1======= 2Tracing 3======= 4 5Introduction 6============ 7 8This document describes the tracing infrastructure in QEMU and how to use it 9for debugging, profiling, and observing execution. 10 11Quickstart 12========== 13 14Enable tracing of ``memory_region_ops_read`` and ``memory_region_ops_write`` 15events:: 16 17 $ qemu --trace "memory_region_ops_*" ... 18 ... 19 719585@1608130130.441188:memory_region_ops_read cpu 0 mr 0x562fdfbb3820 addr 0x3cc value 0x67 size 1 20 719585@1608130130.441190:memory_region_ops_write cpu 0 mr 0x562fdfbd2f00 addr 0x3d4 value 0x70e size 2 21 22This output comes from the "log" trace backend that is enabled by default when 23``./configure --enable-trace-backends=BACKENDS`` was not explicitly specified. 24 25Multiple patterns can be specified by repeating the ``--trace`` option:: 26 27 $ qemu --trace "kvm_*" --trace "virtio_*" ... 28 29When patterns are used frequently it is more convenient to store them in a 30file to avoid long command-line options:: 31 32 $ echo "memory_region_ops_*" >/tmp/events 33 $ echo "kvm_*" >>/tmp/events 34 $ qemu --trace events=/tmp/events ... 35 36Trace events 37============ 38 39Sub-directory setup 40------------------- 41 42Each directory in the source tree can declare a set of trace events in a local 43"trace-events" file. All directories which contain "trace-events" files must be 44listed in the "trace_events_subdirs" variable in the top level meson.build 45file. During build, the "trace-events" file in each listed subdirectory will be 46processed by the "tracetool" script to generate code for the trace events. 47 48The individual "trace-events" files are merged into a "trace-events-all" file, 49which is also installed into "/usr/share/qemu" with the name "trace-events". 50This merged file is to be used by the "simpletrace.py" script to later analyse 51traces in the simpletrace data format. 52 53The following files are automatically generated in <builddir>/trace/ during the 54build: 55 56 - trace-<subdir>.c - the trace event state declarations 57 - trace-<subdir>.h - the trace event enums and probe functions 58 - trace-dtrace-<subdir>.h - DTrace event probe specification 59 - trace-dtrace-<subdir>.dtrace - DTrace event probe helper declaration 60 - trace-dtrace-<subdir>.o - binary DTrace provider (generated by dtrace) 61 - trace-ust-<subdir>.h - UST event probe helper declarations 62 63Here <subdir> is the sub-directory path with '/' replaced by '_'. For example, 64"accel/kvm" becomes "accel_kvm" and the final filename for "trace-<subdir>.c" 65becomes "trace-accel_kvm.c". 66 67Source files in the source tree do not directly include generated files in 68"<builddir>/trace/". Instead they #include the local "trace.h" file, without 69any sub-directory path prefix. eg io/channel-buffer.c would do:: 70 71 #include "trace.h" 72 73The "io/trace.h" file must be created manually with an #include of the 74corresponding "trace/trace-<subdir>.h" file that will be generated in the 75builddir:: 76 77 $ echo '#include "trace/trace-io.h"' >io/trace.h 78 79While it is possible to include a trace.h file from outside a source file's own 80sub-directory, this is discouraged in general. It is strongly preferred that 81all events be declared directly in the sub-directory that uses them. The only 82exception is where there are some shared trace events defined in the top level 83directory trace-events file. The top level directory generates trace files 84with a filename prefix of "trace/trace-root" instead of just "trace". This is 85to avoid ambiguity between a trace.h in the current directory, vs the top level 86directory. 87 88Using trace events 89------------------ 90 91Trace events are invoked directly from source code like this:: 92 93 #include "trace.h" /* needed for trace event prototype */ 94 95 void *qemu_vmalloc(size_t size) 96 { 97 void *ptr; 98 size_t align = QEMU_VMALLOC_ALIGN; 99 100 if (size < align) { 101 align = getpagesize(); 102 } 103 ptr = qemu_memalign(align, size); 104 trace_qemu_vmalloc(size, ptr); 105 return ptr; 106 } 107 108Declaring trace events 109---------------------- 110 111The "tracetool" script produces the trace.h header file which is included by 112every source file that uses trace events. Since many source files include 113trace.h, it uses a minimum of types and other header files included to keep the 114namespace clean and compile times and dependencies down. 115 116Trace events should use types as follows: 117 118 * Use stdint.h types for fixed-size types. Most offsets and guest memory 119 addresses are best represented with uint32_t or uint64_t. Use fixed-size 120 types over primitive types whose size may change depending on the host 121 (32-bit versus 64-bit) so trace events don't truncate values or break 122 the build. 