perf-record.txt 26 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677
  1. perf-record(1)
  2. ==============
  3. NAME
  4. ----
  5. perf-record - Run a command and record its profile into perf.data
  6. SYNOPSIS
  7. --------
  8. [verse]
  9. 'perf record' [-e <EVENT> | --event=EVENT] [-a] <command>
  10. 'perf record' [-e <EVENT> | --event=EVENT] [-a] -- <command> [<options>]
  11. DESCRIPTION
  12. -----------
  13. This command runs a command and gathers a performance counter profile
  14. from it, into perf.data - without displaying anything.
  15. This file can then be inspected later on, using 'perf report'.
  16. OPTIONS
  17. -------
  18. <command>...::
  19. Any command you can specify in a shell.
  20. -e::
  21. --event=::
  22. Select the PMU event. Selection can be:
  23. - a symbolic event name (use 'perf list' to list all events)
  24. - a raw PMU event (eventsel+umask) in the form of rNNN where NNN is a
  25. hexadecimal event descriptor.
  26. - a symbolic or raw PMU event followed by an optional colon
  27. and a list of event modifiers, e.g., cpu-cycles:p. See the
  28. linkperf:perf-list[1] man page for details on event modifiers.
  29. - a symbolically formed PMU event like 'pmu/param1=0x3,param2/' where
  30. 'param1', 'param2', etc are defined as formats for the PMU in
  31. /sys/bus/event_source/devices/<pmu>/format/*.
  32. - a symbolically formed event like 'pmu/config=M,config1=N,config3=K/'
  33. where M, N, K are numbers (in decimal, hex, octal format). Acceptable
  34. values for each of 'config', 'config1' and 'config2' are defined by
  35. corresponding entries in /sys/bus/event_source/devices/<pmu>/format/*
  36. param1 and param2 are defined as formats for the PMU in:
  37. /sys/bus/event_source/devices/<pmu>/format/*
  38. There are also some parameters which are not defined in .../<pmu>/format/*.
  39. These params can be used to overload default config values per event.
  40. Here are some common parameters:
  41. - 'period': Set event sampling period
  42. - 'freq': Set event sampling frequency
  43. - 'time': Disable/enable time stamping. Acceptable values are 1 for
  44. enabling time stamping. 0 for disabling time stamping.
  45. The default is 1.
  46. - 'call-graph': Disable/enable callgraph. Acceptable str are "fp" for
  47. FP mode, "dwarf" for DWARF mode, "lbr" for LBR mode and
  48. "no" for disable callgraph.
  49. - 'stack-size': user stack size for dwarf mode
  50. - 'name' : User defined event name. Single quotes (') may be used to
  51. escape symbols in the name from parsing by shell and tool
  52. like this: name=\'CPU_CLK_UNHALTED.THREAD:cmask=0x1\'.
  53. - 'aux-output': Generate AUX records instead of events. This requires
  54. that an AUX area event is also provided.
  55. - 'aux-sample-size': Set sample size for AUX area sampling. If the
  56. '--aux-sample' option has been used, set aux-sample-size=0 to disable
  57. AUX area sampling for the event.
  58. See the linkperf:perf-list[1] man page for more parameters.
  59. Note: If user explicitly sets options which conflict with the params,
  60. the value set by the parameters will be overridden.
  61. Also not defined in .../<pmu>/format/* are PMU driver specific
  62. configuration parameters. Any configuration parameter preceded by
  63. the letter '@' is not interpreted in user space and sent down directly
  64. to the PMU driver. For example:
  65. perf record -e some_event/@cfg1,@cfg2=config/ ...
  66. will see 'cfg1' and 'cfg2=config' pushed to the PMU driver associated
  67. with the event for further processing. There is no restriction on
  68. what the configuration parameters are, as long as their semantic is
  69. understood and supported by the PMU driver.
  70. - a hardware breakpoint event in the form of '\mem:addr[/len][:access]'
  71. where addr is the address in memory you want to break in.
