xref: /linux/tools/perf/Documentation/perf-config.txt (revision 1dc707e647bc919834eff9636c8d00b78c782545)
1perf-config(1)
2==============
3
4NAME
5----
6perf-config - Get and set variables in a configuration file.
7
8SYNOPSIS
9--------
10[verse]
11'perf config' [<file-option>] [section.name[=value] ...]
12or
13'perf config' [<file-option>] -l | --list
14
15DESCRIPTION
16-----------
17You can manage variables in a configuration file with this command.
18
19OPTIONS
20-------
21
22-l::
23--list::
24	Show current config variables, name and value, for all sections.
25
26--user::
27	For writing and reading options: write to user
28	'$HOME/.perfconfig' file or read it.
29
30--system::
31	For writing and reading options: write to system-wide
32	'$(sysconfdir)/perfconfig' or read it.
33
34CONFIGURATION FILE
35------------------
36
37The perf configuration file contains many variables to change various
38aspects of each of its tools, including output, disk usage, etc.
39The '$HOME/.perfconfig' file is used to store a per-user configuration.
40The file '$(sysconfdir)/perfconfig' can be used to
41store a system-wide default configuration.
42
43One an disable reading config files by setting the PERF_CONFIG environment
44variable to /dev/null, or provide an alternate config file by setting that
45variable.
46
47When reading or writing, the values are read from the system and user
48configuration files by default, and options '--system' and '--user'
49can be used to tell the command to read from or write to only that location.
50
51Syntax
52~~~~~~
53
54The file consist of sections. A section starts with its name
55surrounded by square brackets and continues till the next section
56begins. Each variable must be in a section, and have the form
57'name = value', for example:
58
59	[section]
60		name1 = value1
61		name2 = value2
62
63Section names are case sensitive and can contain any characters except
64newline (double quote `"` and backslash have to be escaped as `\"` and `\\`,
65respectively). Section headers can't span multiple lines.
66
67Example
68~~~~~~~
69
70Given a $HOME/.perfconfig like this:
71
72#
73# This is the config file, and
74# a '#' and ';' character indicates a comment
75#
76
77	[colors]
78		# Color variables
79		top = red, default
80		medium = green, default
81		normal = lightgray, default
82		selected = white, lightgray
83		jump_arrows = blue, default
84		addr = magenta, default
85		root = white, blue
86
87	[tui]
88		# Defaults if linked with libslang
89		report = on
90		annotate = on
91		top = on
92
93	[buildid]
94		# Default, disable using /dev/null
95		dir = ~/.debug
96
97	[annotate]
98		# Defaults
99		hide_src_code = false
100		use_offset = true
101		jump_arrows = true
102		show_nr_jumps = false
103
104	[help]
105		# Format can be man, info, web or html
106		format = man
107		autocorrect = 0
108
109	[ui]
110		show-headers = true
111
112	[call-graph]
113		# fp (framepointer), dwarf
114		record-mode = fp
115		print-type = graph
116		order = caller
117		sort-key = function
118
119	[report]
120		# Defaults
121		sort_order = comm,dso,symbol
122		percent-limit = 0
123		queue-size = 0
124		children = true
125		group = true
126		skip-empty = true
127
128
129You can hide source code of annotate feature setting the config to false with
130
131	% perf config annotate.hide_src_code=true
132
133If you want to add or modify several config items, you can do like
134
135	% perf config ui.show-headers=false kmem.default=slab
136
137To modify the sort order of report functionality in user config file(i.e. `~/.perfconfig`), do
138
139	% perf config --user report.sort-order=srcline
140
141To change colors of selected line to other foreground and background colors
142in system config file (i.e. `$(sysconf)/perfconfig`), do
143
144	% perf config --system colors.selected=yellow,green
145
146To query the record mode of call graph, do
147
148	% perf config call-graph.record-mode
149
150If you want to know multiple config key/value pairs, you can do like
151
152	% perf config report.queue-size call-graph.order report.children
153
154To query the config value of sort order of call graph in user config file (i.e. `~/.perfconfig`), do
155
156	% perf config --user call-graph.sort-order
157
158To query the config value of buildid directory in system config file (i.e. `$(sysconf)/perfconfig`), do
159
160	% perf config --system buildid.dir
161
162Variables
163~~~~~~~~~
164
165colors.*::
166	The variables for customizing the colors used in the output for the
167	'report', 'top' and 'annotate' in the TUI. They should specify the
168	foreground and background colors, separated by a comma, for example:
169
170		medium = green, lightgray
171
172	If you want to use the color configured for you terminal, just leave it
173	as 'default', for example:
174
175		medium = default, lightgray
176
177	Available colors:
178	red, yellow, green, cyan, gray, black, blue,
179	white, default, magenta, lightgray
180
181	colors.top::
182		'top' means a overhead percentage which is more than 5%.
