Lines Matching +full:trace +full:- +full:buffer +full:- +full:extension
1 perf-intel-pt(1)
5 ----
6 perf-intel-pt - Support for Intel Processor Trace within perf tools
9 --------
11 'perf record' -e intel_pt//
14 -----------
16 Intel Processor Trace (Intel PT) is an extension of Intel Architecture that
19 Technical details are documented in the Intel 64 and IA-32 Architectures
20 Software Developer Manuals, Chapter 36 Intel Processor Trace.
23 processors that are based on the Intel micro-architecture code name Broadwell.
25 Trace data is collected by 'perf record' and stored within the perf.data file.
28 Trace data must be 'decoded' which involves walking the object code and matching
29 the trace data packets. For example a TNT packet only tells whether a
33 Decoding is done on-the-fly. The decoder outputs samples in the same format as
43 builds, however the executed images are needed - which makes use in JIT-compiled
44 environments, or with self-modified code, a challenge. Also symbols need to be
47 A limitation of Intel PT is that it produces huge amounts of trace data
51 vary depending on the use-case and architecture.
55 ----------
61 Data is captured with 'perf record' e.g. to trace 'ls' userspace-only:
63 perf record -e intel_pt//u ls
69 To also trace kernel space presents a problem, namely kernel self-modifying
73 --kcore is used, but access to /proc/kcore is restricted e.g.
75 sudo perf record -o pt_ls --kcore -e intel_pt// -- ls
82 sudo perf report -i pt_ls
84 Because samples are synthesized after-the-fact, the sampling period can be
87 sudo perf report pt_ls --itrace=i1usge
89 See the sections below for more information about the --itrace option.
103 perf record -e intel_pt//u ls
104 perf script --itrace=iybxwpe
109 perf script --itrace=iybxwpe -F+flags
112 system, asynchronous, interrupt, transaction abort, trace begin, trace end,
113 in transaction, VM-entry, VM-exit, interrupt disabled, and interrupt disable
118 perf script --insn-trace=disasm
123 perf script --insn-trace --xed
125 Dumping all instructions in a long trace can be fairly slow. It is usually better
128 perf script --call-trace
132 perf script --call-ret-trace
137 perf script --time starttime,stoptime --insn-trace=disasm
139 While examining the trace it's also useful to filter on specific CPUs using
140 the -C option
142 perf script --time starttime,stoptime --insn-trace=disasm -C 1
149 perf script --itrace=be -F+ipc
151 There are two ways that instructions-per-cycle (IPC) can be calculated depending
156 MTC packets are used - refer to the 'mtc' config term. When MTC is used, however,
174 useful to use the 'A' option in conjunction with dlfilter-show-cycles.so to
183 Another note, in the case of "branches" events, non-taken branches are not
185 TNT packet that starts with a non-taken branch. To see every possible IPC
186 value, "instructions" events can be used e.g. --itrace=i0ns
190 Refer to script export-to-sqlite.py or export-to-postgresql.py for more details,
191 and to script exported-sql-viewer.py for an example of using the database.
193 There is also script intel-pt-events.py which provides an example of how to
197 - --insn-trace - instruction trace
198 - --src-trace - source trace
200 The intel-pt-events.py script also has options:
202 - --all-switch-events - display all switch events, not only the last consecutive.
203 - --interleave [<n>] - interleave sample output for the same timestamp so that
213 by inability to access the executed image, self-modified or JIT-ed code, or the
214 inability to match side-band information (such as context switches and mmaps)
218 resulting in data lost because the buffer was full. See 'Buffer handling' below
223 -----------
232 -e intel_pt//
236 -e intel_pt/tsc,noretcomp=0/
240 -e intel_pt/tsc=1,noretcomp=0/
242 Note there are other config terms - see section <<_config_terms,config terms>> further below.
249 $ grep -H . /sys/bus/event_source/devices/intel_pt/format/*
251 /sys/bus/event_source/devices/intel_pt/format/cyc_thresh:config:19-22
253 /sys/bus/event_source/devices/intel_pt/format/mtc_period:config:14-17
255 /sys/bus/event_source/devices/intel_pt/format/psb_period:config:24-27
260 -e intel_pt/noretcomp=0/
264 -e intel_pt/tsc=1,noretcomp=0/
268 -e intel_pt/tsc=0/
272 -e intel_pt/config=0x400/
287 perf_event_attr is displayed if the -vv option is used e.g.
