1 .\" Copyright (c) 2003-2008 Joseph Koshy
2 .\" Copyright (c) 2007 The FreeBSD Foundation
3 .\" All rights reserved.
5 .\" Redistribution and use in source and binary forms, with or without
6 .\" modification, are permitted provided that the following conditions
8 .\" 1. Redistributions of source code must retain the above copyright
9 .\" notice, this list of conditions and the following disclaimer.
10 .\" 2. Redistributions in binary form must reproduce the above copyright
11 .\" notice, this list of conditions and the following disclaimer in the
12 .\" documentation and/or other materials provided with the distribution.
14 .\" This software is provided by Joseph Koshy ``as is'' and
15 .\" any express or implied warranties, including, but not limited to, the
16 .\" implied warranties of merchantability and fitness for a particular purpose
17 .\" are disclaimed. in no event shall Joseph Koshy be liable
18 .\" for any direct, indirect, incidental, special, exemplary, or consequential
19 .\" damages (including, but not limited to, procurement of substitute goods
20 .\" or services; loss of use, data, or profits; or business interruption)
21 .\" however caused and on any theory of liability, whether in contract, strict
22 .\" liability, or tort (including negligence or otherwise) arising in any way
23 .\" out of the use of this software, even if advised of the possibility of
33 .Nd "performance measurement with performance monitoring hardware"
44 .Op Fl M Ar mapfilename
46 .Op Fl O Ar logfilename
47 .Op Fl P Ar event-spec
48 .Op Fl R Ar logfilename
49 .Op Fl S Ar event-spec
63 .Op Fl o Ar outputfile
64 .Op Fl p Ar event-spec
67 .Op Fl s Ar event-spec
68 .Op Fl t Ar process-spec
69 .Op Fl u Ar event-spec
72 .Op Fl z Ar graphdepth
73 .Op Ar command Op Ar args
77 utility measures system performance using the facilities provided by
82 utility can measure both hardware events seen by the system as a
83 whole, and those seen when a specified set of processes are executing
85 If a specific set of processes is being targeted (for example,
88 option is specified, or if a command line is specified using
90 then measurement occurs till
92 exits, or till all target processes specified by the
94 options exit, or till the
96 utility is interrupted by the user.
97 If a specific set of processes is not targeted for measurement, then
99 will perform system-wide measurements till interrupted by the
102 A given invocation of
104 can mix allocations of system-mode and process-mode PMCs, of both
105 counting and sampling flavors.
106 The values of all counting PMCs are printed in human readable form
107 at regular intervals by
111 human-readable textual output is not stable, and could change
113 The output of sampling PMCs may be configured to go to a log file for
114 subsequent offline analysis, or, at the expense of greater
115 overhead, may be configured to be printed in text form on the fly.
117 Hardware events to measure are specified to
119 using event specifier strings
121 The syntax of these event specifiers is machine dependent and is
125 A process-mode PMC may be configured to be inheritable by the target
126 process' current and future children.
128 The following options are available:
129 .Bl -tag -width indent
131 Skip symbol lookup and display address instead.
133 Toggle between showing cumulative or incremental counts for
134 subsequent counting mode PMCs specified on the command line.
135 The default is to show incremental counts.
137 Create files with per-program samples in the directory named
140 The default is to create these files in the current directory.
142 Toggle showing per-process counts at the time a tracked process
143 exits for subsequent process-mode PMCs specified on the command line.
144 This option is useful for mapping the performance characteristics of a
145 complex pipeline of processes when used in conjunction with the
148 The default is to not to enable per-process tracking.
150 Print calltree (Kcachegrind) information to file
156 this information is sent to the output file specified by the
160 Print callchain information to file
166 this information is sent to the output file specified by the
170 Show the offset of the instruction pointer into the symbol.
172 List all event names.
173 .It Fl M Ar mapfilename
174 Write the mapping between executable objects encountered in the event
175 log and the abbreviated pathnames used for
179 If this option is not specified, mapping information is not written.
184 in which case this mapping information is sent to the output
185 file configured by the
189 Toggle capturing callchain information for subsequent sampling PMCs.
190 The default is for sampling PMCs to capture callchain information.
191 .It Fl O Ar logfilename
192 Send logging output to file
197 .Ar hostname Ns : Ns Ar port ,
200 does not start with a
206 will open a network socket to host
213 option is not specified and one of the logging options is requested,
216 will print a textual form of the logged events to the configured
218 .It Fl P Ar event-spec
219 Allocate a process mode sampling PMC measuring hardware events
222 .It Fl R Ar logfilename
223 Perform offline analysis using sampling data in file
225 .It Fl S Ar event-spec
226 Allocate a system mode sampling PMC measuring hardware events
232 mode for sampling PMCs. The following hotkeys
235 .Bl -tag -compact -width "Ctrl+a" -offset 4n
237 Toggle symbol resolution
241 Switch to accumulative mode
250 threshold as a dot (calltree only)
252 Toggle showing offsets into symbols
265 Toggle capturing user-space call traces while in kernel mode.
266 The default is for sampling PMCs to capture user-space callchain information
267 while in user-space mode, and kernel callchain information while in kernel mode.
269 Toggle logging the incremental counts seen by the threads of a
270 tracked process each time they are scheduled on a CPU.
271 This is an experimental feature intended to help analyse the
272 dynamic behaviour of processes in the system.