183		And values of this variable specify percentage colors.
184		Basic key values are foreground-color 'red' and
185		background-color 'default'.
186	colors.medium::
187		'medium' means a overhead percentage which has more than 0.5%.
188		Default values are 'green' and 'default'.
189	colors.normal::
190		'normal' means the rest of overhead percentages
191		except 'top', 'medium', 'selected'.
192		Default values are 'lightgray' and 'default'.
193	colors.selected::
194		This selects the colors for the current entry in a list of entries
195		from sub-commands (top, report, annotate).
196		Default values are 'black' and 'lightgray'.
197	colors.jump_arrows::
198		Colors for jump arrows on assembly code listings
199		such as 'jns', 'jmp', 'jane', etc.
200		Default values are 'blue', 'default'.
201	colors.addr::
202		This selects colors for addresses from 'annotate'.
203		Default values are 'magenta', 'default'.
204	colors.root::
205		Colors for headers in the output of a sub-commands (top, report).
206		Default values are 'white', 'blue'.
207
208core.*::
209	core.proc-map-timeout::
210		Sets a timeout (in milliseconds) for parsing /proc/<pid>/maps files.
211		Can be overridden by the --proc-map-timeout option on supported
212		subcommands. The default timeout is 500ms.
213
214tui.*, gtk.*::
215	Subcommands that can be configured here are 'top', 'report' and 'annotate'.
216	These values are booleans, for example:
217
218	[tui]
219		top = true
220
221	will make the TUI be the default for the 'top' subcommand. Those will be
222	available if the required libs were detected at tool build time.
223
224buildid.*::
225	buildid.dir::
226		Each executable and shared library in modern distributions comes with a
227		content based identifier that, if available, will be inserted in a
228		'perf.data' file header to, at analysis time find what is needed to do
229		symbol resolution, code annotation, etc.
230
231		The recording tools also stores a hard link or copy in a per-user
232		directory, $HOME/.debug/, of binaries, shared libraries, /proc/kallsyms
233		and /proc/kcore files to be used at analysis time.
234
235		The buildid.dir variable can be used to either change this directory
236		cache location, or to disable it altogether. If you want to disable it,
237		set buildid.dir to /dev/null. The default is $HOME/.debug
238
239buildid-cache.*::
240	buildid-cache.debuginfod=URLs
241		Specify debuginfod URLs to be used when retrieving perf.data binaries,
242		it follows the same syntax as the DEBUGINFOD_URLS variable, like:
243
244		  buildid-cache.debuginfod=http://192.168.122.174:8002
245
246annotate.*::
247	These are in control of addresses, jump function, source code
248	in lines of assembly code from a specific program.
249
250	annotate.disassemblers::
251		Choose the disassembler to use: "objdump", "llvm",  "capstone",
252		if not specified it will first try, if available, the "llvm" one,
253		then, if it fails, "capstone", and finally the original "objdump"
254		based one.
255
256		Choosing a different one is useful when handling some feature that
257		is known to be best support at some point by one of the options,
258		to compare the output when in doubt about some bug, etc.
259
260		This can be a list, in order of preference, the first one that works
261		finishes the process.
262
263	annotate.addr2line::
264		addr2line binary to use for file names and line numbers.
265
266	annotate.objdump::
267		objdump binary to use for disassembly and annotations,
268		including in the 'perf test' command.