  72. Access is the memory access type (read, write, execute) it can
  73. be passed as follows: '\mem:addr[:[r][w][x]]'. len is the range,
  74. number of bytes from specified addr, which the breakpoint will cover.
  75. If you want to profile read-write accesses in 0x1000, just set
  76. 'mem:0x1000:rw'.
  77. If you want to profile write accesses in [0x1000~1008), just set
  78. 'mem:0x1000/8:w'.
  79. - a BPF source file (ending in .c) or a precompiled object file (ending
  80. in .o) selects one or more BPF events.
  81. The BPF program can attach to various perf events based on the ELF section
  82. names.
  83. When processing a '.c' file, perf searches an installed LLVM to compile it
  84. into an object file first. Optional clang options can be passed via the
  85. '--clang-opt' command line option, e.g.:
  86. perf record --clang-opt "-DLINUX_VERSION_CODE=0x50000" \
  87. -e tests/bpf-script-example.c
  88. Note: '--clang-opt' must be placed before '--event/-e'.
  89. - a group of events surrounded by a pair of brace ("{event1,event2,...}").
  90. Each event is separated by commas and the group should be quoted to
  91. prevent the shell interpretation. You also need to use --group on
  92. "perf report" to view group events together.
  93. --filter=<filter>::
  94. Event filter. This option should follow an event selector (-e) which
  95. selects either tracepoint event(s) or a hardware trace PMU
  96. (e.g. Intel PT or CoreSight).
  97. - tracepoint filters
  98. In the case of tracepoints, multiple '--filter' options are combined
  99. using '&&'.
  100. - address filters
  101. A hardware trace PMU advertises its ability to accept a number of
  102. address filters by specifying a non-zero value in
  103. /sys/bus/event_source/devices/<pmu>/nr_addr_filters.
  104. Address filters have the format:
  105. filter|start|stop|tracestop <start> [/ <size>] [@<file name>]
  106. Where:
  107. - 'filter': defines a region that will be traced.
  108. - 'start': defines an address at which tracing will begin.
  109. - 'stop': defines an address at which tracing will stop.
  110. - 'tracestop': defines a region in which tracing will stop.
  111. <file name> is the name of the object file, <start> is the offset to the
  112. code to trace in that file, and <size> is the size of the region to
  113. trace. 'start' and 'stop' filters need not specify a <size>.
  114. If no object file is specified then the kernel is assumed, in which case
  115. the start address must be a current kernel memory address.
  116. <start> can also be specified by providing the name of a symbol. If the
  117. symbol name is not unique, it can be disambiguated by inserting #n where
  118. 'n' selects the n'th symbol in address order. Alternately #0, #g or #G
  119. select only a global symbol. <size> can also be specified by providing
  120. the name of a symbol, in which case the size is calculated to the end
  121. of that symbol. For 'filter' and 'tracestop' filters, if <size> is
  122. omitted and <start> is a symbol, then the size is calculated to the end
  123. of that symbol.
  124. If <size> is omitted and <start> is '*', then the start and size will
  125. be calculated from the first and last symbols, i.e. to trace the whole
  126. file.
  127. If symbol names (or '*') are provided, they must be surrounded by white
  128. space.
  129. The filter passed to the kernel is not necessarily the same as entered.
  130. To see the filter that is passed, use the -v option.
  131. The kernel may not be able to configure a trace region if it is not
  132. within a single mapping. MMAP events (or /proc/<pid>/maps) can be
  133. examined to determine if that is a possibility.
  134. Multiple filters can be separated with space or comma.
  135. --exclude-perf::
  136. Don't record events issued by perf itself. This option should follow
  137. an event selector (-e) which selects tracepoint event(s). It adds a
  138. filter expression 'common_pid != $PERFPID' to filters. If other
  139. '--filter' exists, the new filter expression will be combined with
  140. them by '&&'.