123 124 * Use void * for pointers to structs or for arrays. The trace.h header 125 cannot include all user-defined struct declarations and it is therefore 126 necessary to use void * for pointers to structs. 127 128 * For everything else, use primitive scalar types (char, int, long) with the 129 appropriate signedness. 130 131 * Avoid floating point types (float and double) because SystemTap does not 132 support them. In most cases it is possible to round to an integer type 133 instead. This may require scaling the value first by multiplying it by 1000 134 or the like when digits after the decimal point need to be preserved. 135 136Format strings should reflect the types defined in the trace event. Take 137special care to use PRId64 and PRIu64 for int64_t and uint64_t types, 138respectively. This ensures portability between 32- and 64-bit platforms. 139Format strings must not end with a newline character. It is the responsibility 140of backends to adapt line ending for proper logging. 141 142Each event declaration will start with the event name, then its arguments, 143finally a format string for pretty-printing. For example:: 144 145 qemu_vmalloc(size_t size, void *ptr) "size %zu ptr %p" 146 qemu_vfree(void *ptr) "ptr %p" 147 148 149Hints for adding new trace events 150--------------------------------- 151 1521. Trace state changes in the code. Interesting points in the code usually 153 involve a state change like starting, stopping, allocating, freeing. State 154 changes are good trace events because they can be used to understand the 155 execution of the system. 156 1572. Trace guest operations. Guest I/O accesses like reading device registers 158 are good trace events because they can be used to understand guest 159 interactions. 160 1613. Use correlator fields so the context of an individual line of trace output 162 can be understood. For example, trace the pointer returned by malloc and 163 used as an argument to free. This way mallocs and frees can be matched up. 164 Trace events with no context are not very useful. 165 1664. Name trace events after their function. If there are multiple trace events 167 in one function, append a unique distinguisher at the end of the name. 168 169Generic interface and monitor commands 170====================================== 171 172You can programmatically query and control the state of trace events through a 173backend-agnostic interface provided by the header "trace/control.h". 174 175Note that some of the backends do not provide an implementation for some parts 176of this interface, in which case QEMU will just print a warning (please refer to 177header "trace/control.h" to see which routines are backend-dependent). 178 179The state of events can also be queried and modified through monitor commands: 180 181* ``info trace-events`` 182 View available trace events and their state. State 1 means enabled, state 0 183 means disabled. 184 185* ``trace-event NAME on|off`` 186 Enable/disable a given trace event or a group of events (using wildcards). 187 188The "--trace events=<file>" command line argument can be used to enable the 189events listed in <file> from the very beginning of the program. This file must 190contain one event name per line. 191 192If a line in the "--trace events=<file>" file begins with a '-', the trace event 193will be disabled instead of enabled. This is useful when a wildcard was used 194to enable an entire family of events but one noisy event needs to be disabled. 195 196Wildcard matching is supported in both the monitor command "trace-event" and the 197events list file. That means you can enable/disable the events having a common 198prefix in a batch. For example, virtio-blk trace events could be enabled using 199the following monitor command:: 200 201 trace-event virtio_blk_* on 202 203Trace backends 204============== 205 206The "tracetool" script automates tedious trace event code generation and also 207keeps the trace event declarations independent of the trace backend. The trace 208events are not tightly coupled to a specific trace backend, such as LTTng or 209SystemTap. Support for trace backends can be added by extending the "tracetool" 210script. 211 212The trace backends are chosen at configure time:: 213 214 ./configure --enable-trace-backends=simple,dtrace 215 216For a list of supported trace backends, try ./configure --help or see below. 217If multiple backends are enabled, the trace is sent to them all. 218 219If no backends are explicitly selected, configure will default to the 220"log" backend. 