1perf-top(1) 2=========== 3 4NAME 5---- 6perf-top - System profiling tool. 7 8SYNOPSIS 9-------- 10[verse] 11'perf top' [-e <EVENT> | --event=EVENT] [<options>] 12 13DESCRIPTION 14----------- 15This command generates and displays a performance counter profile in real time. 16 17 18OPTIONS 19------- 20-a:: 21--all-cpus:: 22 System-wide collection. (default) 23 24-c <count>:: 25--count=<count>:: 26 Event period to sample. 27 28-C <cpu-list>:: 29--cpu=<cpu>:: 30Monitor only on the list of CPUs provided. Multiple CPUs can be provided as a 31comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2. 32Default is to monitor all CPUS. 33 34-d <seconds>:: 35--delay=<seconds>:: 36 Number of seconds to delay between refreshes. 37 38-e <event>:: 39--event=<event>:: 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/*. 45 46-E <entries>:: 47--entries=<entries>:: 48 Display this many functions. 49 50-f <count>:: 51--count-filter=<count>:: 52 Only display functions with more events than this. 53 54--group:: 55 Put the counters into a counter group. 56 57--group-sort-idx:: 58 Sort the output by the event at the index n in group. If n is invalid, 59 sort by the first event. It can support multiple groups with different 60 amount of events. WARNING: This should be used on grouped events. 61 62-F <freq>:: 63--freq=<freq>:: 64 Profile at this frequency. Use 'max' to use the currently maximum 65 allowed frequency, i.e. the value in the kernel.perf_event_max_sample_rate 66 sysctl. 67 68-i:: 69--inherit:: 70 Child tasks do not inherit counters. 71 72-k <path>:: 73--vmlinux=<path>:: 74 Path to vmlinux. Required for annotation functionality. 75 76--ignore-vmlinux:: 77 Ignore vmlinux files. 78 79--kallsyms=<file>:: 80 kallsyms pathname 81 82-m <pages>:: 83--mmap-pages=<pages>:: 84 Number of mmap data pages (must be a power of two) or size 85 specification with appended unit character - B/K/M/G. The 86 size is rounded up to have nearest pages power of two value. 87 88-p <pid>:: 89--pid=<pid>:: 90 Profile events on existing Process ID (comma separated list). 91 92-t <tid>:: 93--tid=<tid>:: 94 Profile events on existing thread ID (comma separated list). 95 96-u:: 97--uid=:: 98 Record events in threads owned by uid. Name or number. 99 100-r <priority>:: 101--realtime=<priority>:: 102 Collect data with this RT SCHED_FIFO priority. 103 104--sym-annotate=<symbol>:: 105 Annotate this symbol. 106 107-K:: 108--hide_kernel_symbols:: 109 Hide kernel symbols. 110 111-U:: 112--hide_user_symbols:: 113 Hide user symbols. 114 115--demangle-kernel:: 116 Demangle kernel symbols. 117 118-D:: 119--dump-symtab:: 120 Dump the symbol table used for profiling. 121 122-v:: 123--verbose:: 124 Be more verbose (show counter open errors, etc). 125 126-z:: 127--zero:: 128 Zero history across display updates. 129 130-s:: 131--sort:: 132 Sort by key(s): pid, comm, dso, symbol, parent, srcline, weight, 133 local_weight, abort, in_tx, transaction, overhead, sample, period. 134 Please see description of --sort in the perf-report man page. 135 136--fields=:: 137 Specify output field - multiple keys can be specified in CSV format. 138 Following fields are available: 139 overhead, overhead_sys, overhead_us, overhead_children, sample and period. 140 Also it can contain any sort key(s). 141 142 By default, every sort keys not specified in --field will be appended 143 automatically. 144 145-n:: 146--show-nr-samples:: 147 Show a column with the number of samples. 148 149--show-total-period:: 150 Show a column with the sum of periods. 151 152--dsos:: 153 Only consider symbols in these dsos. This option will affect the 154 percentage of the overhead column. See --percentage for more info. 155 156--comms:: 157 Only consider symbols in these comms. This option will affect the 158 percentage of the overhead column. See --percentage for more info. 159 160--symbols:: 161 Only consider these symbols. This option will affect the 162 percentage of the overhead column. See --percentage for more info. 163 164-M:: 165--disassembler-style=:: Set disassembler style for objdump. 166 167--prefix=PREFIX:: 168--prefix-strip=N:: 169 Remove first N entries from source file path names in executables 170 and add PREFIX. This allows to display source code compiled on systems 171 with different file system layout. 172 173--source:: 174 Interleave source code with assembly code. Enabled by default, 175 disable with --no-source. 