  141. -a::
  142. --all-cpus::
  143. System-wide collection from all CPUs (default if no target is specified).
  144. -p::
  145. --pid=::
  146. Record events on existing process ID (comma separated list).
  147. -t::
  148. --tid=::
  149. Record events on existing thread ID (comma separated list).
  150. This option also disables inheritance by default. Enable it by adding
  151. --inherit.
  152. -u::
  153. --uid=::
  154. Record events in threads owned by uid. Name or number.
  155. -r::
  156. --realtime=::
  157. Collect data with this RT SCHED_FIFO priority.
  158. --no-buffering::
  159. Collect data without buffering.
  160. -c::
  161. --count=::
  162. Event period to sample.
  163. -o::
  164. --output=::
  165. Output file name.
  166. -i::
  167. --no-inherit::
  168. Child tasks do not inherit counters.
  169. -F::
  170. --freq=::
  171. Profile at this frequency. Use 'max' to use the currently maximum
  172. allowed frequency, i.e. the value in the kernel.perf_event_max_sample_rate
  173. sysctl. Will throttle down to the currently maximum allowed frequency.
  174. See --strict-freq.
  175. --strict-freq::
  176. Fail if the specified frequency can't be used.
  177. -m::
  178. --mmap-pages=::
  179. Number of mmap data pages (must be a power of two) or size
  180. specification with appended unit character - B/K/M/G. The
  181. size is rounded up to have nearest pages power of two value.
  182. Also, by adding a comma, the number of mmap pages for AUX
  183. area tracing can be specified.
  184. --group::
  185. Put all events in a single event group. This precedes the --event
  186. option and remains only for backward compatibility. See --event.
  187. -g::
  188. Enables call-graph (stack chain/backtrace) recording for both
  189. kernel space and user space.
  190. --call-graph::
  191. Setup and enable call-graph (stack chain/backtrace) recording,
  192. implies -g. Default is "fp" (for user space).
  193. The unwinding method used for kernel space is dependent on the
  194. unwinder used by the active kernel configuration, i.e
  195. CONFIG_UNWINDER_FRAME_POINTER (fp) or CONFIG_UNWINDER_ORC (orc)
  196. Any option specified here controls the method used for user space.
  197. Valid options are "fp" (frame pointer), "dwarf" (DWARF's CFI -
  198. Call Frame Information) or "lbr" (Hardware Last Branch Record
  199. facility).
  200. In some systems, where binaries are build with gcc
  201. --fomit-frame-pointer, using the "fp" method will produce bogus
  202. call graphs, using "dwarf", if available (perf tools linked to
  203. the libunwind or libdw library) should be used instead.
  204. Using the "lbr" method doesn't require any compiler options. It
  205. will produce call graphs from the hardware LBR registers. The
  206. main limitation is that it is only available on new Intel
  207. platforms, such as Haswell. It can only get user call chain. It
  208. doesn't work with branch stack sampling at the same time.
  209. When "dwarf" recording is used, perf also records (user) stack dump
  210. when sampled. Default size of the stack dump is 8192 (bytes).
  211. User can change the size by passing the size after comma like
  212. "--call-graph dwarf,4096".
  213. -q::
  214. --quiet::
  215. Don't print any message, useful for scripting.
  216. -v::
  217. --verbose::
  218. Be more verbose (show counter open errors, etc).
  219. -s::
  220. --stat::
  221. Record per-thread event counts. Use it with 'perf report -T' to see
  222. the values.
  223. -d::
  224. --data::
  225. Record the sample virtual addresses.
  226. --phys-data::
  227. Record the sample physical addresses.
  228. -T::
  229. --timestamp::
  230. Record the sample timestamps. Use it with 'perf report -D' to see the
  231. timestamps, for instance.
