1perf-report(1) 2============== 3 4NAME 5---- 6perf-report - Read perf.data (created by perf record) and display the profile 7 8SYNOPSIS 9-------- 10[verse] 11'perf report' [-i <file> | --input=file] 12 13DESCRIPTION 14----------- 15This command displays the performance counter profile information recorded 16via perf record. 17 18OPTIONS 19------- 20-i:: 21--input=:: 22 Input file name. (default: perf.data unless stdin is a fifo) 23 24-v:: 25--verbose:: 26 Be more verbose. (show symbol address, etc) 27 28-q:: 29--quiet:: 30 Do not show any warnings or messages. (Suppress -v) 31 32-n:: 33--show-nr-samples:: 34 Show the number of samples for each symbol 35 36--show-cpu-utilization:: 37 Show sample percentage for different cpu modes. 38 39-T:: 40--threads:: 41 Show per-thread event counters. The input data file should be recorded 42 with -s option. 43-c:: 44--comms=:: 45 Only consider symbols in these comms. CSV that understands 46 file://filename entries. This option will affect the percentage of 47 the overhead column. See --percentage for more info. 48--pid=:: 49 Only show events for given process ID (comma separated list). 50 51--tid=:: 52 Only show events for given thread ID (comma separated list). 53-d:: 54--dsos=:: 55 Only consider symbols in these dsos. CSV that understands 56 file://filename entries. This option will affect the percentage of 57 the overhead column. See --percentage for more info. 58-S:: 59--symbols=:: 60 Only consider these symbols. CSV that understands 61 file://filename entries. This option will affect the percentage of 62 the overhead column. See --percentage for more info. 63 64--symbol-filter=:: 65 Only show symbols that match (partially) with this filter. 66 67-U:: 68--hide-unresolved:: 69 Only display entries resolved to a symbol. 70 71-s:: 72--sort=:: 73 Sort histogram entries by given key(s) - multiple keys can be specified 74 in CSV format. Following sort keys are available: 75 pid, comm, dso, symbol, parent, cpu, socket, srcline, weight, 76 local_weight, cgroup_id, addr. 77 78 Each key has following meaning: 79 80 - comm: command (name) of the task which can be read via /proc/<pid>/comm 81 - pid: command and tid of the task 82 - dso: name of library or module executed at the time of sample 83 - dso_size: size of library or module executed at the time of sample 84 - symbol: name of function executed at the time of sample 85 - symbol_size: size of function executed at the time of sample 86 - parent: name of function matched to the parent regex filter. Unmatched 87 entries are displayed as "[other]". 88 - cpu: cpu number the task ran at the time of sample 89 - socket: processor socket number the task ran at the time of sample 90 - srcline: filename and line number executed at the time of sample. The 91 DWARF debugging info must be provided. 92 - srcfile: file name of the source file of the samples. Requires dwarf 93 information. 94 - weight: Event specific weight, e.g. memory latency or transaction 95 abort cost. This is the global weight. 96 - local_weight: Local weight version of the weight above. 97 - cgroup_id: ID derived from cgroup namespace device and inode numbers. 98 - cgroup: cgroup pathname in the cgroupfs. 99 - transaction: Transaction abort flags. 100 - overhead: Overhead percentage of sample 101 - overhead_sys: Overhead percentage of sample running in system mode 102 - overhead_us: Overhead percentage of sample running in user mode 103 - overhead_guest_sys: Overhead percentage of sample running in system mode 104 on guest machine 105 - overhead_guest_us: Overhead percentage of sample running in user mode on 106 guest machine 107 - sample: Number of sample 108 - period: Raw number of event count of sample 109 - time: Separate the samples by time stamp with the resolution specified by 110 --time-quantum (default 100ms). Specify with overhead and before it. 111 - code_page_size: the code page size of sampled code address (ip) 112 - ins_lat: Instruction latency in core cycles. This is the global instruction 113 latency 114 - local_ins_lat: Local instruction latency version 115 - p_stage_cyc: On powerpc, this presents the number of cycles spent in a 116 pipeline stage. And currently supported only on powerpc. 