brintos

brintos / linux-shallow public Read only

0
0
Text · 13.7 KiB · af3e423 Raw
440 lines · plain
1perf-top(1)2===========3 4NAME5----6perf-top - System profiling tool.7 8SYNOPSIS9--------10[verse]11'perf top' [-e <EVENT> | --event=EVENT] [<options>]12 13DESCRIPTION14-----------15This command generates and displays a performance counter profile in real time.16 17 18OPTIONS19-------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 a31comma-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 name41	(use 'perf list' to list all events) or a raw PMU event in the form42	of rN where N is a hexadecimal value that represents the raw register43	encoding with the layout of the event control registers as described44	by entries in /sys/bus/event_source/devices/cpu/format/*.45 46--filter=<filter>::47	Event filter.  This option should follow an event selector (-e). For48	syntax see linkperf:perf-record[1].49 50-E <entries>::51--entries=<entries>::52	Display this many functions.53 54-f <count>::55--count-filter=<count>::56	Only display functions with more events than this.57 58--group-sort-idx::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 different61	amount of events. WARNING: This should be used on grouped events.62 63-F <freq>::64--freq=<freq>::65	Profile at this frequency. Use 'max' to use the currently maximum66	allowed frequency, i.e. the value in the kernel.perf_event_max_sample_rate67	sysctl.68 69-i::70--inherit::71	Child tasks do not inherit counters.72 73-k <path>::74--vmlinux=<path>::75	Path to vmlinux.  Required for annotation functionality.76 77--ignore-vmlinux::78	Ignore vmlinux files.79 80--kallsyms=<file>::81	kallsyms pathname82 83-m <pages>::84--mmap-pages=<pages>::85	Number of mmap data pages (must be a power of two) or size86	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.88 89-p <pid>::90--pid=<pid>::91	Profile events on existing Process ID (comma separated list).92 93-t <tid>::94--tid=<tid>::95        Profile events on existing thread ID (comma separated list).96 97-u::98--uid=::99        Record events in threads owned by uid. Name or number.100 101-r <priority>::102--realtime=<priority>::103	Collect data with this RT SCHED_FIFO priority.104 105--sym-annotate=<symbol>::106        Annotate this symbol.107 108-K::109--hide_kernel_symbols::110        Hide kernel symbols.111 112-U::113--hide_user_symbols::114        Hide user symbols.115 116--demangle-kernel::117        Demangle kernel symbols.118 119-D::120--dump-symtab::121        Dump the symbol table used for profiling.122 123-v::124--verbose::125	Be more verbose (show counter open errors, etc).126 127-z::128--zero::129	Zero history across display updates.130 131-s::132--sort::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.136 137--fields=::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).142 143	By default, every sort keys not specified in --field will be appended144	automatically.145 146-n::147--show-nr-samples::148	Show a column with the number of samples.149 150--show-total-period::151	Show a column with the sum of periods.152 153--dsos::154	Only consider symbols in these dsos.  This option will affect the155	percentage of the overhead column.  See --percentage for more info.156 157--comms::158	Only consider symbols in these comms.  This option will affect the159	percentage of the overhead column.  See --percentage for more info.160 161--symbols::162	Only consider these symbols.  This option will affect the163	percentage of the overhead column.  See --percentage for more info.164 165-M::166--disassembler-style=:: Set disassembler style for objdump.167 168--addr2line=<path>::169        Path to addr2line binary.170 171--objdump=<path>::172        Path to objdump binary.173 174--prefix=PREFIX::175--prefix-strip=N::176        Remove first N entries from source file path names in executables177        and add PREFIX. This allows to display source code compiled on systems178        with different file system layout.179 180--source::181	Interleave source code with assembly code. Enabled by default,182	disable with --no-source.183 184--asm-raw::185	Show raw instruction encoding of assembly instructions.186 187-g::188	Enables call-graph (stack chain/backtrace) recording.189 190--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 and193	perf-report man pages for details.194 195--children::196	Accumulate callchain of children to parent entry so that then can197	show up in the output.  The output will have a new "Children" column198	and will be sorted on the data.  It requires -g/--call-graph option199	enabled.  See the `overhead calculation' section for more details.200	Enabled by default, disable with --no-children.201 202--max-stack::203	Set the stack depth limit when parsing the callchain, anything204	beyond the specified depth will be ignored. This is a trade-off205	between information loss and faster processing especially for206	workloads that can have a very long callchain stack.207 208	Default: /proc/sys/kernel/perf_event_max_stack when present, 127 otherwise.209 210--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 such213        function into one place in the call-graph tree.214 215--percent-limit::216	Do not show entries which have an overhead under that percent.217	(Default: 0).218 219--percentage::220	Determine how to display the overhead percentage of filtered entries.221	Filters can be applied by --comms, --dsos and/or --symbols options and222	Zoom operations on the TUI (thread, dso, etc).223 224	"relative" means it's relative to filtered entries only so that the225	sum of shown entries will be always 100%. "absolute" means it retains226	the original value before and after the filter is applied.227 228-w::229--column-widths=<width[,width...]>::230	Force each column width to the provided list, for large terminal231	readability.  0 means no limit (default behavior).232 233--proc-map-timeout::234	When processing pre-existing threads /proc/XXX/mmap, it may take235	a long time, because the file may be huge. A time out is needed236	in such cases.237	This option sets the time out limit. The default value is 500 ms.238 239 240-b::241--branch-any::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.244 245-j::246--branch-filter::247	Enable taken branch stack sampling. Each sample captures a series of consecutive248	taken branches. The number of branches captured with each sample depends on the249	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.252 253	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 associated255	event are applied to the branch filter. Both kernel (k) and hypervisor (hv) privilege256	levels are subject to permissions.  