6 perf-top - System profiling tool.
11 'perf top' [-e <EVENT> | --event=EVENT] [<options>]
15 This command generates and displays a performance counter profile in real time.
22 System-wide collection. (default)
26 Event period to sample.
30 Monitor only on the list of CPUs provided. Multiple CPUs can be provided as a
31 comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2.
32 Default is to monitor all CPUS.
36 Number of seconds to delay between refreshes.
40 Select the PMU event. Selection can be a symbolic event name
41 (use 'perf list' to list all events) or a raw PMU event in the form
42 of rN where N is a hexadecimal value that represents the raw register
43 encoding with the layout of the event control registers as described
44 by entries in /sys/bus/event_source/devices/cpu/format/*.
47 Event filter. This option should follow an event selector (-e). For
48 syntax see linkperf:perf-record[1].
52 Display this many functions.
55 --count-filter=<count>::
56 Only display functions with more events than this.
59 Sort the output by the event at the index n in group. If n is invalid,
60 sort by the first event. It can support multiple groups with different
61 amount of events. WARNING: This should be used on grouped events.
65 Profile at this frequency. Use 'max' to use the currently maximum
66 allowed frequency, i.e. the value in the kernel.perf_event_max_sample_rate
71 Child tasks do not inherit counters.
75 Path to vmlinux. Required for annotation functionality.
84 --mmap-pages=<pages>::
85 Number of mmap data pages (must be a power of two) or size
86 specification in bytes with appended unit character - B/K/M/G.
87 The size is rounded up to the nearest power-of-two page value.
91 Profile events on existing Process ID (comma separated list).
95 Profile events on existing thread ID (comma separated list).
99 Record events in threads owned by uid. Name or number.
102 --realtime=<priority>::
103 Collect data with this RT SCHED_FIFO priority.
105 --sym-annotate=<symbol>::
106 Annotate this symbol.
109 --hide_kernel_symbols::
113 --hide_user_symbols::
117 Demangle kernel symbols.
121 Dump the symbol table used for profiling.
125 Be more verbose (show counter open errors, etc).
129 Zero history across display updates.
133 Sort by key(s): pid, comm, dso, symbol, parent, srcline, weight,
134 local_weight, abort, in_tx, transaction, overhead, sample, period.
135 Please see description of --sort in the perf-report man page.
138 Specify output field - multiple keys can be specified in CSV format.
139 Following fields are available:
140 overhead, overhead_sys, overhead_us, overhead_children, sample and period.
141 Also it can contain any sort key(s).
143 By default, every sort keys not specified in --field will be appended
148 Show a column with the number of samples.
150 --show-total-period::
151 Show a column with the sum of periods.
154 Only consider symbols in these dsos. This option will affect the
155 percentage of the overhead column. See --percentage for more info.
158 Only consider symbols in these comms. This option will affect the
159 percentage of the overhead column. See --percentage for more info.
162 Only consider these symbols. This option will affect the
163 percentage of the overhead column. See --percentage for more info.
166 --disassembler-style=:: Set disassembler style for objdump.
169 Path to addr2line binary.
172 Path to objdump binary.
176 Remove first N entries from source file path names in executables
177 and add PREFIX. This allows to display source code compiled on systems
178 with different file system layout.
181 Interleave source code with assembly code. Enabled by default,
182 disable with --no-source.
185 Show raw instruction encoding of assembly instructions.
188 Enables call-graph (stack chain/backtrace) recording.
190 --call-graph [mode,type,min[,limit],order[,key][,branch]]::
191 Setup and enable call-graph (stack chain/backtrace) recording,
192 implies -g. See `--call-graph` section in perf-record and
193 perf-report man pages for details.
196 Accumulate callchain of children to parent entry so that then can
197 show up in the output. The output will have a new "Children" column
198 and will be sorted on the data. It requires -g/--call-graph option
199 enabled. See the `overhead calculation' section for more details.
200 Enabled by default, disable with --no-children.
203 Set the stack depth limit when parsing the callchain, anything
204 beyond the specified depth will be ignored. This is a trade-off
205 between information loss and faster processing especially for
206 workloads that can have a very long callchain stack.
208 Default: /proc/sys/kernel/perf_event_max_stack when present, 127 otherwise.
210 --ignore-callees=<regex>::
211 Ignore callees of the function(s) matching the given regex.
212 This has the effect of collecting the callers of each such
213 function into one place in the call-graph tree.
216 Do not show entries which have an overhead under that percent.
220 Determine how to display the overhead percentage of filtered entries.
221 Filters can be applied by --comms, --dsos and/or --symbols options and
222 Zoom operations on the TUI (thread, dso, etc).
224 "relative" means it's relative to filtered entries only so that the
225 sum of shown entries will be always 100%. "absolute" means it retains
226 the original value before and after the filter is applied.
229 --column-widths=<width[,width...]>::
230 Force each column width to the provided list, for large terminal
231 readability. 0 means no limit (default behavior).
234 When processing pre-existing threads /proc/XXX/mmap, it may take
235 a long time, because the file may be huge. A time out is needed
237 This option sets the time out limit. The default value is 500 ms.
242 Enable taken branch stack sampling. Any type of taken branch may be sampled.
243 This is a shortcut for --branch-filter any. See --branch-filter for more infos.
247 Enable taken branch stack sampling. Each sample captures a series of consecutive
248 taken branches. The number of branches captured with each sample depends on the
249 underlying hardware, the type of branches of interest, and the executed code.