289 ------------------------------------------------------------
303 ------------------------------------------------------------
304 sys_perf_event_open: pid 31104 cpu 0 group_fd -1 flags 0x8
305 sys_perf_event_open: pid 31104 cpu 1 group_fd -1 flags 0x8
306 sys_perf_event_open: pid 31104 cpu 2 group_fd -1 flags 0x8
307 sys_perf_event_open: pid 31104 cpu 3 group_fd -1 flags 0x8
308 ------------------------------------------------------------
314 Config terms are parameters specified with the -e intel_pt// event option,
317 -e intel_pt/cyc/
322 -e intel_pt/cyc=1/
327 -e intel_pt/branch=0/
331 -e intel_pt/cyc,mtc_period=9/
333 There are also common config terms, see linkperf:perf-record[1] documentation.
340 without timing information, for example a per-thread context
373 trace bytes between PSB packets as:
382 $ perf record -e intel_pt/psb_period=15/u uname
383 Invalid psb_period for intel_pt. Valid values are: 0-5
389 If decoding is expected to be reliable and the buffer is large
410 The frequency of MTC packets can also be specified - see
414 Specifies how frequently MTC packets are produced - see mtc
426 CTC-frequency / (2 ^ value)
428 e.g. value 3 means one eighth of CTC-frequency
436 $ perf record -e intel_pt/mtc_period=15/u uname
458 a threshold - see cyc_thresh below.
461 Specifies how frequently CYC packets are produced - see cyc
475 2 ^ (value - 1)
484 $ perf record -e intel_pt/cyc,cyc_thresh=15/u uname
485 Invalid cyc_thresh for intel_pt. Valid values are: 0-12
490 Specifies pass-through which enables the 'branch' config term.
523 changes to the CPU C-state.
533 Enable Event Trace. The events provide information about asynchronous
547 return compression is disabled - see noretcomp) return statements.
549 trace and corresponding tracing overhead.
558 *aux-action=start-paused*::
566 and PEBS-via-PT. In those cases, the other events can have config terms below:
568 *aux-sample-size*::
572 *aux-output*::
573 Used to select PEBS-via-PT, refer to the
576 *aux-action*::
585 --aux-sample
589 --aux-sample=8192
593 -e intel_pt//u
596 following will create Intel PT samples on the branch-misses event, note the
599 perf record --aux-sample -e '{intel_pt//u,branch-misses:u}'
601 An alternative to '--aux-sample' is to add the config term 'aux-sample-size' to
604 perf record -e intel_pt//u -e branch-misses/aux-sample-size=8192/u
608 perf record -e '{intel_pt//u,branch-misses/aux-sample-size=8192/u}'
612 perf record -e intel_pt//u --filter 'filter * @/bin/ls' -e branch-misses/aux-sample-size=8192/u -- ls
626 difficult to know how big the event might be without the trace sample attached,
633 The difference between full trace and snapshot from the kernel's perspective is
634 that in full trace we don't overwrite trace data that the user hasn't collected
636 the trace run and overwrite older data in the buffer so that whenever something
642 -S
646 -S0x100000
655 The snapshot size is displayed if the option -vv is used e.g.
663 Intel PT buffer size is specified by an addition to the -m option e.g.
665 -m,16
667 selects a buffer size of 16 pages i.e. 64KiB.
669 Note that the existing functionality of -m is unchanged. The auxtrace mmap size
683 In full-trace mode, powers of two are allowed for buffer size, with a minimum
687 The mmap size and auxtrace mmap size are displayed if the -vv option is used e.g.
697 full-trace mode
701 Full-trace mode traces continuously e.g.
703 perf record -e intel_pt//u uname
707 perf record --aux-sample -e intel_pt//u -e branch-misses:u
712 perf record -v -e intel_pt//u -S ./loopy 1000000000 &
714 kill -USR2 11435
718 Note that "Recording AUX area tracing snapshot" is displayed because the -v
728 $ sudo ~/bin/perf record --control fifo:perf.control,perf.ack -S -e intel_pt//u -- sleep 60 &
730 $ ps -e | grep perf
732 $ kill -USR2 15244
733 bash: kill: (15244) - Operation not permitted
740 Buffer handling
743 There may be buffer limitations (i.e. single ToPa entry) which means that actual
744 buffer sizes are limited to powers of 2 up to 4MiB (MAX_PAGE_ORDER). In order to
748 a) the interrupt may not be handled in time so that the current buffer
749 becomes full and some trace data is lost.
