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.
74 --mmap-pages=<pages>::
75 Number of mmap data pages (must be a power of two) or size
76 specification with appended unit character - B/K/M/G. The
77 size is rounded up to have nearest pages power of two value.
81 Profile events on existing Process ID (comma separated list).
85 Profile events on existing thread ID (comma separated list).
89 Record events in threads owned by uid. Name or number.
92 --realtime=<priority>::
93 Collect data with this RT SCHED_FIFO priority.
95 --sym-annotate=<symbol>::
99 --hide_kernel_symbols::
103 --hide_user_symbols::
107 Demangle kernel symbols.
111 Dump the symbol table used for profiling.
115 Be more verbose (show counter open errors, etc).
119 Zero history across display updates.
123 Sort by key(s): pid, comm, dso, symbol, parent, srcline, weight,
124 local_weight, abort, in_tx, transaction, overhead, sample, period.
125 Please see description of --sort in the perf-report man page.
128 Specify output field - multiple keys can be specified in CSV format.
129 Following fields are available:
130 overhead, overhead_sys, overhead_us, overhead_children, sample and period.
131 Also it can contain any sort key(s).
133 By default, every sort keys not specified in --field will be appended
138 Show a column with the number of samples.
140 --show-total-period::
141 Show a column with the sum of periods.
144 Only consider symbols in these dsos. This option will affect the
145 percentage of the overhead column. See --percentage for more info.
148 Only consider symbols in these comms. This option will affect the
149 percentage of the overhead column. See --percentage for more info.
152 Only consider these symbols. This option will affect the
153 percentage of the overhead column. See --percentage for more info.
156 --disassembler-style=:: Set disassembler style for objdump.
159 Interleave source code with assembly code. Enabled by default,
160 disable with --no-source.
163 Show raw instruction encoding of assembly instructions.
166 Enables call-graph (stack chain/backtrace) recording.
168 --call-graph [mode,type,min[,limit],order[,key][,branch]]::
169 Setup and enable call-graph (stack chain/backtrace) recording,
170 implies -g. See `--call-graph` section in perf-record and
171 perf-report man pages for details.
174 Accumulate callchain of children to parent entry so that then can
175 show up in the output. The output will have a new "Children" column
176 and will be sorted on the data. It requires -g/--call-graph option
177 enabled. See the `overhead calculation' section for more details.
178 Enabled by default, disable with --no-children.
181 Set the stack depth limit when parsing the callchain, anything
182 beyond the specified depth will be ignored. This is a trade-off
183 between information loss and faster processing especially for
184 workloads that can have a very long callchain stack.
186 Default: /proc/sys/kernel/perf_event_max_stack when present, 127 otherwise.
188 --ignore-callees=<regex>::
189 Ignore callees of the function(s) matching the given regex.
190 This has the effect of collecting the callers of each such
191 function into one place in the call-graph tree.
194 Do not show entries which have an overhead under that percent.
198 Determine how to display the overhead percentage of filtered entries.
199 Filters can be applied by --comms, --dsos and/or --symbols options and
200 Zoom operations on the TUI (thread, dso, etc).
202 "relative" means it's relative to filtered entries only so that the
203 sum of shown entries will be always 100%. "absolute" means it retains
204 the original value before and after the filter is applied.
207 --column-widths=<width[,width...]>::
208 Force each column width to the provided list, for large terminal
209 readability. 0 means no limit (default behavior).
212 When processing pre-existing threads /proc/XXX/mmap, it may take
213 a long time, because the file may be huge. A time out is needed
215 This option sets the time out limit. The default value is 500 ms.
220 Enable taken branch stack sampling. Any type of taken branch may be sampled.
221 This is a shortcut for --branch-filter any. See --branch-filter for more infos.
225 Enable taken branch stack sampling. Each sample captures a series of consecutive
226 taken branches. The number of branches captured with each sample depends on the
227 underlying hardware, the type of branches of interest, and the executed code.
228 It is possible to select the types of branches captured by enabling filters.
229 For a full list of modifiers please see the perf record manpage.
231 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond.
232 The privilege levels may be omitted, in which case, the privilege levels of the associated
233 event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege
234 levels are subject to permissions. When sampling on multiple events, branch stack sampling
235 is enabled for all the sampling events. The sampled branch type is the same for all events.
236 The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k
237 Note that this feature may not be available on all processors.
240 When displaying traceevent output, do not use print fmt or plugins.
243 Enable hierarchy output.
246 Enable this to use just the most recent records, which helps in high core count
247 machines such as Knights Landing/Mill, but right now is disabled by default as
248 the pausing used in this technique is leading to loss of metadata events such
249 as PERF_RECORD_MMAP which makes 'perf top' unable to resolve samples, leading
250 to lots of unknown samples appearing on the UI. Enable this if you are in such
251 machines and profiling a workload that doesn't creates short lived threads and/or
252 doesn't uses many executable mmap operations. Work is being planed to solve
253 this situation, till then, this will remain disabled by default.
256 Don't do ownership validation.
258 --num-thread-synthesize::
259 The number of threads to run when synthesizing events for existing processes.
260 By default, the number of threads equals to the number of online CPUs.
262 INTERACTIVE PROMPTING KEYS
263 --------------------------
266 Display refresh delay.
269 Number of entries to display.
272 Event to display when multiple counters are active.
275 Profile display filter (>= hit count).
278 Annotation display filter (>= % of total).
284 Stop annotation, return to full profile display.
293 Toggle event count zeroing across display updates.
298 Pressing any unmapped key displays a menu, and prompts for input.
300 include::callchain-overhead-calculation.txt[]
304 linkperf:perf-stat[1], linkperf:perf-list[1], linkperf:perf-report[1]