250 It is possible to select the types of branches captured by enabling filters.
251 For a full list of modifiers please see the perf record manpage.
253 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
254 The privilege levels may be omitted, in which case, the privilege levels of the associated
255 event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
256 levels are subject to permissions. When sampling on multiple events, branch stack sampling
257 is enabled for all the sampling events. The sampled branch type is the same for all events.
258 The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
259 Note that this feature may not be available on all processors.
262 Add the addresses of sampled taken branches to the callstack.
263 This allows to examine the path the program took to each sample.
266 When displaying traceevent output, do not use print fmt or plugins.
270 Enable hierarchical output. In the hierarchy mode, each sort key groups
271 samples based on the criteria and then sub-divide it using the lower
274 For example, in normal output:
276 perf report -s dso,sym
278 # Overhead Shared Object Symbol
279 # ........ ................. ...........
280 50.00% [kernel.kallsyms] [k] kfunc1
282 15.00% [kernel.kallsyms] [k] kfunc2
284 5.00% libc.so [.] libcall
288 perf report -s dso,sym --hierarchy
290 # Overhead Shared Object / Symbol
291 # .......... ......................
292 65.00% [kernel.kallsyms]
302 Enable this to use just the most recent records, which helps in high core count
303 machines such as Knights Landing/Mill, but right now is disabled by default as
304 the pausing used in this technique is leading to loss of metadata events such
305 as PERF_RECORD_MMAP which makes 'perf top' unable to resolve samples, leading
306 to lots of unknown samples appearing on the UI. Enable this if you are in such
307 machines and profiling a workload that doesn't creates short lived threads and/or
308 doesn't uses many executable mmap operations. Work is being planed to solve
309 this situation, till then, this will remain disabled by default.
312 Don't do ownership validation.
314 --num-thread-synthesize::
315 The number of threads to run when synthesizing events for existing processes.
316 By default, the number of threads equals to the number of online CPUs.
319 Record events of type PERF_RECORD_NAMESPACES and display it with the
320 'cgroup_id' sort key.
324 monitor only in the container (cgroup) called "name". This option is available only
325 in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to
326 container "name" are monitored when they run on the monitored CPUs. Multiple cgroups
327 can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup
328 to first event, second cgroup to second event and so on. It is possible to provide
329 an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have
330 corresponding events, i.e., they always refer to events defined earlier on the command
331 line. If the user wants to track multiple events for a specific cgroup, the user can
332 use '-e e1 -e e2 -G foo,foo' or just use '-e e1 -e e2 -G foo'.
335 Record events of type PERF_RECORD_CGROUP and display it with the
338 --switch-on EVENT_NAME::
339 Only consider events after this event is found.
343 Find out where broadcast packets are handled
345 perf probe -L icmp_rcv
347 Insert a probe there:
349 perf probe icmp_rcv:59
351 Start perf top and ask it to only consider the cycles events when a
352 broadcast packet arrives This will show a menu with two entries and
353 will start counting when a broadcast packet arrives:
355 perf top -e cycles,probe:icmp_rcv --switch-on=probe:icmp_rcv
357 Alternatively one can ask for a group and then two overhead columns
358 will appear, the first for cycles and the second for the switch-on event.
360 perf top -e '{cycles,probe:icmp_rcv}' --switch-on=probe:icmp_rcv
362 This may be interesting to measure a workload only after some initialization
363 phase is over, i.e. insert a perf probe at that point and use the above
364 examples replacing probe:icmp_rcv with the just-after-init probe.
366 --switch-off EVENT_NAME::
367 Stop considering events after this event is found.
369 --show-on-off-events::
370 Show the --switch-on/off events too. This has no effect in 'perf top' now
371 but probably we'll make the default not to show the switch-on/off events
372 on the --group mode and if there is only one event besides the off/on ones,
373 go straight to the histogram browser, just like 'perf top' with no events
374 explicitly specified does.
377 Show callgraph with stitched LBRs, which may have more complete
378 callgraph. The option must be used with --call-graph lbr recording.
379 Disabled by default. In common cases with call stack overflows,
380 it can recreate better call stacks than the default lbr call stack
381 output. But this approach is not foolproof. There can be cases
382 where it creates incorrect call stacks from incorrect matches.
383 The known limitations include exception handing such as
384 setjmp/longjmp will have calls/returns not match.
387 --pfm-events events::
388 Select a PMU event using libpfm4 syntax (see http://perfmon2.sf.net)
389 including support for event filters. For example '--pfm-events
390 inst_retired:any_p:u:c=1:i'. More than one event can be passed to the
391 option using the comma separator. Hardware events and generic hardware
392 events cannot be mixed together. The latter must be used with the -e
393 option. The -e option and this one can be mixed and matched. Events
394 can be grouped using the {} notation.
397 INTERACTIVE PROMPTING KEYS
398 --------------------------
401 Display refresh delay.
404 Number of entries to display.
407 Event to display when multiple counters are active.
410 Profile display filter (>= hit count).
413 Annotation display filter (>= % of total).
419 Stop annotation, return to full profile display.
428 Toggle event count zeroing across display updates.
433 Pressing any unmapped key displays a menu, and prompts for input.
435 include::callchain-overhead-calculation.txt[]
439 linkperf:perf-stat[1], linkperf:perf-list[1], linkperf:perf-report[1]