753 If trace data is lost, the driver sets 'truncated' in the PERF_RECORD_AUX event
756 In full-trace mode, the driver waits for data to be copied out before allowing
757 the (logical) buffer to wrap-around. If data is not copied out quickly enough,
760 that happens, perf tools always re-enable the intel_pt event after copying out
767 By default "perf record" post-processes the event stream to find all build ids
775 perf buildid-list
779 perf buildid-list --with-hits
787 collection of side-band information. In order to prevent that, a dummy
790 there is complete side-band information to allow the decoding of subsequent
813 "per thread" mode is selected by -t or by --per-thread (with -p or -u or just a
815 "per cpu" is selected by -C or -a.
819 In per-thread mode an exact list of threads is traced. There is no inheritance.
820 Each thread has its own event buffer.
822 In per-cpu mode all processes (or processes from the selected cgroup i.e. -G
823 option, or processes selected with -p or -u) are traced. Each cpu has its own
824 buffer. Inheritance is allowed.
826 In workload-only mode, the workload is traced but with per-cpu buffers.
827 Inheritance is allowed. Note that you can now trace a workload in per-thread
828 mode by using the --per-thread option.
831 Privileged vs non-privileged users
834 Unless /proc/sys/kernel/perf_event_paranoid is set to -1, unprivileged users
835 have memory limits imposed upon them. That affects what buffer sizes they can
850 Unless /proc/sys/kernel/perf_event_paranoid is set to -1, unprivileged users are
851 not permitted to use tracepoints which means there is insufficient side-band
852 information to decode Intel PT in per-cpu mode, and potentially workload-only
855 Note also, that to use tracepoints, read-access to debugfs is required. So if
856 debugfs is not mounted or the user does not have read-access, it will again not
857 be possible to decode Intel PT in per-cpu mode.
863 The sched_switch tracepoint is used to provide side-band data for Intel PT
870 $ perf record -vv -e intel_pt//u uname
871 ------------------------------------------------------------
885 ------------------------------------------------------------
886 sys_perf_event_open: pid 31104 cpu 0 group_fd -1 flags 0x8
887 sys_perf_event_open: pid 31104 cpu 1 group_fd -1 flags 0x8
888 sys_perf_event_open: pid 31104 cpu 2 group_fd -1 flags 0x8
889 sys_perf_event_open: pid 31104 cpu 3 group_fd -1 flags 0x8
890 ------------------------------------------------------------
901 ------------------------------------------------------------
902 sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8
903 sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8
904 sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0x8
905 sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0x8
906 ------------------------------------------------------------
925 ------------------------------------------------------------
926 sys_perf_event_open: pid 31104 cpu 0 group_fd -1 flags 0x8
927 sys_perf_event_open: pid 31104 cpu 1 group_fd -1 flags 0x8
928 sys_perf_event_open: pid 31104 cpu 2 group_fd -1 flags 0x8
929 sys_perf_event_open: pid 31104 cpu 3 group_fd -1 flags 0x8
932 perf event ring buffer mmapped per cpu
939 and only in per-cpu mode.
943 cannot be matched against the Intel PT trace.
947 -----------
949 By default, perf script will decode trace data found in the perf.data file.
950 This can be further controlled by new option --itrace.
953 New --itrace option
958 --itrace
962 --itrace=cepwxy
974 o synthesize PEBS-via-PT events
975 I synthesize Event Trace events
985 Z prefer to ignore timestamps (so-called "timeless" decoding)
987 "Instructions" events look like they were recorded by "perf record -e
990 "Cycles" events look like they were recorded by "perf record -e cycles"
1000 "Branches" events look like they were recorded by "perf record -e branches". "c"
1018 "Power" events correspond to power event packets and CBR (core-to-bus ratio)
1022 C-state changes, whereas CBR is indicative of CPU frequency. perf script
1027 pwre: hw: 0 cstate: 2 sub-cstate: 0
1033 "cbr" includes the frequency and the percentage of maximum non-turbo
1035 "pwre" shows C-state transitions (to a C-state deeper than C0) and
1041 For more details refer to the Intel 64 and IA-32 Architectures Software
1044 PSB events show when a PSB+ occurred and also the byte-offset in the trace.
1049 Error events show where the decoder lost the trace. Error events
1052 will or will not be reported. Each flag must be preceded by either '+' or '-'.