  232. -P::
  233. --period::
  234. Record the sample period.
  235. --sample-cpu::
  236. Record the sample cpu.
  237. -n::
  238. --no-samples::
  239. Don't sample.
  240. -R::
  241. --raw-samples::
  242. Collect raw sample records from all opened counters (default for tracepoint counters).
  243. -C::
  244. --cpu::
  245. Collect samples only on the list of CPUs provided. Multiple CPUs can be provided as a
  246. comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
  247. In per-thread mode with inheritance mode on (default), samples are captured only when
  248. the thread executes on the designated CPUs. Default is to monitor all CPUs.
  249. -B::
  250. --no-buildid::
  251. Do not save the build ids of binaries in the perf.data files. This skips
  252. post processing after recording, which sometimes makes the final step in
  253. the recording process to take a long time, as it needs to process all
  254. events looking for mmap records. The downside is that it can misresolve
  255. symbols if the workload binaries used when recording get locally rebuilt
  256. or upgraded, because the only key available in this case is the
  257. pathname. You can also set the "record.build-id" config variable to
  258. 'skip to have this behaviour permanently.
  259. -N::
  260. --no-buildid-cache::
  261. Do not update the buildid cache. This saves some overhead in situations
  262. where the information in the perf.data file (which includes buildids)
  263. is sufficient. You can also set the "record.build-id" config variable to
  264. 'no-cache' to have the same effect.
  265. -G name,...::
  266. --cgroup name,...::
  267. monitor only in the container (cgroup) called "name". This option is available only
  268. in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
  269. container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
  270. can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
  271. to first event, second cgroup to second event and so on. It is possible to provide
  272. an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
  273. corresponding events, i.e., they always refer to events defined earlier on the command
  274. line. If the user wants to track multiple events for a specific cgroup, the user can
  275. use '-e e1 -e e2 -G foo,foo' or just use '-e e1 -e e2 -G foo'.
  276. If wanting to monitor, say, 'cycles' for a cgroup and also for system wide, this
  277. command line can be used: 'perf stat -e cycles -G cgroup_name -a -e cycles'.
  278. -b::
  279. --branch-any::
  280. Enable taken branch stack sampling. Any type of taken branch may be sampled.
  281. This is a shortcut for --branch-filter any. See --branch-filter for more infos.
  282. -j::
  283. --branch-filter::
  284. Enable taken branch stack sampling. Each sample captures a series of consecutive
  285. taken branches. The number of branches captured with each sample depends on the
  286. underlying hardware, the type of branches of interest, and the executed code.
  287. It is possible to select the types of branches captured by enabling filters. The
  288. following filters are defined:
  289. - any: any type of branches
  290. - any_call: any function call or system call
  291. - any_ret: any function return or system call return
  292. - ind_call: any indirect branch
  293. - call: direct calls, including far (to/from kernel) calls
  294. - u: only when the branch target is at the user level
  295. - k: only when the branch target is in the kernel
  296. - hv: only when the target is at the hypervisor level
  297. - in_tx: only when the target is in a hardware transaction
  298. - no_tx: only when the target is not in a hardware transaction
  299. - abort_tx: only when the target is a hardware transaction abort
  300. - cond: conditional branches
  301. - save_type: save branch type during sampling in case binary is not available later
  302. +
  303. The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
  304. The privilege levels may be omitted, in which case, the privilege levels of the associated
  305. event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
  306. levels are subject to permissions. When sampling on multiple events, branch stack sampling
  307. is enabled for all the sampling events. The sampled branch type is the same for all events.
  308. The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
  309. Note that this feature may not be available on all processors.
  310. --weight::
  311. Enable weightened sampling. An additional weight is recorded per sample and can be
  312. displayed with the weight and local_weight sort keys. This currently works for TSX
  313. abort events and some memory events in precise mode on modern Intel CPUs.