When sampling on multiple events, branch stack sampling257	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,k259	Note that this feature may not be available on all processors.260 261--branch-history::262	Add the addresses of sampled taken branches to the callstack.263	This allows to examine the path the program took to each sample.264 265--raw-trace::266	When displaying traceevent output, do not use print fmt or plugins.267 268-H::269--hierarchy::270	Enable hierarchical output.  In the hierarchy mode, each sort key groups271	samples based on the criteria and then sub-divide it using the lower272	level sort key.273 274	For example, in normal output:275 276	  perf report -s dso,sym277	  #278	  # Overhead  Shared Object      Symbol279	  # ........  .................  ...........280	      50.00%  [kernel.kallsyms]  [k] kfunc1281	      20.00%  perf               [.] foo282	      15.00%  [kernel.kallsyms]  [k] kfunc2283	      10.00%  perf               [.] bar284	       5.00%  libc.so            [.] libcall285 286	In hierarchy output:287 288	  perf report -s dso,sym --hierarchy289	  #290	  #   Overhead  Shared Object / Symbol291	  # ..........  ......................292	      65.00%    [kernel.kallsyms]293	        50.00%    [k] kfunc1294	        15.00%    [k] kfunc2295	      30.00%    perf296	        20.00%    [.] foo297	        10.00%    [.] bar298	       5.00%    libc.so299	         5.00%    [.] libcall300 301--overwrite::302	Enable this to use just the most recent records, which helps in high core count303	machines such as Knights Landing/Mill, but right now is disabled by default as304	the pausing used in this technique is leading to loss of metadata events such305	as PERF_RECORD_MMAP which makes 'perf top' unable to resolve samples, leading306	to lots of unknown samples appearing on the UI. Enable this if you are in such307	machines and profiling a workload that doesn't creates short lived threads and/or308	doesn't uses many executable mmap operations. Work is being planed to solve309	this situation, till then, this will remain disabled by default.310 311--force::312	Don't do ownership validation.313 314--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.317 318--namespaces::319	Record events of type PERF_RECORD_NAMESPACES and display it with the320	'cgroup_id' sort key.321 322-G name::323--cgroup name::324monitor only in the container (cgroup) called "name". This option is available only325in per-cpu mode. The cgroup filesystem must be mounted. All threads belonging to326container "name" are monitored when they run on the monitored CPUs. Multiple cgroups327can be provided. Each cgroup is applied to the corresponding event, i.e., first cgroup328to first event, second cgroup to second event and so on. It is possible to provide329an empty cgroup (monitor all the time) using, e.g., -G foo,,bar. Cgroups must have330corresponding events, i.e., they always refer to events defined earlier on the command331line. If the user wants to track multiple events for a specific cgroup, the user can332use '-e e1 -e e2 -G foo,foo' or just use '-e e1 -e e2 -G foo'.333 334--all-cgroups::335	Record events of type PERF_RECORD_CGROUP and display it with the336	'cgroup' sort key.337 338--switch-on EVENT_NAME::339	Only consider events after this event is found.340 341	E.g.:342 343           Find out where broadcast packets are handled344 345		perf probe -L icmp_rcv346 347	   Insert a probe there:348 349		perf probe icmp_rcv:59350 351	   Start perf top and ask it to only consider the cycles events when a352           broadcast packet arrives This will show a menu with two entries and353           will start counting when a broadcast packet arrives:354 355		perf top -e cycles,probe:icmp_rcv --switch-on=probe:icmp_rcv356 357	   Alternatively one can ask for a group and then two overhead columns358           will appear, the first for cycles and the second for the switch-on event.359 360		perf top -e '{cycles,probe:icmp_rcv}' --switch-on=probe:icmp_rcv361 362	This may be interesting to measure a workload only after some initialization363	phase is over, i.e. insert a perf probe at that point and use the above364	examples replacing probe:icmp_rcv with the just-after-init probe.365 366--switch-off EVENT_NAME::367	Stop considering events after this event is found.368 369--show-on-off-events::370	Show the --switch-on/off events too. This has no effect in 'perf top' now371	but probably we'll make the default not to show the switch-on/off events372        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 events374	explicitly specified does.375 376--stitch-lbr::377	Show callgraph with stitched LBRs, which may have more complete378	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 stack381	output. But this approach is not foolproof. There can be cases382	where it creates incorrect call stacks from incorrect matches.383	The known limitations include exception handing such as384	setjmp/longjmp will have calls/returns not match.385 386ifdef::HAVE_LIBPFM[]387--pfm-events events::388Select a PMU event using libpfm4 syntax (see http://perfmon2.sf.net)389including support for event filters. For example '--pfm-events390inst_retired:any_p:u:c=1:i'. More than one event can be passed to the391option using the comma separator. Hardware events and generic hardware392events cannot be mixed together. The latter must be used with the -e393option. The -e option and this one can be mixed and matched.  Events394can be grouped using the {} notation.395endif::HAVE_LIBPFM[]396 397INTERACTIVE PROMPTING KEYS398--------------------------399 400[d]::401	Display refresh delay.402 403[e]::404	Number of entries to display.405 406[E]::407	Event to display when multiple counters are active.408 409[f]::410	Profile display filter (>= hit count).411 412[F]::413	Annotation display filter (>= % of total).414 415[s]::416	Annotate symbol.417 418[S]::419	Stop annotation, return to full profile display.420 421[K]::422	Hide kernel symbols.423 424[U]::425	Hide user symbols.426 427[z]::428	Toggle event count zeroing across display updates.429 430[qQ]::431	Quit.432 433Pressing any unmapped key displays a menu, and prompts for input.434 435include::callchain-overhead-calculation.txt[]436 437SEE ALSO438--------439linkperf:perf-stat[1], linkperf:perf-list[1], linkperf:perf-report[1]440