1055 -o Suppress overflow errors
1056 -l Suppress trace data lost errors
1060 --itrace=e-o-l
1066 must be preceded by either '+' or '-'. The flags support by Intel PT are:
1068 -a Suppress logging of perf events
1076 linkperf:perf-config[1] e.g. perf config itrace.debug-log-buffer-size=30000
1080 --itrace=i10us
1083 microseconds of trace. Alternatives to "us" are "ms" (milliseconds),
1098 'instructions' (i.e. --itrace=i1i).
1103 --itrace=ig32
1104 --itrace=xg32
1109 --itrace=il10
1110 --itrace=xl10
1117 instead of synthesized events. For example, to record branch-misses events for
1118 'ls' and then add a call chain derived from the Intel PT trace:
1120 perf record --aux-sample -e '{intel_pt//u,branch-misses:u}' -- ls
1121 perf report --itrace=Ge
1129 into the event buffer in one go. That reduces interrupts, but can give very
1130 late timestamps. Because the Intel PT trace is synchronized by timestamps,
1131 the PEBS events do not match the trace. Currently, Large PEBS is used only in
1133 - hardware supports it
1134 - PEBS is used
1135 - event period is specified, instead of frequency
1136 - the sample type is limited to the following flags:
1145 cases, avoid specifying the event period i.e. avoid the 'perf record' -c option,
1146 --count option, or 'period' config term.
1148 To disable trace decoding entirely, use the option --no-itrace.
1153 --itrace=i0nss1000000
1157 The q option changes the way the trace is decoded. The decoding is much faster
1164 ranges that could then be decoded fully using the --time option.
1168 - direct calls and jmps
1169 - conditional branches
1170 - non-branch instructions
1174 - asynchronous branches such as interrupts
1175 - indirect branches
1176 - function return target address *if* the noretcomp config term (refer
1178 - start of (control-flow) tracing
1179 - end of (control-flow) tracing, if it is not out of context
1180 - power events, ptwrite, transaction start and abort
1181 - instruction pointer associated with PSB packets
1186 Repeating the q option (double-q i.e. qq) results in even faster decoding and even
1189 PSBEND). Note PSB packets occur regularly in the trace based on the psb_period
1195 - everything except instruction pointer associated with PSB packets
1199 - instruction pointer associated with PSB packets
1201 The Z option is equivalent to having recorded a trace without TSC
1203 decoding a trace of a virtual machine.
1206 dlfilter-show-cycles.so
1209 Cycles can be displayed using dlfilter-show-cycles.so in which case the itrace A
1212 perf script --itrace=A --call-trace --dlfilter dlfilter-show-cycles.so
1216 perf script -v --list-dlfilters
1218 See also linkperf:perf-dlfilters[1]
1224 perf script has an option (-D) to "dump" the events i.e. display the binary
1227 When -D is used, Intel PT packets are displayed. The packet decoder does not
1228 pay attention to PSB packets, but just decodes the bytes - so the packets seen
1230 One example of that would be when the buffer-switching interrupt has been too
1231 slow, and the buffer has been filled completely. In that case, the last packet
1232 in the buffer might be truncated and immediately followed by a PSB as the trace
1233 continues in the next buffer.
1235 To disable the display of Intel PT packets, combine the -D option with
1236 --no-itrace.
1240 -----------
1242 By default, perf report will decode trace data found in the perf.data file.
1243 This can be further controlled by new option --itrace exactly the same as
1244 perf script, with the exception that the default is --itrace=igxe.
1248 -----------
1250 perf inject also accepts the --itrace option in which case tracing data is
1253 perf inject --itrace -i perf.data -o perf.data.new
1260 $ gcc-5 -O3 sort.c -o sort_optimized
1266 [intel-pt]
1267 mispred-all = on
1269 $ perf record -e intel_pt//u ./sort 3000
1274 $ perf inject -i perf.data -o inj --itrace=i100usle --strip
1275 $ ./create_gcov --binary=./sort --profile=inj --gcov=sort.gcov -gcov_version=1
1276 $ gcc-5 -O3 -fauto-profile=sort.gcov sort.c -o sort_autofdo
1286 -----------------
1288 Some hardware has the feature to redirect PEBS records to the Intel PT trace.
1289 Recording is selected by using the aux-output config term e.g.
1291 perf record -c 10000 -e '{intel_pt/branch=0/,cycles/aux-output/ppp}' uname
1295 kernels and perf tools add support for the PERF_RECORD_AUX_OUTPUT_HW_ID side-band event.