  314. --namespaces::
  315. Record events of type PERF_RECORD_NAMESPACES. This enables 'cgroup_id' sort key.
  316. --all-cgroups::
  317. Record events of type PERF_RECORD_CGROUP. This enables 'cgroup' sort key.
  318. --transaction::
  319. Record transaction flags for transaction related events.
  320. --per-thread::
  321. Use per-thread mmaps. By default per-cpu mmaps are created. This option
  322. overrides that and uses per-thread mmaps. A side-effect of that is that
  323. inheritance is automatically disabled. --per-thread is ignored with a warning
  324. if combined with -a or -C options.
  325. -D::
  326. --delay=::
  327. After starting the program, wait msecs before measuring (-1: start with events
  328. disabled). This is useful to filter out the startup phase of the program, which
  329. is often very different.
  330. -I::
  331. --intr-regs::
  332. Capture machine state (registers) at interrupt, i.e., on counter overflows for
  333. each sample. List of captured registers depends on the architecture. This option
  334. is off by default. It is possible to select the registers to sample using their
  335. symbolic names, e.g. on x86, ax, si. To list the available registers use
  336. --intr-regs=\?. To name registers, pass a comma separated list such as
  337. --intr-regs=ax,bx. The list of register is architecture dependent.
  338. --user-regs::
  339. Similar to -I, but capture user registers at sample time. To list the available
  340. user registers use --user-regs=\?.
  341. --running-time::
  342. Record running and enabled time for read events (:S)
  343. -k::
  344. --clockid::
  345. Sets the clock id to use for the various time fields in the perf_event_type
  346. records. See clock_gettime(). In particular CLOCK_MONOTONIC and
  347. CLOCK_MONOTONIC_RAW are supported, some events might also allow
  348. CLOCK_BOOTTIME, CLOCK_REALTIME and CLOCK_TAI.
  349. -S::
  350. --snapshot::
  351. Select AUX area tracing Snapshot Mode. This option is valid only with an
  352. AUX area tracing event. Optionally, certain snapshot capturing parameters
  353. can be specified in a string that follows this option:
  354. 'e': take one last snapshot on exit; guarantees that there is at least one
  355. snapshot in the output file;
  356. <size>: if the PMU supports this, specify the desired snapshot size.
  357. In Snapshot Mode trace data is captured only when signal SIGUSR2 is received
  358. and on exit if the above 'e' option is given.
  359. --aux-sample[=OPTIONS]::
  360. Select AUX area sampling. At least one of the events selected by the -e option
  361. must be an AUX area event. Samples on other events will be created containing
  362. data from the AUX area. Optionally sample size may be specified, otherwise it
  363. defaults to 4KiB.
  364. --proc-map-timeout::
  365. When processing pre-existing threads /proc/XXX/mmap, it may take a long time,
  366. because the file may be huge. A time out is needed in such cases.
  367. This option sets the time out limit. The default value is 500 ms.
  368. --switch-events::
  369. Record context switch events i.e. events of type PERF_RECORD_SWITCH or
  370. PERF_RECORD_SWITCH_CPU_WIDE. In some cases (e.g. Intel PT or CoreSight)
  371. switch events will be enabled automatically, which can be suppressed by
  372. by the option --no-switch-events.
  373. --clang-path=PATH::
  374. Path to clang binary to use for compiling BPF scriptlets.
  375. (enabled when BPF support is on)
  376. --clang-opt=OPTIONS::
  377. Options passed to clang when compiling BPF scriptlets.
  378. (enabled when BPF support is on)
  379. --vmlinux=PATH::
  380. Specify vmlinux path which has debuginfo.
  381. (enabled when BPF prologue is on)
  382. --buildid-all::
  383. Record build-id of all DSOs regardless whether it's actually hit or not.
  384. --aio[=n]::
  385. Use <n> control blocks in asynchronous (Posix AIO) trace writing mode (default: 1, max: 4).
  386. Asynchronous mode is supported only when linking Perf tool with libc library
  387. providing implementation for Posix AIO API.