273 It may incur substantial overhead if enabled.
274 The default is for this feature to be disabled.
276 Perform a symbol and file:line lookup for each address in each
277 callgraph and save the output to
281 that only resolves the first symbol in the graph, this resolves
282 every node in the callgraph, or prints out addresses if no
283 lookup information is available.
284 This option requires the
286 option to read in samples that were previously collected and
291 Set the cpus for subsequent system mode PMCs specified on the
296 is a comma separated list of CPU numbers, or the literal
298 denoting all available CPUs.
299 The default is to allocate system mode PMCs on all available
302 Toggle between process mode PMCs measuring events for the target
303 process' current and future children or only measuring events for
305 The default is to measure events for the target process alone.
306 (it has to be passed in the command line prior to
313 Specify that the gprof profile files will use a wide history counter.
314 These files are produced in a format compatible with
316 However, other tools that cannot fully parse a BSD-style
317 gmon header might be unable to correctly parse these files.
318 .It Fl f Ar pluginopt
319 Pass option string to the active plugin.
321 threshold=<float> do not display cost under specified value (Top).
323 skiplink=0|1 replace node with cost under threshold by a dot (Top).
325 Produce profiles in a format compatible with
327 A separate profile file is generated for each executable object
329 Profile files are placed in sub-directories named by their PMC
334 which you can get from
339 Set system-wide performance measurement duration for
344 may be a fractional value.
346 Print the sampled PCs with the name, the start and ending addresses
347 of the function within they live.
350 argument is mandatory and indicates where the information will be stored.
355 this information is sent to the output file specified by the
358 This option requires the
360 option to read in samples that were previously collected and
365 Set the default sampling rate for subsequent sampling mode
366 PMCs specified on the command line.
367 The default is to configure PMCs to sample the CPU's instruction
368 pointer every 65536 events.
369 .It Fl o Ar outputfile
370 Send counter readings and textual representations of logged data
373 The default is to send output to
375 when collecting live data and to
377 when processing a pre-existing logfile.
378 .It Fl p Ar event-spec
379 Allocate a process mode counting PMC measuring hardware events
385 Set the top of the filesystem hierarchy under which executables
386 are located to argument
390 .It Fl s Ar event-spec
391 Allocate a system mode counting PMC measuring hardware events
394 .It Fl t Ar process-spec
395 Attach process mode PMCs to the processes named by argument
399 may be a non-negative integer denoting a specific process id, or a
400 regular expression for selecting processes based on their command names.
401 .It Fl u Ar event-spec
402 Provide short description of event.
406 Print the values of all counting mode PMCs or sampling mode PMCs
412 may be a fractional value.
413 The default interval is 5 seconds.
414 .It Fl z Ar graphdepth
415 When printing system-wide callgraphs, limit callgraphs to the depth
416 specified by argument
422 is specified, it is executed using
425 To perform system-wide statistical sampling on an AMD Athlon CPU with
426 samples taken every 32768 instruction retirals and data being sampled
430 .Dl "pmcstat -O sample.stat -n 32768 -S k7-retired-instructions"
434 and measure the number of data cache misses suffered
435 by it and its children every 12 seconds on an AMD Athlon, use:
436 .Dl "pmcstat -d -w 12 -p k7-dc-misses firefox"
438 To measure instructions retired for all processes named
441 .Dl "pmcstat -t '^emacs$' -p instructions"
443 To measure instructions retired for processes named
445 for a period of 10 seconds use:
446 .Dl "pmcstat -t '^emacs$' -p instructions sleep 10"
448 To count instruction tlb-misses on CPUs 0 and 2 on a Intel
449 Pentium Pro/Pentium III SMP system use:
450 .Dl "pmcstat -c 0,2 -s p6-itlb-miss"
452 To collect profiling information for a specific process with pid 1234
453 based on instruction cache misses seen by it use:
454 .Dl "pmcstat -P ic-misses -t 1234 -O /tmp/sample.out"
456 To perform system-wide sampling on all configured processors
457 based on processor instructions retired use:
458 .Dl "pmcstat -S instructions -O /tmp/sample.out"
459 If callgraph capture is not desired use:
460 .Dl "pmcstat -N -S instructions -O /tmp/sample.out"
462 To send the generated event log to a remote machine use:
463 .Dl "pmcstat -S instructions -O remotehost:port"
464 On the remote machine, the sample log can be collected using
466 .Dl "nc -l remotehost port > /tmp/sample.out"
470 compatible profiles from a sample file use:
471 .Dl "pmcstat -R /tmp/sample.out -g"
473 To print a system-wide profile with callgraphs to file
476 .Dl "pmcstat -R /tmp/sample.out -G foo.graph"
482 may issue the following diagnostic messages:
484 .It "#callchain/dubious-frames"
485 The number of callchain records that had an
487 value for a return address.
488 .It "#exec handling errors"
491 events in the log file that named executables that could not be
496 events that named ELF executables.
500 events that named executables with unrecognized formats.
502 The total number of samples in the log file.
503 .It "#samples/unclaimed"
504 The number of samples that could not be correlated to a known
505 executable object (i.e., to an executable, shared library, the
506 kernel or the runtime loader).
507 .It "#samples/unknown-object"
508 The number of samples that were associated with an executable
509 with an unrecognized object format.
514 Due to the limitations of the
518 compatible profiles generated by the
520 option do not contain information about calls that cross executable
524 files are also only meaningful for native executables.
537 utility first appeared in
542 .An Joseph Koshy Aq Mt jkoshy@FreeBSD.org
546 utility cannot yet analyse
548 logs generated by non-native architectures.