1296 To check for the presence of that event in a PEBS-via-PT trace:
1298 perf script -D --no-itrace | grep PERF_RECORD_AUX_OUTPUT_HW_ID
1300 To display PEBS events from the Intel PT trace, use the itrace 'o' option e.g.
1302 perf script --itrace=oe
1305 ---
1307 include::build-xed.txt[]
1311 --------------------------------------
1320 VMX controls may block the perf NMI to the host potentially resulting in lost trace data
1321 Guest kernel self-modifying code (e.g. jump labels or JIT-compiled eBPF) will result in decoding errors
1333 Mount the guest file system. Note sshfs needs -o direct_io to enable reading of proc files. root access is needed to read /proc/kcore.
1336 $ sshfs -o direct_io root@vm0:/ vm0
1340 $ perf buildid-cache -v --kcore vm0/proc/kcore
1341 kcore added to build-id cache directory /home/user/.debug/[kernel.kcore]/9600f316a53a0f54278885e8d9710538ec5f6a08/2021021807494306
1346 $ ps -eLl | grep 'KVM\|PID'
1348 3 S 64055 1430 1 1440 1 80 0 - 1921718 - ? 00:02:47 CPU 0/KVM
1349 3 S 64055 1430 1 1441 1 80 0 - 1921718 - ? 00:02:41 CPU 1/KVM
1350 3 S 64055 1430 1 1442 1 80 0 - 1921718 - ? 00:02:38 CPU 2/KVM
1351 3 S 64055 1430 1 1443 2 80 0 - 1921718 - ? 00:03:18 CPU 3/KVM
1353 Start an open-ended perf record, tracing the VM process, do something on the VM, and then ctrl-C to stop.
1357 Intel PT traces both the host and the guest so --guest and --host need to be specified.
1358 Without timestamps, --per-thread must be specified to distinguish threads.
1360 $ sudo perf kvm --guest --host --guestkallsyms $KALLSYMS record --kcore -e intel_pt/tsc=0,mtc=0,cyc=1/k -p 1430 --per-thread
1365 perf script can be used to provide an instruction trace
1367 $ perf script --guestkallsyms $KALLSYMS --insn-trace=disasm -F+ipc | grep -C10 vmresume | head -21
1397 Mount the guest file system. Note sshfs needs -o direct_io to enable reading of proc files. root access is needed to read /proc/kcore.
1399 $ mkdir -p vm0
1400 $ sshfs -o direct_io root@vm0:/ vm0
1404 $ perf buildid-cache -v --kcore vm0/proc/kcore
1410 $ ps -eLl | grep 'KVM\|PID'
1412 3 S 64055 16998 1 17005 13 80 0 - 1818189 - ? 00:00:16 CPU 0/KVM
1413 3 S 64055 16998 1 17006 4 80 0 - 1818189 - ? 00:00:05 CPU 1/KVM
1414 3 S 64055 16998 1 17007 3 80 0 - 1818189 - ? 00:00:04 CPU 2/KVM
1415 3 S 64055 16998 1 17008 4 80 0 - 1818189 - ? 00:00:05 CPU 3/KVM
1417 Start an open-ended perf record, tracing the VM process, do something on the VM, and then ctrl-C to stop.
1420 Intel PT traces both the host and the guest so --guest and --host need to be specified.
1422 $ sudo perf kvm --guest --host --guestkallsyms $KALLSYMS record --kcore -e intel_pt/cyc=1/k -p 16998
1427 only 7-bytes, so the TSC Offset might differ from the actual value in the 8th byte. That will
1430 $ perf inject -i perf.data.kvm --vm-time-correlation=dry-run
1446 $ perf inject -i perf.data.kvm --vm-time-correlation="dry-run 0xffffe42722c64c41"
1448 Note the options for 'perf inject' --vm-time-correlation are:
1450 [ dry-run ] [ <TSC Offset> [ : <VMCS> [ , <VMCS> ]... ] ]...