221 222The following subsections describe the supported trace backends. 223 224Nop 225--- 226 227The "nop" backend generates empty trace event functions so that the compiler 228can optimize out trace events completely. This imposes no performance 229penalty. 230 231Note that regardless of the selected trace backend, events with the "disable" 232property will be generated with the "nop" backend. 233 234Log 235--- 236 237The "log" backend sends trace events directly to standard error. This 238effectively turns trace events into debug printfs. 239 240This is the simplest backend and can be used together with existing code that 241uses DPRINTF(). 242 243The -msg timestamp=on|off command-line option controls whether or not to print 244the tid/timestamp prefix for each trace event. 245 246Simpletrace 247----------- 248 249The "simple" backend writes binary trace logs to a file from a thread, making 250it lower overhead than the "log" backend. A Python API is available for writing 251offline trace file analysis scripts. It may not be as powerful as 252platform-specific or third-party trace backends but it is portable and has no 253special library dependencies. 254 255Monitor commands 256~~~~~~~~~~~~~~~~ 257 258* ``trace-file on|off|flush|set <path>`` 259 Enable/disable/flush the trace file or set the trace file name. 260 261Analyzing trace files 262~~~~~~~~~~~~~~~~~~~~~ 263 264The "simple" backend produces binary trace files that can be formatted with the 265simpletrace.py script. The script takes the "trace-events-all" file and the 266binary trace:: 267 268 ./scripts/simpletrace.py trace-events-all trace-12345 269 270You must ensure that the same "trace-events-all" file was used to build QEMU, 271otherwise trace event declarations may have changed and output will not be 272consistent. 273 274Ftrace 275------ 276 277The "ftrace" backend writes trace data to ftrace marker. This effectively 278sends trace events to ftrace ring buffer, and you can compare qemu trace 279data and kernel(especially kvm.ko when using KVM) trace data. 280 281if you use KVM, enable kvm events in ftrace:: 282 283 # echo 1 > /sys/kernel/debug/tracing/events/kvm/enable 284 285After running qemu by root user, you can get the trace:: 286 287 # cat /sys/kernel/debug/tracing/trace 288 289Restriction: "ftrace" backend is restricted to Linux only. 290 291Syslog 292------ 293 294The "syslog" backend sends trace events using the POSIX syslog API. The log 295is opened specifying the LOG_DAEMON facility and LOG_PID option (so events 296are tagged with the pid of the particular QEMU process that generated 297them). All events are logged at LOG_INFO level. 298 299NOTE: syslog may squash duplicate consecutive trace events and apply rate 300 limiting. 301 302Restriction: "syslog" backend is restricted to POSIX compliant OS. 303 304LTTng Userspace Tracer 305---------------------- 306 307The "ust" backend uses the LTTng Userspace Tracer library. There are no 308monitor commands built into QEMU, instead UST utilities should be used to list, 309enable/disable, and dump traces. 310 311Package lttng-tools is required for userspace tracing. You must ensure that the 312current user belongs to the "tracing" group, or manually launch the 313lttng-sessiond daemon for the current user prior to running any instance of 314QEMU. 315 316While running an instrumented QEMU, LTTng should be able to list all available 317events:: 318 319 lttng list -u 320 321Create tracing session:: 322 323 lttng create mysession 324 325Enable events:: 326 327 lttng enable-event qemu:g_malloc -u 328 329Where the events can either be a comma-separated list of events, or "-a" to 330enable all tracepoint events. Start and stop tracing as needed:: 331 332 lttng start 333 lttng stop 334 335View the trace:: 336 337 lttng view 338 339Destroy tracing session:: 340 341 lttng destroy 342 343Babeltrace can be used at any later time to view the trace:: 344 345 babeltrace $HOME/lttng-traces/mysession-<date>-<time> 346 347SystemTap 348--------- 349 350The "dtrace" backend uses DTrace sdt probes but has only been tested with 351SystemTap. When SystemTap support is detected a .stp file with wrapper probes 352is generated to make use in scripts more convenient. This step can also be 353performed manually after a build in order to change the binary name in the .stp 354probes:: 355 356 scripts/tracetool.py --backends=dtrace --format=stap \ 357 --binary path/to/qemu-binary \ 358 --target-type system \ 359 --target-name x86_64 \ 360 --group=all \ 361 trace-events-all \ 362 qemu.stp 363 364To facilitate simple usage of systemtap where there merely needs to be printf 365logging of certain probes, a helper script "qemu-trace-stap" is provided. 