117 - addr: (Full) virtual address of the sampled instruction 118 - retire_lat: On X86, this reports pipeline stall of this instruction compared 119 to the previous instruction in cycles. And currently supported only on X86 120 - simd: Flags describing a SIMD operation. "e" for empty Arm SVE predicate. "p" for partial Arm SVE predicate 121 - type: Data type of sample memory access. 122 - typeoff: Offset in the data type of sample memory access. 123 - symoff: Offset in the symbol. 124 - weight1: Average value of event specific weight (1st field of weight_struct). 125 - weight2: Average value of event specific weight (2nd field of weight_struct). 126 - weight3: Average value of event specific weight (3rd field of weight_struct). 127 128 By default, comm, dso and symbol keys are used. 129 (i.e. --sort comm,dso,symbol) 130 131 If --branch-stack option is used, following sort keys are also 132 available: 133 134 - dso_from: name of library or module branched from 135 - dso_to: name of library or module branched to 136 - symbol_from: name of function branched from 137 - symbol_to: name of function branched to 138 - srcline_from: source file and line branched from 139 - srcline_to: source file and line branched to 140 - mispredict: "N" for predicted branch, "Y" for mispredicted branch 141 - in_tx: branch in TSX transaction 142 - abort: TSX transaction abort. 143 - cycles: Cycles in basic block 144 145 And default sort keys are changed to comm, dso_from, symbol_from, dso_to 146 and symbol_to, see '--branch-stack'. 147 148 When the sort key symbol is specified, columns "IPC" and "IPC Coverage" 149 are enabled automatically. Column "IPC" reports the average IPC per function 150 and column "IPC coverage" reports the percentage of instructions with 151 sampled IPC in this function. IPC means Instruction Per Cycle. If it's low, 152 it indicates there may be a performance bottleneck when the function is 153 executed, such as a memory access bottleneck. If a function has high overhead 154 and low IPC, it's worth further analyzing it to optimize its performance. 155 156 If the --mem-mode option is used, the following sort keys are also available 157 (incompatible with --branch-stack): 158 symbol_daddr, dso_daddr, locked, tlb, mem, snoop, dcacheline, blocked. 159 160 - symbol_daddr: name of data symbol being executed on at the time of sample 161 - dso_daddr: name of library or module containing the data being executed 162 on at the time of the sample 163 - locked: whether the bus was locked at the time of the sample 164 - tlb: type of tlb access for the data at the time of the sample 165 - mem: type of memory access for the data at the time of the sample 166 - snoop: type of snoop (if any) for the data at the time of the sample 167 - dcacheline: the cacheline the data address is on at the time of the sample 168 - phys_daddr: physical address of data being executed on at the time of sample 169 - data_page_size: the data page size of data being executed on at the time of sample 170 - blocked: reason of blocked load access for the data at the time of the sample 171 172 And the default sort keys are changed to local_weight, mem, sym, dso, 173 symbol_daddr, dso_daddr, snoop, tlb, locked, blocked, local_ins_lat, 174 see '--mem-mode'. 175 176 If the data file has tracepoint event(s), following (dynamic) sort keys 177 are also available: 178 trace, trace_fields, [<event>.]<field>[/raw] 179 180 - trace: pretty printed trace output in a single column 181 - trace_fields: fields in tracepoints in separate columns 182 - <field name>: optional event and field name for a specific field 183 184 The last form consists of event and field names. If event name is 185 omitted, it searches all events for matching field name. The matched 186 field will be shown only for the event has the field. The event name 187 supports substring match so user doesn't need to specify full subsystem 188 and event name everytime. For example, 'sched:sched_switch' event can 189 be shortened to 'switch' as long as it's not ambiguous. Also event can 190 be specified by its index (starting from 1) preceded by the '%'. 