1453 The option "dry-run" will cause the file to be processed but without updating it.
1454 Note it is also possible to get a intel_pt.log file by adding option --itrace=d
1458 $ perf inject -i perf.data.kvm --vm-time-correlation=0xffffe42722c64c41 --force
1462 $ perf script -i perf.data.kvm --guestkallsyms $KALLSYMS --itrace=e-o
1466 'perf script' can be used to provide an instruction trace showing timestamps
1468 $ perf script -i perf.data.kvm --guestkallsyms $KALLSYMS --insn-trace=disasm -F+ipc | grep -C10 vmresume | head -21
1493 -----------------------------------------------
1495 It is possible to use perf record to record sideband events within a virtual machine, so that an Intel PT trace on the host can be decoded.
1502 Check that no-kvmclock kernel command line option was used to boot:
1507 BOOT_IMAGE=/boot/vmlinuz-5.10.0-16-amd64 root=UUID=cb49c910-e573-47e0-bce7-79e293df8e1d ro no-kvmclock
1516 $ sudo perf record -o guest-sideband-testing-guest-perf.data --sample-identifier --buildid-all --switch-events --kcore -a -e dummy
1520 Start perf record to collect Intel PT trace:
1522 Note, the host trace will get very big, very fast, so the steps from starting to stopping the host trace really need to be done so that they happen in the shortest time possible.
1524 $ sudo perf record -o guest-sideband-testing-host-perf.data -m,64M --kcore -a -e intel_pt/cyc/
1535 Stop the Intel PT trace:
1539 [ perf record: Captured and wrote 76.122 MB guest-sideband-testing-host-perf.data ]
1543 Stop the Intel PT trace:
1547 [ perf record: Captured and wrote 1.247 MB guest-sideband-testing-guest-perf.data ]
1549 And then copy guest-sideband-testing-guest-perf.data to the host (not shown here).
1557 $ perf inject -i guest-sideband-testing-host-perf.data --vm-time-correlation=dry-run
1565 $ perf inject -i guest-sideband-testing-host-perf.data --vm-time-correlation=0xfffffa6ae070cb20 --force
1569 $ perf script -i guest-sideband-testing-host-perf.data --no-itrace --show-task-events | grep KVM
1575 Note, the QEMU option -name debug-threads=on is needed so that thread names
1580 $ mkdir -p ~/guestmount/13376
1581 $ sshfs -o direct_io vm_to_test:/ ~/guestmount/13376
1586 If needed, VDSO can be copied manually in a fashion similar to that used by the perf-archive script.
1588 $ perf inject -i guest-sideband-testing-host-perf.data -o inj --guestmount ~/guestmount --guest-data=guest-sideband-testing-guest-perf.data,13376,0xfffffa6ae070cb20
1594 - the CPU displayed, [002] in this case, is always the host CPU
1595 - events happening in the virtual machine start with VM:13376 VCPU:003, which shows the hypervisor PID 13376 and the VCPU number
1596 - only calls and errors are displayed i.e. --itrace=ce
1597 - branches entering and exiting the virtual machine are split, and show as 2 branches to/from "0 [unknown] ([unknown])"
1599 $ perf script -i inj --itrace=ce -F+machine_pid,+vcpu,+addr,+pid,+tid,-period --ns --time 7919.408803365,7919.408804631 -C 2
1604 VM:13376 VCPU:003 uname 3404/3404 [002] 7919.408803461: branches: 0 [unknown] ([unknown]) => 7f851c9b5a5c init_cacheinfo+0x3ac (/usr/lib/x86_64-linux-gnu/libc-2.31.so)
1605 VM:13376 VCPU:003 uname 3404/3404 [002] 7919.408803567: branches: 7f851c9b5a5a init_cacheinfo+0x3aa (/usr/lib/x86_64-linux-gnu/libc-2.31.so) => 0 [unknown] ([unknown])
1655 VM:13376 VCPU:003 uname 3404/3404 [002] 7919.408804162: branches: 0 [unknown] ([unknown]) => 7f851c9b5a5c init_cacheinfo+0x3ac (/usr/lib/x86_64-linux-gnu/libc-2.31.so)
1656 VM:13376 VCPU:003 uname 3404/3404 [002] 7919.408804273: branches: 7f851cb7c0e4 _dl_init+0x74 (/usr/lib/x86_64-linux-gnu/ld-2.31.so) => 7f851cb7bf50 call_init.part.0+0x0 (/usr/lib/x86_64-linux-gnu/ld-2.31.so)
1665 Tracing Virtual Machines - Guest Code
1666 -------------------------------------
1673 addresses. To support that, option "--guest-code" has been added to perf script
1678 # perf record --kcore -e intel_pt/cyc/ -- tools/testing/selftests/kselftest_install/kvm/tsc_msrs_test
1681 # perf script --guest-code --itrace=bep --ns -F-period,+addr,+flags
1711 # perf kvm --guest-code --guest --host report -i perf.data --stdio | head -20
1713 # To display the perf.data header info, please use --header/--header-only options.