366Consult its manual page for guidance on its usage. 367 368Trace event properties 369====================== 370 371Each event in the "trace-events-all" file can be prefixed with a space-separated 372list of zero or more of the following event properties. 373 374"disable" 375--------- 376 377If a specific trace event is going to be invoked a huge number of times, this 378might have a noticeable performance impact even when the event is 379programmatically disabled. 380 381In this case you should declare such event with the "disable" property. This 382will effectively disable the event at compile time (by using the "nop" backend), 383thus having no performance impact at all on regular builds (i.e., unless you 384edit the "trace-events-all" file). 385 386In addition, there might be cases where relatively complex computations must be 387performed to generate values that are only used as arguments for a trace 388function. In these cases you can use 'trace_event_get_state_backends()' to 389guard such computations, so they are skipped if the event has been either 390compile-time disabled or run-time disabled. If the event is compile-time 391disabled, this check will have no performance impact. 392 393:: 394 395 #include "trace.h" /* needed for trace event prototype */ 396 397 void *qemu_vmalloc(size_t size) 398 { 399 void *ptr; 400 size_t align = QEMU_VMALLOC_ALIGN; 401 402 if (size < align) { 403 align = getpagesize(); 404 } 405 ptr = qemu_memalign(align, size); 406 if (trace_event_get_state_backends(TRACE_QEMU_VMALLOC)) { 407 void *complex; 408 /* some complex computations to produce the 'complex' value */ 409 trace_qemu_vmalloc(size, ptr, complex); 410 } 411 return ptr; 412 } 413 414"tcg" 415----- 416 417Guest code generated by TCG can be traced by defining an event with the "tcg" 418event property. Internally, this property generates two events: 419"<eventname>_trans" to trace the event at translation time, and 420"<eventname>_exec" to trace the event at execution time. 421 422Instead of using these two events, you should instead use the function 423"trace_<eventname>_tcg" during translation (TCG code generation). This function 424will automatically call "trace_<eventname>_trans", and will generate the 425necessary TCG code to call "trace_<eventname>_exec" during guest code execution. 426 427Events with the "tcg" property can be declared in the "trace-events" file with a 428mix of native and TCG types, and "trace_<eventname>_tcg" will gracefully forward 429them to the "<eventname>_trans" and "<eventname>_exec" events. Since TCG values 430are not known at translation time, these are ignored by the "<eventname>_trans" 431event. Because of this, the entry in the "trace-events" file needs two printing 432formats (separated by a comma):: 433 434 tcg foo(uint8_t a1, TCGv_i32 a2) "a1=%d", "a1=%d a2=%d" 435 436For example:: 437 438 #include "trace-tcg.h" 439 440 void some_disassembly_func (...) 441 { 442 uint8_t a1 = ...; 443 TCGv_i32 a2 = ...; 444 trace_foo_tcg(a1, a2); 445 } 446 447This will immediately call:: 448 449 void trace_foo_trans(uint8_t a1); 450 451and will generate the TCG code to call:: 452 453 void trace_foo(uint8_t a1, uint32_t a2); 454 455"vcpu" 456------ 457 458Identifies events that trace vCPU-specific information. It implicitly adds a 459"CPUState*" argument, and extends the tracing print format to show the vCPU 460information. If used together with the "tcg" property, it adds a second 461"TCGv_env" argument that must point to the per-target global TCG register that 462points to the vCPU when guest code is executed (usually the "cpu_env" variable). 463 464The "tcg" and "vcpu" properties are currently only honored in the root 465./trace-events file. 466 467The following example events:: 468 469 foo(uint32_t a) "a=%x" 470 vcpu bar(uint32_t a) "a=%x" 471 tcg vcpu baz(uint32_t a) "a=%x", "a=%x" 472 473Can be used as:: 474 475 #include "trace-tcg.h" 476 477 CPUArchState *env; 478 TCGv_ptr cpu_env; 479 480 void some_disassembly_func(...) 481 { 482 /* trace emitted at this point */ 483 trace_foo(0xd1); 484 /* trace emitted at this point */ 485 trace_bar(env_cpu(env), 0xd2); 486 /* trace emitted at this point (env) and when guest code is executed (cpu_env) */ 487 trace_baz_tcg(env_cpu(env), cpu_env, 0xd3); 488 } 489 490If the translating vCPU has address 0xc1 and code is later executed by vCPU 4910xc2, this would be an example output:: 492 493 // at guest code translation 494 foo a=0xd1 495 bar cpu=0xc1 a=0xd2 496 baz_trans cpu=0xc1 a=0xd3 497 // at guest code execution 498 baz_exec cpu=0xc2 a=0xd3 499