191 So '%1' is the first event, '%2' is the second, and so on. 192 193 The field name can have '/raw' suffix which disables pretty printing 194 and shows raw field value like hex numbers. The --raw-trace option 195 has the same effect for all dynamic sort keys. 196 197 The default sort keys are changed to 'trace' if all events in the data 198 file are tracepoint. 199 200-F:: 201--fields=:: 202 Specify output field - multiple keys can be specified in CSV format. 203 Following fields are available: 204 overhead, overhead_sys, overhead_us, overhead_children, sample, period, 205 weight1, weight2, weight3, ins_lat, p_stage_cyc and retire_lat. The 206 last 3 names are alias for the corresponding weights. When the weight 207 fields are used, they will show the average value of the weight. 208 209 Also it can contain any sort key(s). 210 211 By default, every sort keys not specified in -F will be appended 212 automatically. 213 214 If the keys starts with a prefix '+', then it will append the specified 215 field(s) to the default field order. For example: perf report -F +period,sample. 216 217-p:: 218--parent=<regex>:: 219 A regex filter to identify parent. The parent is a caller of this 220 function and searched through the callchain, thus it requires callchain 221 information recorded. The pattern is in the extended regex format and 222 defaults to "\^sys_|^do_page_fault", see '--sort parent'. 223 224-x:: 225--exclude-other:: 226 Only display entries with parent-match. 227 228-w:: 229--column-widths=<width[,width...]>:: 230 Force each column width to the provided list, for large terminal 231 readability. 0 means no limit (default behavior). 232 233-t:: 234--field-separator=:: 235 Use a special separator character and don't pad with spaces, replacing 236 all occurrences of this separator in symbol names (and other output) 237 with a '.' character, that thus it's the only non valid separator. 238 239-D:: 240--dump-raw-trace:: 241 Dump raw trace in ASCII. 242 243--disable-order:: 244 Disable raw trace ordering. 245 246-g:: 247--call-graph=<print_type,threshold[,print_limit],order,sort_key[,branch],value>:: 248 Display call chains using type, min percent threshold, print limit, 249 call order, sort key, optional branch and value. Note that ordering 250 is not fixed so any parameter can be given in an arbitrary order. 251 One exception is the print_limit which should be preceded by threshold. 252 253 print_type can be either: 254 - flat: single column, linear exposure of call chains. 255 - graph: use a graph tree, displaying absolute overhead rates. (default) 256 - fractal: like graph, but displays relative rates. Each branch of 257 the tree is considered as a new profiled object. 258 - folded: call chains are displayed in a line, separated by semicolons 259 - none: disable call chain display. 260 261 threshold is a percentage value which specifies a minimum percent to be 262 included in the output call graph. Default is 0.5 (%). 263 264 print_limit is only applied when stdio interface is used. It's to limit 265 number of call graph entries in a single hist entry. Note that it needs 266 to be given after threshold (but not necessarily consecutive). 267 Default is 0 (unlimited). 268 269 order can be either: 270 - callee: callee based call graph. 271 - caller: inverted caller based call graph. 272 Default is 'caller' when --children is used, otherwise 'callee'. 273 274 sort_key can be: 275 - function: compare on functions (default) 276 - address: compare on individual code addresses 277 - srcline: compare on source filename and line number 278 279 branch can be: 280 - branch: include last branch information in callgraph when available. 281 Usually more convenient to use --branch-history for this. 282 283 value can be: 284 - percent: display overhead percent (default) 285 - period: display event period 286 - count: display event count 287 288--children:: 289 Accumulate callchain of children to parent entry so that then can 290 show up in the output. The output will have a new "Children" column 291 and will be sorted on the data. It requires callchains are recorded. 