1726 ---entry_SYSCALL_64_after_hwframe
1729 |--29.44%--syscall_exit_to_user_mode
1735 Event Trace
1736 -----------
1738 Event Trace records information about asynchronous events, for example interrupts,
1749 7 VMENTRY VM-Entry
1750 8 VMEXIT VM-Entry
1751 9 VMEXIT_INTR VM-Exit due to interrupt
1754 For more details, refer to the Intel 64 and IA-32 Architectures Software
1757 The capability to do Event Trace is indicated by the
1760 Event trace is selected for recording using the "event" config term. e.g.
1762 perf record -e intel_pt/event/u uname
1764 Event trace events are output using the --itrace I option. e.g.
1766 perf script --itrace=Ie
1779 iflag: t IFLAG: 1->0 via branch
1789 t interrupts become disabled IF=1 -> IF=0
1791 Dt interrupts become enabled IF=0 -> IF=1
1793 The intel-pt-events.py script illustrates how to access Event Trace information
1798 -----------
1802 perf record -e intel_pt/notnt/u uname
1804 In that case the --itrace q option is forced because walking executable code
1809 ----------------
1815 a trace that encodes the payload data into TNT packets. Here is an example
1885 $ gcc -Wall -Wextra -O3 -g -o eg_ptw eg_ptw.c
1886 $ perf record -e intel_pt//u ./eg_ptw 0x1234567890abcdef
1889 $ perf script --itrace=ew
1895 ---------
1905 In the Intel PT case, the head and tail are updated only when the trace
1907 - full-trace, system wide : when buffer passes watermark
1908 - full-trace, not system-wide : when buffer passes watermark or
1910 - snapshot mode : as above but also when a snapshot is made
1911 - sample mode : as above but also when a sample is made
1913 That means finished-round ordering doesn't work. An auxtrace buffer
1917 For a perf.data file, that problem is solved by going through the trace
1925 -----------------------
1928 or resume Intel PT tracing. This is configured by using the "aux-action"
1931 "aux-action=pause" is used with events that are to pause Intel PT tracing.
1933 "aux-action=resume" is used with events that are to resume Intel PT tracing.
1935 "aux-action=start-paused" is used with the Intel PT event to start in a
1938 For example, to trace only the uname system call (sys_newuname) when running the
1941 $ perf record --kcore -e intel_pt/aux-action=start-paused/k,syscalls:sys_enter_newuname/aux-action=resume/,syscalls:sys_exit_newuname/aux-action=pause/ uname
1945 $ perf script --call-trace
1996 $ sudo perf record --kcore -a -e intel_pt/aux-action=start-paused/k -e mem:0xffffffffb605bf60:x/aux-action=resume/ -e mem:0xffffffffb614d8a0:x/aux-action=pause/ -- sleep 1
2002 $ sudo perf probe --add '__alloc_pages order'
2004 $ sudo perf probe --add __alloc_pages%return
2006 $ sudo perf record --kcore -aR -e intel_pt/aux-action=start-paused/k -e probe:__alloc_pages/aux-action=resume/ -e probe:__alloc_pages__return/aux-action=pause/ -- sleep 1
2012 $ sudo perf probe -x /usr/bin/uname main
2014 $ sudo perf record -e intel_pt/-aux-action=start-paused/u -e probe_uname:main/aux-action=resume/ -- uname
2021 $ perf record --kcore -a -m,64M -e intel_pt/aux-action=start-paused/k -e cycles/aux-action=pause,period=1000000/Pk -e cycles/aux-action=resume,period=10500000/Pk -- firefox
2027 -------
2029 Examples can be found on perf wiki page "Perf tools support for Intel® Processor Trace":
2035 --------
2037 linkperf:perf-record[1], linkperf:perf-script[1], linkperf:perf-report[1],
2038 linkperf:perf-inject[1]