269
270	annotate.disassembler_style::
271		Use this to change the default disassembler style to some other value
272		supported by binutils, such as "intel", see the '-M' option help in the
273		'objdump' man page.
274
275	annotate.hide_src_code::
276		If a program which is analyzed has source code,
277		this option lets 'annotate' print a list of assembly code with the source code.
278		For example, let's see a part of a program. There're four lines.
279		If this option is 'true', they can be printed
280		without source code from a program as below.
281
282		│        push   %rbp
283		│        mov    %rsp,%rbp
284		│        sub    $0x10,%rsp
285		│        mov    (%rdi),%rdx
286
287		But if this option is 'false', source code of the part
288		can be also printed as below. Default is 'false'.
289
290		│      struct rb_node *rb_next(const struct rb_node *node)
291		│      {
292		│        push   %rbp
293		│        mov    %rsp,%rbp
294		│        sub    $0x10,%rsp
295		│              struct rb_node *parent;
296297		│              if (RB_EMPTY_NODE(node))
298		│        mov    (%rdi),%rdx
299		│              return n;
300
301		This option works with tui, stdio2 browsers.
302
303        annotate.use_offset::
304		Basing on a first address of a loaded function, offset can be used.
305		Instead of using original addresses of assembly code,
306		addresses subtracted from a base address can be printed.
307		Let's illustrate an example.
308		If a base address is 0XFFFFFFFF81624d50 as below,
309
310		ffffffff81624d50 <load0>
311
312		an address on assembly code has a specific absolute address as below
313
314		ffffffff816250b8:│  mov    0x8(%r14),%rdi
315
316		but if use_offset is 'true', an address subtracted from a base address is printed.
317		Default is true. This option is only applied to TUI.
318
319		             368:│  mov    0x8(%r14),%rdi
320
321		This option works with tui, stdio2 browsers.
322
323	annotate.jump_arrows::
324		There can be jump instruction among assembly code.
325		Depending on a boolean value of jump_arrows,
326		arrows can be printed or not which represent
327		where do the instruction jump into as below.
328
329		│     ┌──jmp    1333
330		│     │  xchg   %ax,%ax
331		│1330:│  mov    %r15,%r10
332		│1333:└─→cmp    %r15,%r14
333
334		If jump_arrow is 'false', the arrows isn't printed as below.
335		Default is 'false'.
336
337		│      ↓ jmp    1333
338		│        xchg   %ax,%ax
339		│1330:   mov    %r15,%r10
340		│1333:   cmp    %r15,%r14
341
342		This option works with tui browser.
343
344        annotate.show_linenr::
345		When showing source code if this option is 'true',
346		line numbers are printed as below.
347
348		│1628         if (type & PERF_SAMPLE_IDENTIFIER) {
349		│     ↓ jne    508
350		│1628                 data->id = *array;
351		│1629                 array++;
352		│1630         }
353
354		However if this option is 'false', they aren't printed as below.
355		Default is 'false'.
356
357		│             if (type & PERF_SAMPLE_IDENTIFIER) {
358		│     ↓ jne    508
359		│                     data->id = *array;
360		│                     array++;
361		│             }
362
363		This option works with tui, stdio2 browsers.
364
365        annotate.show_nr_jumps::
366		Let's see a part of assembly code.
367
368		│1382:   movb   $0x1,-0x270(%rbp)
369
370		If use this, the number of branches jumping to that address can be printed as below.
371		Default is 'false'.
372
373		│1 1382:   movb   $0x1,-0x270(%rbp)
374
375		This option works with tui, stdio2 browsers.
376
377        annotate.show_total_period::
378		To compare two records on an instruction base, with this option
379		provided, display total number of samples that belong to a line
380		in assembly code. If this option is 'true', total periods are printed
381		instead of percent values as below.