292 See the `overhead calculation' section for more details. Enabled by 293 default, disable with --no-children. 294 295--max-stack:: 296 Set the stack depth limit when parsing the callchain, anything 297 beyond the specified depth will be ignored. This is a trade-off 298 between information loss and faster processing especially for 299 workloads that can have a very long callchain stack. 300 Note that when using the --itrace option the synthesized callchain size 301 will override this value if the synthesized callchain size is bigger. 302 303 Default: 127 304 305-G:: 306--inverted:: 307 alias for inverted caller based call graph. 308 309--ignore-callees=<regex>:: 310 Ignore callees of the function(s) matching the given regex. 311 This has the effect of collecting the callers of each such 312 function into one place in the call-graph tree. 313 314--pretty=<key>:: 315 Pretty printing style. key: normal, raw 316 317--stdio:: Use the stdio interface. 318 319--stdio-color:: 320 'always', 'never' or 'auto', allowing configuring color output 321 via the command line, in addition to via "color.ui" .perfconfig. 322 Use '--stdio-color always' to generate color even when redirecting 323 to a pipe or file. Using just '--stdio-color' is equivalent to 324 using 'always'. 325 326--tui:: Use the TUI interface, that is integrated with annotate and allows 327 zooming into DSOs or threads, among other features. Use of --tui 328 requires a tty, if one is not present, as when piping to other 329 commands, the stdio interface is used. 330 331--gtk:: Use the GTK2 interface. 332 333-k:: 334--vmlinux=<file>:: 335 vmlinux pathname 336 337--ignore-vmlinux:: 338 Ignore vmlinux files. 339 340--kallsyms=<file>:: 341 kallsyms pathname 342 343-m:: 344--modules:: 345 Load module symbols. WARNING: This should only be used with -k and 346 a LIVE kernel. 347 348-f:: 349--force:: 350 Don't do ownership validation. 351 352--symfs=<directory>:: 353 Look for files with symbols relative to this directory. 354 355-C:: 356--cpu:: Only report samples for the list of CPUs provided. Multiple CPUs can 357 be provided as a comma-separated list with no space: 0,1. Ranges of 358 CPUs are specified with -: 0-2. Default is to report samples on all 359 CPUs. 360 361-M:: 362--disassembler-style=:: Set disassembler style for objdump. 363 364--source:: 365 Interleave source code with assembly code. Enabled by default, 366 disable with --no-source. 367 368--asm-raw:: 369 Show raw instruction encoding of assembly instructions. 370 371--show-total-period:: Show a column with the sum of periods. 372 373-I:: 374--show-info:: 375 Display extended information about the perf.data file. This adds 376 information which may be very large and thus may clutter the display. 377 It currently includes: cpu and numa topology of the host system. 378 379-b:: 380--branch-stack:: 381 Use the addresses of sampled taken branches instead of the instruction 382 address to build the histograms. To generate meaningful output, the 383 perf.data file must have been obtained using perf record -b or 384 perf record --branch-filter xxx where xxx is a branch filter option. 385 perf report is able to auto-detect whether a perf.data file contains 386 branch stacks and it will automatically switch to the branch view mode, 387 unless --no-branch-stack is used. 388 389--branch-history:: 390 Add the addresses of sampled taken branches to the callstack. 391 This allows to examine the path the program took to each sample. 392 The data collection must have used -b (or -j) and -g. 393 394--addr2line=<path>:: 395 Path to addr2line binary. 396 397--objdump=<path>:: 398 Path to objdump binary. 399 400--prefix=PREFIX:: 401--prefix-strip=N:: 402 Remove first N entries from source file path names in executables 403 and add PREFIX. This allows to display source code compiled on systems 404 with different file system layout. 