176 177--asm-raw:: 178 Show raw instruction encoding of assembly instructions. 179 180-g:: 181 Enables call-graph (stack chain/backtrace) recording. 182 183--call-graph [mode,type,min[,limit],order[,key][,branch]]:: 184 Setup and enable call-graph (stack chain/backtrace) recording, 185 implies -g. See `--call-graph` section in perf-record and 186 perf-report man pages for details. 187 188--children:: 189 Accumulate callchain of children to parent entry so that then can 190 show up in the output. The output will have a new "Children" column 191 and will be sorted on the data. It requires -g/--call-graph option 192 enabled. See the `overhead calculation' section for more details. 193 Enabled by default, disable with --no-children. 194 195--max-stack:: 196 Set the stack depth limit when parsing the callchain, anything 197 beyond the specified depth will be ignored. This is a trade-off 198 between information loss and faster processing especially for 199 workloads that can have a very long callchain stack. 200 201 Default: /proc/sys/kernel/perf_event_max_stack when present, 127 otherwise. 202 203--ignore-callees=<regex>:: 204 Ignore callees of the function(s) matching the given regex. 205 This has the effect of collecting the callers of each such 206 function into one place in the call-graph tree. 207 208--percent-limit:: 209 Do not show entries which have an overhead under that percent. 210 (Default: 0). 211 212--percentage:: 213 Determine how to display the overhead percentage of filtered entries. 214 Filters can be applied by --comms, --dsos and/or --symbols options and 215 Zoom operations on the TUI (thread, dso, etc). 216 217 "relative" means it's relative to filtered entries only so that the 218 sum of shown entries will be always 100%. "absolute" means it retains 219 the original value before and after the filter is applied. 220 221-w:: 222--column-widths=<width[,width...]>:: 223 Force each column width to the provided list, for large terminal 224 readability. 0 means no limit (default behavior). 225 226--proc-map-timeout:: 227 When processing pre-existing threads /proc/XXX/mmap, it may take 228 a long time, because the file may be huge. A time out is needed 229 in such cases. 230 This option sets the time out limit. The default value is 500 ms. 231 232 233-b:: 234--branch-any:: 235 Enable taken branch stack sampling. Any type of taken branch may be sampled. 236 This is a shortcut for --branch-filter any. See --branch-filter for more infos. 237 238-j:: 239--branch-filter:: 240 Enable taken branch stack sampling. Each sample captures a series of consecutive 241 taken branches. The number of branches captured with each sample depends on the 242 underlying hardware, the type of branches of interest, and the executed code. 243 It is possible to select the types of branches captured by enabling filters. 244 For a full list of modifiers please see the perf record manpage. 245 246 The option requires at least one branch type among any, any_call, any_ret, ind_call, cond. 247 The privilege levels may be omitted, in which case, the privilege levels of the associated 248 event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege 249 levels are subject to permissions. When sampling on multiple events, branch stack sampling 250 is enabled for all the sampling events. The sampled branch type is the same for all events. 251 The various filters must be specified as a comma separated list: --branch-filter any_ret,u,k 252 Note that this feature may not be available on all processors. 253 254--raw-trace:: 255 When displaying traceevent output, do not use print fmt or plugins. 256 257--hierarchy:: 258 Enable hierarchy output. 259 260--overwrite:: 261 Enable this to use just the most recent records, which helps in high core count 262 machines such as Knights Landing/Mill, but right now is disabled by default as 263 the pausing used in this technique is leading to loss of metadata events such 264 as PERF_RECORD_MMAP which makes 'perf top' unable to resolve samples, leading 265 to lots of unknown samples appearing on the UI. Enable this if you are in such 266 machines and profiling a workload that doesn't creates short lived threads and/or 267 doesn't uses many executable mmap operations. Work is being planed to solve 268 this situation, till then, this will remain disabled by default. 269 270--force:: 271 Don't do ownership validation. 272 273--num-thread-synthesize:: 274 The number of threads to run when synthesizing events for existing processes. 