace event code generation and also
 149 keeps the trace event declarations independent of the trace backend.  The trace
 150 events are not tightly coupled to a specific trace backend, such as LTTng or
 151 SystemTap.  Support for trace backends can be added by extending the "tracetool"
 152 script.
 153
 154 The trace backend is chosen at configure time and only one trace backend can
 155 be built into the binary:
 156
 157     ./configure --trace-backend=simple
 158
 159 For a list of supported trace backends, try ./configure --help or see below.
 160
 161 The following subsections describe the supported trace backends.
 162
 163 === Nop ===
 164
 165 The "nop" backend generates empty trace event functions so that the compiler
 166 can optimize out trace events completely.  This is the default and imposes no
 167 performance penalty.
 168
 169 Note that regardless of the selected trace backend, events with the "disable"
 170 property will be generated with the "nop" backend.
 171
 172 === Stderr ===
 173
 174 The "stderr" backend sends trace events directly to standard error.  This
 175 effectively turns trace events into debug printfs.
 176
 177 This is the simplest backend and can be used together with existing code that
 178 uses DPRINTF().
 179
 180 === Simpletrace ===
 181
 182 The "simple" backend supports common use cases and comes as part of the QEMU
 183 source tree.  It may not be as powerful as platform-specific or third-party
 184 trace backends but it is portable.  This is the recommended trace backend
 185 unless you have specific needs for more advanced backends.
 186
 187 The "simple" backend currently does not capture string arguments, it simply
 188 records the char* pointer value instead of the string that is pointed to.
 189
 190 ==== Monitor commands ====
 191
 192 * info trace
 193   Display the contents of trace buffer.  This command dumps the trace buffer
 194   with simple formatting.  For full pretty-printing, use the simpletrace.py
 195   script on a binary trace file.
 196
 197   The trace buffer is written into until full.  The full trace buffer is
 198   flushed and emptied.  This means the 'info trace' will display few or no
 199   entries if the buffer has just been flushed.
 200
 201 * trace-file on|off|flush|set <path>
 202   Enable/disable/flush the trace file or set the trace file name.
 203
 204 ==== Analyzing trace files ====
 205
 206 The "simple" backend produces binary trace files that can be formatted with the
 207 simpletrace.py script.  The script takes the "trace-events" file and the binary
 208 trace:
 209
 210     ./simpletrace.py trace-events trace-12345
 211
 212 You must ensure that the same "trace-events" file was used to build QEMU,
 213 otherwise trace event declarations may have changed and output will not be
 214 consistent.
 215
 216 === LTTng Userspace Tracer ===
 217
 218 The "ust" backend uses the LTTng Userspace Tracer library.  There are no
 219 monitor commands built into QEMU, instead UST utilities should be used to list,
 220 enable/disable, and dump traces.
 221
 222 === SystemTap ===
 223
 224 The "dtrace" backend uses DTrace sdt probes but has only been tested with
 225 SystemTap.  When SystemTap support is detected a .stp file with wrapper probes
 226 is generated to make use in scripts more convenient.  This step can also be
 227 performed manually after a build in order to change the binary name in the .stp
 228 probes:
 229
 230     scripts/tracetool --dtrace --stap \
 231                       --binary path/to/qemu-binary \
 232                       --target-type system \
 233                       --target-arch x86_64 \
 234                       <trace-events >qemu.stp
 235
 236 == Trace event properties ==
 237
 238 Each event in the "trace-events" file can be prefixed with a space-separated
 239 list of zero or more of the following event properties.
 240
 241 === "disable" ===
 242
 243 If a specific trace event is going to be invoked a huge number of times, this
 244 might have a noticeable performance impact even when the event is
 245 programmatically disabled.
 246
 247 In this case you should declare such event with the "disable" property. This
 248 will effectively disable the event at compile time (by using the "nop" backend),
 249 thus having no performance impact at all on regular builds (i.e., unless you
 250 edit the "trace-events" file).
 251
 252 In addition, there might be cases where relatively complex computations must be
 253 performed to generate values that are only used as arguments for a trace
 254 function. In these cases you can use the macro 'TRACE_${EVENT_NAME}_ENABLED' to
 255 guard such computations and avoid its compilation when the event is disabled:
 256
 257     #include "trace.h"  /* needed for trace event prototype */
 258
 259     void *qemu_vmalloc(size_t size)
 260     {
 261         void *ptr;
 262         size_t align = QEMU_VMALLOC_ALIGN;
 263
 264         if (size < align) {
 265             align = getpagesize();
 266         }
 267         ptr = qemu_memalign(align, size);
 268         if (TRACE_QEMU_VMALLOC_ENABLED) { /* preprocessor macro */
 269             void *complex;
 270             /* some complex computations to produce the 'complex' value */
 271             trace_qemu_vmalloc(size, ptr, complex);
 272         }
 273         return ptr;
 274     }