405 406--group:: 407 Show event group information together. It forces group output also 408 if there are no groups defined in data file. 409 410--group-sort-idx:: 411 Sort the output by the event at the index n in group. If n is invalid, 412 sort by the first event. It can support multiple groups with different 413 amount of events. WARNING: This should be used on grouped events. 414 415--demangle:: 416 Demangle symbol names to human readable form. It's enabled by default, 417 disable with --no-demangle. 418 419--demangle-kernel:: 420 Demangle kernel symbol names to human readable form (for C++ kernels). 421 422--mem-mode:: 423 Use the data addresses of samples in addition to instruction addresses 424 to build the histograms. To generate meaningful output, the perf.data 425 file must have been obtained using perf record -d -W and using a 426 special event -e cpu/mem-loads/p or -e cpu/mem-stores/p. See 427 'perf mem' for simpler access. 428 429--percent-limit:: 430 Do not show entries which have an overhead under that percent. 431 (Default: 0). Note that this option also sets the percent limit (threshold) 432 of callchains. However the default value of callchain threshold is 433 different than the default value of hist entries. Please see the 434 --call-graph option for details. 435 436--percentage:: 437 Determine how to display the overhead percentage of filtered entries. 438 Filters can be applied by --comms, --dsos and/or --symbols options and 439 Zoom operations on the TUI (thread, dso, etc). 440 441 "relative" means it's relative to filtered entries only so that the 442 sum of shown entries will be always 100%. "absolute" means it retains 443 the original value before and after the filter is applied. 444 445--header:: 446 Show header information in the perf.data file. This includes 447 various information like hostname, OS and perf version, cpu/mem 448 info, perf command line, event list and so on. Currently only 449 --stdio output supports this feature. 450 451--header-only:: 452 Show only perf.data header (forces --stdio). 453 454--time:: 455 Only analyze samples within given time window: <start>,<stop>. Times 456 have the format seconds.nanoseconds. If start is not given (i.e. time 457 string is ',x.y') then analysis starts at the beginning of the file. If 458 stop time is not given (i.e. time string is 'x.y,') then analysis goes 459 to end of file. Multiple ranges can be separated by spaces, which 460 requires the argument to be quoted e.g. --time "1234.567,1234.789 1235," 461 462 Also support time percent with multiple time ranges. Time string is 463 'a%/n,b%/m,...' or 'a%-b%,c%-%d,...'. 464 465 For example: 466 Select the second 10% time slice: 467 468 perf report --time 10%/2 469 470 Select from 0% to 10% time slice: 471 472 perf report --time 0%-10% 473 474 Select the first and second 10% time slices: 475 476 perf report --time 10%/1,10%/2 477 478 Select from 0% to 10% and 30% to 40% slices: 479 480 perf report --time 0%-10%,30%-40% 481 482--switch-on EVENT_NAME:: 483 Only consider events after this event is found. 484 485 This may be interesting to measure a workload only after some initialization 486 phase is over, i.e. insert a perf probe at that point and then using this 487 option with that probe. 488 489--switch-off EVENT_NAME:: 490 Stop considering events after this event is found. 491 492--show-on-off-events:: 493 Show the --switch-on/off events too. This has no effect in 'perf report' now 494 but probably we'll make the default not to show the switch-on/off events 495 on the --group mode and if there is only one event besides the off/on ones, 496 go straight to the histogram browser, just like 'perf report' with no events 497 explicitly specified does. 498 499--itrace:: 500 Options for decoding instruction tracing data. The options are: 501 502include::itrace.txt[] 503 504 To disable decoding entirely, use --no-itrace. 505 506--full-source-path:: 507 Show the full path for source files for srcline output. 508 509--show-ref-call-graph:: 510 When multiple events are sampled, it may not be needed to collect 511 callgraphs for all of them. The sample sites are usually nearby, 512 and it's enough to collect the callgraphs on a reference event. 