  388. --affinity=mode::
  389. Set affinity mask of trace reading thread according to the policy defined by 'mode' value:
  390. node - thread affinity mask is set to NUMA node cpu mask of the processed mmap buffer
  391. cpu - thread affinity mask is set to cpu of the processed mmap buffer
  392. --mmap-flush=number::
  393. Specify minimal number of bytes that is extracted from mmap data pages and
  394. processed for output. One can specify the number using B/K/M/G suffixes.
  395. The maximal allowed value is a quarter of the size of mmaped data pages.
  396. The default option value is 1 byte which means that every time that the output
  397. writing thread finds some new data in the mmaped buffer the data is extracted,
  398. possibly compressed (-z) and written to the output, perf.data or pipe.
  399. Larger data chunks are compressed more effectively in comparison to smaller
  400. chunks so extraction of larger chunks from the mmap data pages is preferable
  401. from the perspective of output size reduction.
  402. Also at some cases executing less output write syscalls with bigger data size
  403. can take less time than executing more output write syscalls with smaller data
  404. size thus lowering runtime profiling overhead.
  405. -z::
  406. --compression-level[=n]::
  407. Produce compressed trace using specified level n (default: 1 - fastest compression,
  408. 22 - smallest trace)
  409. --all-kernel::
  410. Configure all used events to run in kernel space.
  411. --all-user::
  412. Configure all used events to run in user space.
  413. --kernel-callchains::
  414. Collect callchains only from kernel space. I.e. this option sets
  415. perf_event_attr.exclude_callchain_user to 1.
  416. --user-callchains::
  417. Collect callchains only from user space. I.e. this option sets
  418. perf_event_attr.exclude_callchain_kernel to 1.
  419. Don't use both --kernel-callchains and --user-callchains at the same time or no
  420. callchains will be collected.
  421. --timestamp-filename
  422. Append timestamp to output file name.
  423. --timestamp-boundary::
  424. Record timestamp boundary (time of first/last samples).
  425. --switch-output[=mode]::
  426. Generate multiple perf.data files, timestamp prefixed, switching to a new one
  427. based on 'mode' value:
  428. "signal" - when receiving a SIGUSR2 (default value) or
  429. <size> - when reaching the size threshold, size is expected to
  430. be a number with appended unit character - B/K/M/G
  431. <time> - when reaching the time threshold, size is expected to
  432. be a number with appended unit character - s/m/h/d
  433. Note: the precision of the size threshold hugely depends
  434. on your configuration - the number and size of your ring
  435. buffers (-m). It is generally more precise for higher sizes
  436. (like >5M), for lower values expect different sizes.
  437. A possible use case is to, given an external event, slice the perf.data file
  438. that gets then processed, possibly via a perf script, to decide if that
  439. particular perf.data snapshot should be kept or not.
  440. Implies --timestamp-filename, --no-buildid and --no-buildid-cache.
  441. The reason for the latter two is to reduce the data file switching
  442. overhead. You can still switch them on with:
  443. --switch-output --no-no-buildid --no-no-buildid-cache
  444. --switch-output-event::
  445. Events that will cause the switch of the perf.data file, auto-selecting
  446. --switch-output=signal, the results are similar as internally the side band
  447. thread will also send a SIGUSR2 to the main one.
  448. Uses the same syntax as --event, it will just not be recorded, serving only to
  449. switch the perf.data file as soon as the --switch-output event is processed by
  450. a separate sideband thread.
  451. This sideband thread is also used to other purposes, like processing the
  452. PERF_RECORD_BPF_EVENT records as they happen, asking the kernel for extra BPF
  453. information, etc.
  454. --switch-max-files=N::
  455. When rotating perf.data with --switch-output, only keep N files.
  456. --dry-run::
  457. Parse options then exit. --dry-run can be used to detect errors in cmdline
  458. options.
  459. 'perf record --dry-run -e' can act as a BPF script compiler if llvm.dump-obj
  460. in config file is set to true.
  461. --tail-synthesize::
  462. Instead of collecting non-sample events (for example, fork, comm, mmap) at
  463. the beginning of record, collect them during finalizing an output file.