275 By default, the number of threads equals to the number of online CPUs. 276 277--namespaces:: 278 Record events of type PERF_RECORD_NAMESPACES and display it with the 279 'cgroup_id' sort key. 280 281-G name:: 282--cgroup name:: 283monitor only in the container (cgroup) called "name". This option is available only 284in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to 285container "name" are monitored when they run on the monitored CPUs. Multiple cgroups 286can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup 287to first event, second cgroup to second event and so on. It is possible to provide 288an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have 289corresponding events, i.e., they always refer to events defined earlier on the command 290line. If the user wants to track multiple events for a specific cgroup, the user can 291use '-e e1 -e e2 -G foo,foo' or just use '-e e1 -e e2 -G foo'. 292 293--all-cgroups:: 294 Record events of type PERF_RECORD_CGROUP and display it with the 295 'cgroup' sort key. 296 297--switch-on EVENT_NAME:: 298 Only consider events after this event is found. 299 300 E.g.: 301 302 Find out where broadcast packets are handled 303 304 perf probe -L icmp_rcv 305 306 Insert a probe there: 307 308 perf probe icmp_rcv:59 309 310 Start perf top and ask it to only consider the cycles events when a 311 broadcast packet arrives This will show a menu with two entries and 312 will start counting when a broadcast packet arrives: 313 314 perf top -e cycles,probe:icmp_rcv --switch-on=probe:icmp_rcv 315 316 Alternatively one can ask for --group and then two overhead columns 317 will appear, the first for cycles and the second for the switch-on event. 318 319 perf top --group -e cycles,probe:icmp_rcv --switch-on=probe:icmp_rcv 320 321 This may be interesting to measure a workload only after some initialization 322 phase is over, i.e. insert a perf probe at that point and use the above 323 examples replacing probe:icmp_rcv with the just-after-init probe. 324 325--switch-off EVENT_NAME:: 326 Stop considering events after this event is found. 327 328--show-on-off-events:: 329 Show the --switch-on/off events too. This has no effect in 'perf top' now 330 but probably we'll make the default not to show the switch-on/off events 331 on the --group mode and if there is only one event besides the off/on ones, 332 go straight to the histogram browser, just like 'perf top' with no events 333 explicitly specified does. 334 335--stitch-lbr:: 336 Show callgraph with stitched LBRs, which may have more complete 337 callgraph. The option must be used with --call-graph lbr recording. 338 Disabled by default. In common cases with call stack overflows, 339 it can recreate better call stacks than the default lbr call stack 340 output. But this approach is not full proof. There can be cases 341 where it creates incorrect call stacks from incorrect matches. 342 The known limitations include exception handing such as 343 setjmp/longjmp will have calls/returns not match. 344 345ifdef::HAVE_LIBPFM[] 346--pfm-events events:: 347Select a PMU event using libpfm4 syntax (see http://perfmon2.sf.net) 348including support for event filters. For example '--pfm-events 349inst_retired:any_p:u:c=1:i'. More than one event can be passed to the 350option using the comma separator. Hardware events and generic hardware 351events cannot be mixed together. The latter must be used with the -e 352option. The -e option and this one can be mixed and matched. Events 353can be grouped using the {} notation. 354endif::HAVE_LIBPFM[] 355 356INTERACTIVE PROMPTING KEYS 357-------------------------- 358 359[d]:: 360 Display refresh delay. 361 362[e]:: 363 Number of entries to display. 364 365[E]:: 366 Event to display when multiple counters are active. 367 368[f]:: 369 Profile display filter (>= hit count). 370 371[F]:: 372 Annotation display filter (>= % of total). 373 374[s]:: 375 Annotate symbol. 376 377[S]:: 378 Stop annotation, return to full profile display. 379 380[K]:: 381 Hide kernel symbols. 382 383[U]:: 384 Hide user symbols. 385 386[z]:: 387 Toggle event count zeroing across display updates. 388 389[qQ]:: 390 Quit. 391 392Pressing any unmapped key displays a menu, and prompts for input. 393 394include::callchain-overhead-calculation.txt[] 395 396SEE ALSO 397-------- 398linkperf:perf-stat[1], linkperf:perf-list[1], linkperf:perf-report[1] 399