513 So user can use "call-graph=no" event modifier to disable callgraph 514 for other events to reduce the overhead. 515 However, perf report cannot show callgraphs for the event which 516 disable the callgraph. 517 This option extends the perf report to show reference callgraphs, 518 which collected by reference event, in no callgraph event. 519 520--stitch-lbr:: 521 Show callgraph with stitched LBRs, which may have more complete 522 callgraph. The perf.data file must have been obtained using 523 perf record --call-graph lbr. 524 Disabled by default. In common cases with call stack overflows, 525 it can recreate better call stacks than the default lbr call stack 526 output. But this approach is not foolproof. There can be cases 527 where it creates incorrect call stacks from incorrect matches. 528 The known limitations include exception handing such as 529 setjmp/longjmp will have calls/returns not match. 530 531--socket-filter:: 532 Only report the samples on the processor socket that match with this filter 533 534--samples=N:: 535 Save N individual samples for each histogram entry to show context in perf 536 report tui browser. 537 538--raw-trace:: 539 When displaying traceevent output, do not use print fmt or plugins. 540 541-H:: 542--hierarchy:: 543 Enable hierarchical output. In the hierarchy mode, each sort key groups 544 samples based on the criteria and then sub-divide it using the lower 545 level sort key. 546 547 For example: 548 In normal output: 549 550 perf report -s dso,sym 551 # Overhead Shared Object Symbol 552 50.00% [kernel.kallsyms] [k] kfunc1 553 20.00% perf [.] foo 554 15.00% [kernel.kallsyms] [k] kfunc2 555 10.00% perf [.] bar 556 5.00% libc.so [.] libcall 557 558 In hierarchy output: 559 560 perf report -s dso,sym --hierarchy 561 # Overhead Shared Object / Symbol 562 65.00% [kernel.kallsyms] 563 50.00% [k] kfunc1 564 15.00% [k] kfunc2 565 30.00% perf 566 20.00% [.] foo 567 10.00% [.] bar 568 5.00% libc.so 569 5.00% [.] libcall 570 571--inline:: 572 If a callgraph address belongs to an inlined function, the inline stack 573 will be printed. Each entry is function name or file/line. Enabled by 574 default, disable with --no-inline. 575 576--mmaps:: 577 Show --tasks output plus mmap information in a format similar to 578 /proc/<PID>/maps. 579 580 Please note that not all mmaps are stored, options affecting which ones 581 are include 'perf record --data', for instance. 582 583--ns:: 584 Show time stamps in nanoseconds. 585 586--stats:: 587 Display overall events statistics without any further processing. 588 (like the one at the end of the perf report -D command) 589 590--tasks:: 591 Display monitored tasks stored in perf data. Displaying pid/tid/ppid 592 plus the command string aligned to distinguish parent and child tasks. 593 594--percent-type:: 595 Set annotation percent type from following choices: 596 global-period, local-period, global-hits, local-hits 597 598 The local/global keywords set if the percentage is computed 599 in the scope of the function (local) or the whole data (global). 600 The period/hits keywords set the base the percentage is computed 601 on - the samples period or the number of samples (hits). 602 603--time-quantum:: 604 Configure time quantum for time sort key. Default 100ms. 605 Accepts s, us, ms, ns units. 606 607--total-cycles:: 608 When --total-cycles is specified, it supports sorting for all blocks by 609 'Sampled Cycles%'. This is useful to concentrate on the globally hottest 610 blocks. In output, there are some new columns: 611 612 'Sampled Cycles%' - block sampled cycles aggregation / total sampled cycles 613 'Sampled Cycles' - block sampled cycles aggregation 614 'Avg Cycles%' - block average sampled cycles / sum of total block average 615 sampled cycles 616 'Avg Cycles' - block average sampled cycles 617 'Branch Counter' - block branch counter histogram (with -v showing the number) 618 619--skip-empty:: 620 Do not print 0 results in the --stat output. 621 622include::callchain-overhead-calculation.txt[] 623 624SEE ALSO 625-------- 626linkperf:perf-stat[1], linkperf:perf-annotate[1], linkperf:perf-record[1], 627linkperf:perf-intel-pt[1] 628