382
383		  302 │      mov    %eax,%eax
384
385		But if this option is 'false', percent values for overhead are printed i.e.
386		Default is 'false'.
387
388		99.93 │      mov    %eax,%eax
389
390		This option works with tui, stdio2, stdio browsers.
391
392	annotate.show_nr_samples::
393		By default perf annotate shows percentage of samples. This option
394		can be used to print absolute number of samples. Ex, when set as
395		false:
396
397		Percent│
398		 74.03 │      mov    %fs:0x28,%rax
399
400		When set as true:
401
402		Samples│
403		     6 │      mov    %fs:0x28,%rax
404
405		This option works with tui, stdio2, stdio browsers.
406
407	annotate.offset_level::
408		Default is '1', meaning just jump targets will have offsets show right beside
409		the instruction. When set to '2' 'call' instructions will also have its offsets
410		shown, 3 or higher will show offsets for all instructions.
411
412		This option works with tui, stdio2 browsers.
413
414	annotate.demangle::
415		Demangle symbol names to human readable form. Default is 'true'.
416
417	annotate.demangle_kernel::
418		Demangle kernel symbol names to human readable form. Default is 'true'.
419
420hist.*::
421	hist.percentage::
422		This option control the way to calculate overhead of filtered entries -
423		that means the value of this option is effective only if there's a
424		filter (by comm, dso or symbol name). Suppose a following example:
425
426		       Overhead  Symbols
427		       ........  .......
428		        33.33%     foo
429		        33.33%     bar
430		        33.33%     baz
431
432	       This is an original overhead and we'll filter out the first 'foo'
433	       entry. The value of 'relative' would increase the overhead of 'bar'
434	       and 'baz' to 50.00% for each, while 'absolute' would show their
435	       current overhead (33.33%).
436
437ui.*::
438	ui.show-headers::
439		This option controls display of column headers (like 'Overhead' and 'Symbol')
440		in 'report' and 'top'. If this option is false, they are hidden.
441		This option is only applied to TUI.
442
443call-graph.*::
444	The following controls the handling of call-graphs (obtained via the
445	-g/--call-graph options).
446
447	call-graph.record-mode::
448		The mode for user space can be 'fp' (frame pointer), 'dwarf'
449		and 'lbr'.  The value 'dwarf' is effective only if libunwind
450		(or a recent version of libdw) is present on the system;
451		the value 'lbr' only works for certain cpus. The method for
452		kernel space is controlled not by this option but by the
453		kernel config (CONFIG_UNWINDER_*).
454
455	call-graph.dump-size::
456		The size of stack to dump in order to do post-unwinding. Default is 8192 (byte).
457		When using dwarf into record-mode, the default size will be used if omitted.
458
459	call-graph.print-type::
460		The print-types can be graph (graph absolute), fractal (graph relative),
461		flat and folded. This option controls a way to show overhead for each callchain
462		entry. Suppose a following example.
463
464                Overhead  Symbols
465                ........  .......
466                  40.00%  foo
467                          |
468                          ---foo
469                             |
470                             |--50.00%--bar
471                             |          main
472                             |
473                              --50.00%--baz
474                                        main
475
476		This output is a 'fractal' format. The 'foo' came from 'bar' and 'baz' exactly
477		half and half so 'fractal' shows 50.00% for each
478		(meaning that it assumes 100% total overhead of 'foo').
479
480		The 'graph' uses absolute overhead value of 'foo' as total so each of
481		'bar' and 'baz' callchain will have 20.00% of overhead.
482		If 'flat' is used, single column and linear exposure of call chains.
483		'folded' mean call chains are displayed in a line, separated by semicolons.
484
485	call-graph.order::
486		This option controls print order of callchains. The default is
487		'callee' which means callee is printed at top and then followed by its
488		caller and so on. The 'caller' prints it in reverse order.
489
490		If this option is not set and report.children or top.children is
491		set to true (or the equivalent command line option is given),
492		the default value of this option is changed to 'caller' for the
493		execution of 'perf report' or 'perf top'. Other commands will
494		still default to 'callee'.
495
496	call-graph.sort-key::
497		The callchains are merged if they contain same information.
498		The sort-key option determines a way to compare the callchains.
499		A value of 'sort-key' can be 'function' or 'address'.
500		The default is 'function'.
501
502	call-graph.threshold::
503		When there're many callchains it'd print tons of lines. So perf omits
504		small callchains under a certain overhead (threshold) and this option
505		control the threshold. Default is 0.5 (%). The overhead is calculated
506		by value depends on call-graph.print-type.
