1.. _tracing: 2 3======= 4Tracing 5======= 6 7Introduction 8============ 9 10This document describes the tracing infrastructure in QEMU and how to use it 11for debugging, profiling, and observing execution. 12 13Quickstart 14========== 15 16Enable tracing of ``memory_region_ops_read`` and ``memory_region_ops_write`` 17events:: 18 19 $ qemu --trace "memory_region_ops_*" ... 20 ... 21 719585@1608130130.441188:memory_region_ops_read cpu 0 mr 0x562fdfbb3820 addr 0x3cc value 0x67 size 1 22 719585@1608130130.441190:memory_region_ops_write cpu 0 mr 0x562fdfbd2f00 addr 0x3d4 value 0x70e size 2 23 24This output comes from the "log" trace backend that is enabled by default when 25``./configure --enable-trace-backends=BACKENDS`` was not explicitly specified. 26 27Multiple patterns can be specified by repeating the ``--trace`` option:: 28 29 $ qemu --trace "kvm_*" --trace "virtio_*" ... 30 31When patterns are used frequently it is more convenient to store them in a 32file to avoid long command-line options:: 33 34 $ echo "memory_region_ops_*" >/tmp/events 35 $ echo "kvm_*" >>/tmp/events 36 $ qemu --trace events=/tmp/events ... 37 38Trace events 39============ 40 41Sub-directory setup 42------------------- 43 44Each directory in the source tree can declare a set of trace events in a local 45"trace-events" file. All directories which contain "trace-events" files must be 46listed in the "trace_events_subdirs" variable in the top level meson.build 47file. During build, the "trace-events" file in each listed subdirectory will be 48processed by the "tracetool" script to generate code for the trace events. 49 50The individual "trace-events" files are merged into a "trace-events-all" file, 51which is also installed into "/usr/share/qemu". 52This merged file is to be used by the "simpletrace.py" script to later analyse 53traces in the simpletrace data format. 54 55The following files are automatically generated in <builddir>/trace/ during the 56build: 57 58 - trace-<subdir>.c - the trace event state declarations 59 - trace-<subdir>.h - the trace event enums and probe functions 60 - trace-dtrace-<subdir>.h - DTrace event probe specification 61 - trace-dtrace-<subdir>.dtrace - DTrace event probe helper declaration 62 - trace-dtrace-<subdir>.o - binary DTrace provider (generated by dtrace) 63 - trace-ust-<subdir>.h - UST event probe helper declarations 64 65Here <subdir> is the sub-directory path with '/' replaced by '_'. For example, 66"accel/kvm" becomes "accel_kvm" and the final filename for "trace-<subdir>.c" 67becomes "trace-accel_kvm.c". 68 69Source files in the source tree do not directly include generated files in 70"<builddir>/trace/". Instead they #include the local "trace.h" file, without 71any sub-directory path prefix. eg io/channel-buffer.c would do:: 72 73 #include "trace.h" 74 75The "io/trace.h" file must be created manually with an #include of the 76corresponding "trace/trace-<subdir>.h" file that will be generated in the 77builddir:: 78 79 $ echo '#include "trace/trace-io.h"' >io/trace.h 80 81While it is possible to include a trace.h file from outside a source file's own 82sub-directory, this is discouraged in general. It is strongly preferred that 83all events be declared directly in the sub-directory that uses them. The only 84exception is where there are some shared trace events defined in the top level 85directory trace-events file. The top level directory generates trace files 86with a filename prefix of "trace/trace-root" instead of just "trace". This is 87to avoid ambiguity between a trace.h in the current directory, vs the top level 88directory. 89 90Using trace events 91------------------ 92 93Trace events are invoked directly from source code like this:: 94 95 #include "trace.h" /* needed for trace event prototype */ 96 97 void *qemu_vmalloc(size_t size) 98 { 99 void *ptr; 100 size_t align = QEMU_VMALLOC_ALIGN; 101 102 if (size < align) { 103 align = getpagesize(); 104 } 105 ptr = qemu_memalign(align, size); 106 trace_qemu_vmalloc(size, ptr); 107 return ptr; 108 } 109 110Declaring trace events 111---------------------- 112 113The "tracetool" script produces the trace.h header file which is included by 114every source file that uses trace events. Since many source files include 115trace.h, it uses a minimum of types and other header files included to keep the 116namespace clean and compile times and dependencies down. 117 118Trace events should use types as follows: 119 120 * Use stdint.h types for fixed-size types. Most offsets and guest memory 121 addresses are best represented with uint32_t or uint64_t. Use fixed-size 122 types over primitive types whose size may change depending on the host 123 (32-bit versus 64-bit) so trace events don't truncate values or break 124 the build. 125 126 * Use void * for pointers to structs or for arrays. The trace.h header 127 cannot include all user-defined struct declarations and it is therefore 128 necessary to use void * for pointers to structs. 129 130 * For everything else, use primitive scalar types (char, int, long) with the 131 appropriate signedness. 