  464. The collected non-sample events reflects the status of the system when
  465. record is finished.
  466. --overwrite::
  467. Makes all events use an overwritable ring buffer. An overwritable ring
  468. buffer works like a flight recorder: when it gets full, the kernel will
  469. overwrite the oldest records, that thus will never make it to the
  470. perf.data file.
  471. When '--overwrite' and '--switch-output' are used perf records and drops
  472. events until it receives a signal, meaning that something unusual was
  473. detected that warrants taking a snapshot of the most current events,
  474. those fitting in the ring buffer at that moment.
  475. 'overwrite' attribute can also be set or canceled for an event using
  476. config terms. For example: 'cycles/overwrite/' and 'instructions/no-overwrite/'.
  477. Implies --tail-synthesize.
  478. --kcore::
  479. Make a copy of /proc/kcore and place it into a directory with the perf data file.
  480. --max-size=<size>::
  481. Limit the sample data max size, <size> is expected to be a number with
  482. appended unit character - B/K/M/G
  483. --num-thread-synthesize::
  484. The number of threads to run when synthesizing events for existing processes.
  485. By default, the number of threads equals 1.
  486. ifdef::HAVE_LIBPFM[]
  487. --pfm-events events::
  488. Select a PMU event using libpfm4 syntax (see http://perfmon2.sf.net)
  489. including support for event filters. For example '--pfm-events
  490. inst_retired:any_p:u:c=1:i'. More than one event can be passed to the
  491. option using the comma separator. Hardware events and generic hardware
  492. events cannot be mixed together. The latter must be used with the -e
  493. option. The -e option and this one can be mixed and matched. Events
  494. can be grouped using the {} notation.
  495. endif::HAVE_LIBPFM[]
  496. --control=fifo:ctl-fifo[,ack-fifo]::
  497. --control=fd:ctl-fd[,ack-fd]::
  498. ctl-fifo / ack-fifo are opened and used as ctl-fd / ack-fd as follows.
  499. Listen on ctl-fd descriptor for command to control measurement ('enable': enable events,
  500. 'disable': disable events, 'snapshot': AUX area tracing snapshot). Measurements can be
  501. started with events disabled using --delay=-1 option. Optionally send control command
  502. completion ('ack\n') to ack-fd descriptor to synchronize with the controlling process.
  503. Example of bash shell script to enable and disable events during measurements:
  504. #!/bin/bash
  505. ctl_dir=/tmp/
  506. ctl_fifo=${ctl_dir}perf_ctl.fifo
  507. test -p ${ctl_fifo} && unlink ${ctl_fifo}
  508. mkfifo ${ctl_fifo}
  509. exec {ctl_fd}<>${ctl_fifo}
  510. ctl_ack_fifo=${ctl_dir}perf_ctl_ack.fifo
  511. test -p ${ctl_ack_fifo} && unlink ${ctl_ack_fifo}
  512. mkfifo ${ctl_ack_fifo}
  513. exec {ctl_fd_ack}<>${ctl_ack_fifo}
  514. perf record -D -1 -e cpu-cycles -a \
  515. --control fd:${ctl_fd},${ctl_fd_ack} \
  516. -- sleep 30 &
  517. perf_pid=$!
  518. sleep 5 && echo 'enable' >&${ctl_fd} && read -u ${ctl_fd_ack} e1 && echo "enabled(${e1})"
  519. sleep 10 && echo 'disable' >&${ctl_fd} && read -u ${ctl_fd_ack} d1 && echo "disabled(${d1})"
  520. exec {ctl_fd_ack}>&-
  521. unlink ${ctl_ack_fifo}
  522. exec {ctl_fd}>&-
  523. unlink ${ctl_fifo}
  524. wait -n ${perf_pid}
  525. exit $?
  526. SEE ALSO
  527. --------
  528. linkperf:perf-stat[1], linkperf:perf-list[1], linkperf:perf-intel-pt[1]