507
508	call-graph.print-limit::
509		This is a maximum number of lines of callchain printed for a single
510		histogram entry. Default is 0 which means no limitation.
511
512report.*::
513	report.sort_order::
514		Allows changing the default sort order from "comm,dso,symbol" to
515		some other default, for instance "sym,dso" may be more fitting for
516		kernel developers.
517	report.percent-limit::
518		This one is mostly the same as call-graph.threshold but works for
519		histogram entries. Entries having an overhead lower than this
520		percentage will not be printed. Default is '0'. If percent-limit
521		is '10', only entries which have more than 10% of overhead will be
522		printed.
523
524	report.queue-size::
525		This option sets up the maximum allocation size of the internal
526		event queue for ordering events. Default is 0, meaning no limit.
527
528	report.children::
529		'Children' means functions called from another function.
530		If this option is true, 'perf report' cumulates callchains of children
531		and show (accumulated) total overhead as well as 'Self' overhead.
532		Please refer to the 'perf report' manual. The default is 'true'.
533
534	report.group::
535		This option is to show event group information together.
536		Example output with this turned on, notice that there is one column
537		per event in the group, ref-cycles and cycles:
538
539		# group: {ref-cycles,cycles}
540		# ========
541		#
542		# Samples: 7K of event 'anon group { ref-cycles, cycles }'
543		# Event count (approx.): 6876107743
544		#
545		#         Overhead  Command      Shared Object               Symbol
546		# ................  .......  .................  ...................
547		#
548		    99.84%  99.76%  noploop  noploop            [.] main
549		     0.07%   0.00%  noploop  ld-2.15.so         [.] strcmp
550		     0.03%   0.00%  noploop  [kernel.kallsyms]  [k] timerqueue_del
551
552	report.skip-empty::
553		This option can change default stat behavior with empty results.
554		If it's set true, 'perf report --stat' will not show 0 stats.
555
556top.*::
557	top.children::
558		Same as 'report.children'. So if it is enabled, the output of 'top'
559		command will have 'Children' overhead column as well as 'Self' overhead
560		column by default.
561		The default is 'true'.
562
563	top.call-graph::
564		This is identical to 'call-graph.record-mode', except it is
565		applicable only for 'top' subcommand. This option ONLY setup
566		the unwind method. To enable 'perf top' to actually use it,
567		the command line option -g must be specified.
568
569man.*::
570	man.viewer::
571		This option can assign a tool to view manual pages when 'help'
572		subcommand was invoked. Supported tools are 'man', 'woman'
573		(with emacs client) and 'konqueror'. Default is 'man'.
574
575		New man viewer tool can be also added using 'man.<tool>.cmd'
576		or use different path using 'man.<tool>.path' config option.
577
578pager.*::
579	pager.<subcommand>::
580		When the subcommand is run on stdio, determine whether it uses
581		pager or not based on this value. Default is 'unspecified'.
582
583kmem.*::
584	kmem.default::
585		This option decides which allocator is to be analyzed if neither
586		'--slab' nor '--page' option is used. Default is 'slab'.
587
588record.*::
589	record.build-id::
590		This option can be 'cache', 'no-cache', 'skip' or 'mmap'.
591		'cache' is to post-process data and save/update the binaries into
592		the build-id cache (in ~/.debug). This is the default.
593		But if this option is 'no-cache', it will not update the build-id cache.
594		'skip' skips post-processing and does not update the cache.
595		'mmap' skips post-processing and reads build-ids from MMAP events.
596
597	record.call-graph::
598		This is identical to 'call-graph.record-mode', except it is
599		applicable only for 'record' subcommand. This option ONLY setup
600		the unwind method. To enable 'perf record' to actually use it,
601		the command line option -g must be specified.
602
603	record.aio::
604		Use 'n' control blocks in asynchronous (Posix AIO) trace writing
605		mode ('n' default: 1, max: 4).
606
607	record.debuginfod::
608		Specify debuginfod URL to be used when cacheing perf.data binaries,
609		it follows the same syntax as the DEBUGINFOD_URLS variable, like:
610
611		  http://192.168.122.174:8002
612
613		If the URLs is 'system', the value of DEBUGINFOD_URLS system environment
614		variable is used.