132 133 * Avoid floating point types (float and double) because SystemTap does not 134 support them. In most cases it is possible to round to an integer type 135 instead. This may require scaling the value first by multiplying it by 1000 136 or the like when digits after the decimal point need to be preserved. 137 138Format strings should reflect the types defined in the trace event. Take 139special care to use PRId64 and PRIu64 for int64_t and uint64_t types, 140respectively. This ensures portability between 32- and 64-bit platforms. 141Format strings must not end with a newline character. It is the responsibility 142of backends to adapt line ending for proper logging. 143 144Each event declaration will start with the event name, then its arguments, 145finally a format string for pretty-printing. For example:: 146 147 qemu_vmalloc(size_t size, void *ptr) "size %zu ptr %p" 148 qemu_vfree(void *ptr) "ptr %p" 149 150 151Hints for adding new trace events 152--------------------------------- 153 1541. Trace state changes in the code. Interesting points in the code usually 155 involve a state change like starting, stopping, allocating, freeing. State 156 changes are good trace events because they can be used to understand the 157 execution of the system. 158 1592. Trace guest operations. Guest I/O accesses like reading device registers 160 are good trace events because they can be used to understand guest 161 interactions. 162 1633. Use correlator fields so the context of an individual line of trace output 164 can be understood. For example, trace the pointer returned by malloc and 165 used as an argument to free. This way mallocs and frees can be matched up. 166 Trace events with no context are not very useful. 167 1684. Name trace events after their function. If there are multiple trace events 169 in one function, append a unique distinguisher at the end of the name. 170 171Generic interface and monitor commands 172====================================== 173 174You can programmatically query and control the state of trace events through a 175backend-agnostic interface provided by the header "trace/control.h". 176 177Note that some of the backends do not provide an implementation for some parts 178of this interface, in which case QEMU will just print a warning (please refer to 179header "trace/control.h" to see which routines are backend-dependent). 180 181The state of events can also be queried and modified through monitor commands: 182 183* ``info trace-events`` 184 View available trace events and their state. State 1 means enabled, state 0 185 means disabled. 186 187* ``trace-event NAME on|off`` 188 Enable/disable a given trace event or a group of events (using wildcards). 189 190The "--trace events=<file>" command line argument can be used to enable the 191events listed in <file> from the very beginning of the program. This file must 192contain one event name per line. 193 194If a line in the "--trace events=<file>" file begins with a '-', the trace event 195will be disabled instead of enabled. This is useful when a wildcard was used 196to enable an entire family of events but one noisy event needs to be disabled. 197 198Wildcard matching is supported in both the monitor command "trace-event" and the 199events list file. That means you can enable/disable the events having a common 200prefix in a batch. For example, virtio-blk trace events could be enabled using 201the following monitor command:: 202 203 trace-event virtio_blk_* on 204 205Trace backends 206============== 207 208The "tracetool" script automates tedious trace event code generation and also 209keeps the trace event declarations independent of the trace backend. The trace 210events are not tightly coupled to a specific trace backend, such as LTTng or 211SystemTap. Support for trace backends can be added by extending the "tracetool" 212script. 213 214The trace backends are chosen at configure time:: 215 216 ./configure --enable-trace-backends=simple,dtrace 217 218For a list of supported trace backends, try ./configure --help or see below. 219If multiple backends are enabled, the trace is sent to them all. 220 221If no backends are explicitly selected, configure will default to the 222"log" backend. 223 224The following subsections describe the supported trace backends. 225 226Nop 227--- 228 229The "nop" backend generates empty trace event functions so that the compiler 230can optimize out trace events completely. This imposes no performance 231penalty. 232 233Note that regardless of the selected trace backend, events with the "disable" 234property will be generated with the "nop" backend. 235 236Log 237--- 238 239The "log" backend sends trace events directly to standard error. This 240effectively turns trace events into debug printfs. 241 242This is the simplest backend and can be used together with existing code that 243uses DPRINTF(). 244 245The -msg timestamp=on|off command-line option controls whether or not to print 246the tid/timestamp prefix for each trace event. 247 248Simpletrace 249----------- 250 251The "simple" backend writes binary trace logs to a file from a thread, making 252it lower overhead than the "log" backend. A Python API is available for writing 253offline trace file analysis scripts. It may not be as powerful as 254platform-specific or third-party trace backends but it is portable and has no 255special library dependencies. 