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
42 event (eventsel+umask) in the form of rNNN where NNN is a
43 hexadecimal event descriptor.
47 Display this many functions.
50 --count-filter=<count>::
51 Only display functions with more events than this.
54 Put the counters into a counter group.
58 Profile at this frequency. Use 'max' to use the currently maximum
59 allowed frequency, i.e. the value in the kernel.perf_event_max_sample_rate
64 Child tasks do not inherit counters.
68 Path to vmlinux. Required for annotation functionality.
77 --mmap-pages=<pages>::
78 Number of mmap data pages (must be a power of two) or size
79 specification with appended unit character - B/K/M/G. The
80 size is rounded up to have nearest pages power of two value.
84 Profile events on existing Process ID (comma separated list).
88 Profile events on existing thread ID (comma separated list).
92 Record events in threads owned by uid. Name or number.
95 --realtime=<priority>::
96 Collect data with this RT SCHED_FIFO priority.
98 --sym-annotate=<symbol>::
102 --hide_kernel_symbols::
106 --hide_user_symbols::
110 Demangle kernel symbols.
114 Dump the symbol table used for profiling.
118 Be more verbose (show counter open errors, etc).
122 Zero history across display updates.
126 Sort by key(s): pid, comm, dso, symbol, parent, srcline, weight,
127 local_weight, abort, in_tx, transaction, overhead, sample, period.
128 Please see description of --sort in the perf-report man page.
131 Specify output field - multiple keys can be specified in CSV format.
132 Following fields are available:
133 overhead, overhead_sys, overhead_us, overhead_children, sample and period.
134 Also it can contain any sort key(s).
136 By default, every sort keys not specified in --field will be appended
141 Show a column with the number of samples.
143 --show-total-period::
144 Show a column with the sum of periods.
147 Only consider symbols in these dsos. This option will affect the
148 percentage of the overhead column. See --percentage for more info.
151 Only consider symbols in these comms. This option will affect the
152 percentage of the overhead column. See --percentage for more info.
155 Only consider these symbols. This option will affect the
156 percentage of the overhead column. See --percentage for more info.
159 --disassembler-style=:: Set disassembler style for objdump.
163 Remove first N entries from source file path names in executables
164 and add PREFIX. This allows to display source code compiled on systems
165 with different file system layout.
168 Interleave source code with assembly code. Enabled by default,
169 disable with --no-source.
172 Show raw instruction encoding of assembly instructions.
175 Enables call-graph (stack chain/backtrace) recording.
177 --call-graph [mode,type,min[,limit],order[,key][,branch]]::
178 Setup and enable call-graph (stack chain/backtrace) recording,
179 implies -g. See `--call-graph` section in perf-record and
180 perf-report man pages for details.
183 Accumulate callchain of children to parent entry so that then can
184 show up in the output. The output will have a new "Children" column
185 and will be sorted on the data. It requires -g/--call-graph option
186 enabled. See the `overhead calculation' section for more details.
187 Enabled by default, disable with --no-children.
190 Set the stack depth limit when parsing the callchain, anything
191 beyond the specified depth will be ignored. This is a trade-off
192 between information loss and faster processing especially for
193 workloads that can have a very long callchain stack.
195 Default: /proc/sys/kernel/perf_event_max_stack when present, 127 otherwise.
197 --ignore-callees=<regex>::
198 Ignore callees of the function(s) matching the given regex.
199 This has the effect of collecting the callers of each such
200 function into one place in the call-graph tree.
203 Do not show entries which have an overhead under that percent.
207 Determine how to display the overhead percentage of filtered entries.
208 Filters can be applied by --comms, --dsos and/or --symbols options and
209 Zoom operations on the TUI (thread, dso, etc).
211 "relative" means it's relative to filtered entries only so that the
212 sum of shown entries will be always 100%. "absolute" means it retains
213 the original value before and after the filter is applied.
216 --column-widths=<width[,width...]>::
217 Force each column width to the provided list, for large terminal
218 readability. 0 means no limit (default behavior).
221 When processing pre-existing threads /proc/XXX/mmap, it may take
222 a long time, because the file may be huge. A time out is needed
224 This option sets the time out limit. The default value is 500 ms.
229 Enable taken branch stack sampling. Any type of taken branch may be sampled.
230 This is a shortcut for --branch-filter any. See --branch-filter for more infos.
234 Enable taken branch stack sampling. Each sample captures a series of consecutive
235 taken branches. The number of branches captured with each sample depends on the
236 underlying hardware, the type of branches of interest, and the executed code.
237 It is possible to select the types of branches captured by enabling filters.
238 For a full list of modifiers please see the perf record manpage.
240 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
241 The privilege levels may be omitted, in which case, the privilege levels of the associated
242 event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
243 levels are subject to permissions. When sampling on multiple events, branch stack sampling
244 is enabled for all the sampling events. The sampled branch type is the same for all events.
245 The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
246 Note that this feature may not be available on all processors.
249 When displaying traceevent output, do not use print fmt or plugins.
252 Enable hierarchy output.
255 Enable this to use just the most recent records, which helps in high core count
256 machines such as Knights Landing/Mill, but right now is disabled by default as
257 the pausing used in this technique is leading to loss of metadata events such
258 as PERF_RECORD_MMAP which makes 'perf top' unable to resolve samples, leading
259 to lots of unknown samples appearing on the UI. Enable this if you are in such
260 machines and profiling a workload that doesn't creates short lived threads and/or
261 doesn't uses many executable mmap operations. Work is being planed to solve
262 this situation, till then, this will remain disabled by default.
265 Don't do ownership validation.
267 --num-thread-synthesize::
268 The number of threads to run when synthesizing events for existing processes.
269 By default, the number of threads equals to the number of online CPUs.
272 Record events of type PERF_RECORD_NAMESPACES and display it with the
273 'cgroup_id' sort key.
275 --switch-on EVENT_NAME::
276 Only consider events after this event is found.
280 Find out where broadcast packets are handled
282 perf probe -L icmp_rcv
284 Insert a probe there:
286 perf probe icmp_rcv:59
288 Start perf top and ask it to only consider the cycles events when a
289 broadcast packet arrives This will show a menu with two entries and
290 will start counting when a broadcast packet arrives:
292 perf top -e cycles,probe:icmp_rcv --switch-on=probe:icmp_rcv
294 Alternatively one can ask for --group and then two overhead columns
295 will appear, the first for cycles and the second for the switch-on event.
297 perf top --group -e cycles,probe:icmp_rcv --switch-on=probe:icmp_rcv
299 This may be interesting to measure a workload only after some initialization
300 phase is over, i.e. insert a perf probe at that point and use the above
301 examples replacing probe:icmp_rcv with the just-after-init probe.
303 --switch-off EVENT_NAME::
304 Stop considering events after this event is found.
306 --show-on-off-events::
307 Show the --switch-on/off events too. This has no effect in 'perf top' now
308 but probably we'll make the default not to show the switch-on/off events
309 on the --group mode and if there is only one event besides the off/on ones,
310 go straight to the histogram browser, just like 'perf top' with no events
311 explicitely specified does.
314 INTERACTIVE PROMPTING KEYS
315 --------------------------
318 Display refresh delay.
321 Number of entries to display.
324 Event to display when multiple counters are active.
327 Profile display filter (>= hit count).
330 Annotation display filter (>= % of total).
336 Stop annotation, return to full profile display.
345 Toggle event count zeroing across display updates.
350 Pressing any unmapped key displays a menu, and prompts for input.
352 include::callchain-overhead-calculation.txt[]
356 linkperf:perf-stat[1], linkperf:perf-list[1], linkperf:perf-report[1]