615
616diff.*::
617	diff.order::
618		This option sets the number of columns to sort the result.
619		The default is 0, which means sorting by baseline.
620		Setting it to 1 will sort the result by delta (or other
621		compute method selected).
622
623	diff.compute::
624		This options sets the method for computing the diff result.
625		Possible values are 'delta', 'delta-abs', 'ratio' and
626		'wdiff'.  Default is 'delta'.
627
628trace.*::
629	trace.add_events::
630		Allows adding a set of events to add to the ones specified
631		by the user, or use as a default one if none was specified.
632		The initial use case is to add augmented_raw_syscalls.o to
633		activate the 'perf trace' logic that looks for syscall
634		pointer contents after the normal tracepoint payload.
635
636	trace.args_alignment::
637		Number of columns to align the argument list, default is 70,
638		use 40 for the strace default, zero to no alignment.
639
640	trace.no_inherit::
641		Do not follow children threads.
642
643	trace.show_arg_names::
644		Should syscall argument names be printed? If not then trace.show_zeros
645		will be set.
646
647	trace.show_duration::
648		Show syscall duration.
649
650	trace.show_prefix::
651		If set to 'yes' will show common string prefixes in tables. The default
652		is to remove the common prefix in things like "MAP_SHARED", showing just "SHARED".
653
654	trace.show_timestamp::
655		Show syscall start timestamp.
656
657	trace.show_zeros::
658		Do not suppress syscall arguments that are equal to zero.
659
660	trace.tracepoint_beautifiers::
661		Use "libtraceevent" to use that library to augment the tracepoint arguments,
662		"libbeauty", the default, to use the same argument beautifiers used in the
663		strace-like sys_enter+sys_exit lines.
664
665ftrace.*::
666	ftrace.tracer::
667		Can be used to select the default tracer when neither -G nor
668		-F option is not specified. Possible values are 'function' and
669		'function_graph'.
670
671samples.*::
672
673	samples.context::
674		Define how many ns worth of time to show
675		around samples in perf report sample context browser.
676
677scripts.*::
678
679	Any option defines a script that is added to the scripts menu
680	in the interactive perf browser and whose output is displayed.
681	The name of the option is the name, the value is a script command line.
682	The script gets the same options passed as a full perf script,
683	in particular -i perfdata file, --cpu, --tid
684
685convert.*::
686
687	convert.queue-size::
688		Limit the size of ordered_events queue, so we could control
689		allocation size of perf data files without proper finished
690		round events.
691stat.*::
692
693	stat.big-num::
694		(boolean) Change the default for "--big-num". To make
695		"--no-big-num" the default, set "stat.big-num=false".
696
697intel-pt.*::
698
699	intel-pt.cache-divisor::
700
701	intel-pt.mispred-all::
702		If set, Intel PT decoder will set the mispred flag on all
703		branches.
704
705	intel-pt.max-loops::
706		If set and non-zero, the maximum number of unconditional
707		branches decoded without consuming any trace packets. If
708		the maximum is exceeded there will be a "Never-ending loop"
709		error. The default is 100000.
710
711auxtrace.*::
712
713	auxtrace.dumpdir::
714		s390 only. The directory to save the auxiliary trace buffer
715		can be changed using this option. Ex, auxtrace.dumpdir=/tmp.
716		If the directory does not exist or has the wrong file type,
717		the current directory is used.
718
719itrace.*::
720
721	debug-log-buffer-size::
722		Log size in bytes to output when using the option --itrace=d+e
723		Refer 'itrace' option of linkperf:perf-script[1] or
724		linkperf:perf-report[1]. The default is 16384.
725
726daemon.*::
727
728	daemon.base::
729		Base path for daemon data. All sessions data are stored under
730		this path.
731
732session-<NAME>.*::
733
734	session-<NAME>.run::
735
736		Defines new record session for daemon. The value is record's
737		command line without the 'record' keyword.
738
739SEE ALSO
740--------
741linkperf:perf[1]
742