256 257Monitor commands 258~~~~~~~~~~~~~~~~ 259 260* ``trace-file on|off|flush|set <path>`` 261 Enable/disable/flush the trace file or set the trace file name. 262 263Analyzing trace files 264~~~~~~~~~~~~~~~~~~~~~ 265 266The "simple" backend produces binary trace files that can be formatted with the 267simpletrace.py script. The script takes the "trace-events-all" file and the 268binary trace:: 269 270 ./scripts/simpletrace.py trace-events-all trace-12345 271 272You must ensure that the same "trace-events-all" file was used to build QEMU, 273otherwise trace event declarations may have changed and output will not be 274consistent. 275 276Ftrace 277------ 278 279The "ftrace" backend writes trace data to ftrace marker. This effectively 280sends trace events to ftrace ring buffer, and you can compare qemu trace 281data and kernel(especially kvm.ko when using KVM) trace data. 282 283if you use KVM, enable kvm events in ftrace:: 284 285 # echo 1 > /sys/kernel/debug/tracing/events/kvm/enable 286 287After running qemu by root user, you can get the trace:: 288 289 # cat /sys/kernel/debug/tracing/trace 290 291Restriction: "ftrace" backend is restricted to Linux only. 292 293Syslog 294------ 295 296The "syslog" backend sends trace events using the POSIX syslog API. The log 297is opened specifying the LOG_DAEMON facility and LOG_PID option (so events 298are tagged with the pid of the particular QEMU process that generated 299them). All events are logged at LOG_INFO level. 300 301NOTE: syslog may squash duplicate consecutive trace events and apply rate 302 limiting. 303 304Restriction: "syslog" backend is restricted to POSIX compliant OS. 305 306LTTng Userspace Tracer 307---------------------- 308 309The "ust" backend uses the LTTng Userspace Tracer library. There are no 310monitor commands built into QEMU, instead UST utilities should be used to list, 311enable/disable, and dump traces. 312 313Package lttng-tools is required for userspace tracing. You must ensure that the 314current user belongs to the "tracing" group, or manually launch the 315lttng-sessiond daemon for the current user prior to running any instance of 316QEMU. 317 318While running an instrumented QEMU, LTTng should be able to list all available 319events:: 320 321 lttng list -u 322 323Create tracing session:: 324 325 lttng create mysession 326 327Enable events:: 328 329 lttng enable-event qemu:g_malloc -u 330 331Where the events can either be a comma-separated list of events, or "-a" to 332enable all tracepoint events. Start and stop tracing as needed:: 333 334 lttng start 335 lttng stop 336 337View the trace:: 338 339 lttng view 340 341Destroy tracing session:: 342 343 lttng destroy 344 345Babeltrace can be used at any later time to view the trace:: 346 347 babeltrace $HOME/lttng-traces/mysession-<date>-<time> 348 349SystemTap 350--------- 351 352The "dtrace" backend uses DTrace sdt probes but has only been tested with 353SystemTap. When SystemTap support is detected a .stp file with wrapper probes 354is generated to make use in scripts more convenient. This step can also be 355performed manually after a build in order to change the binary name in the .stp 356probes:: 357 358 scripts/tracetool.py --backends=dtrace --format=stap \ 359 --binary path/to/qemu-binary \ 360 --probe-prefix qemu.system.x86_64 \ 361 --group=all \ 362 trace-events-all \ 363 qemu.stp 364 365To facilitate simple usage of systemtap where there merely needs to be printf 366logging of certain probes, a helper script "qemu-trace-stap" is provided. 367Consult its manual page for guidance on its usage. 368 369Trace event properties 370====================== 371 372Each event in the "trace-events-all" file can be prefixed with a space-separated 373list of zero or more of the following event properties. 374 375"disable" 376--------- 377 378If a specific trace event is going to be invoked a huge number of times, this 379might have a noticeable performance impact even when the event is 380programmatically disabled. 381 382In this case you should declare such event with the "disable" property. This 383will effectively disable the event at compile time (by using the "nop" backend), 384thus having no performance impact at all on regular builds (i.e., unless you 385edit the "trace-events-all" file). 386 387In addition, there might be cases where relatively complex computations must be 388performed to generate values that are only used as arguments for a trace 389function. In these cases you can use 'trace_event_get_state_backends()' to 390guard such computations, so they are skipped if the event has been either 391compile-time disabled or run-time disabled. If the event is compile-time 392disabled, this check will have no performance impact. 393 394:: 395 396 #include "trace.h" /* needed for trace event prototype */ 397 398 void *qemu_vmalloc(size_t size) 399 { 400 void *ptr; 401 size_t align = QEMU_VMALLOC_ALIGN; 402 403 if (size < align) { 404 align = getpagesize(); 405 } 406 ptr = qemu_memalign(align, size); 407 if (trace_event_get_state_backends(TRACE_QEMU_VMALLOC)) { 408 void *complex; 409 /* some complex computations to produce the 'complex' value */ 410 trace_qemu_vmalloc(size, ptr, complex); 411 } 412 return ptr; 413 } 414 415