xref: /freebsd/usr.sbin/pmcstat/pmcstat.8 (revision 38f0b757fd84d17d0fc24739a7cda160c4516d81)
1.\" Copyright (c) 2003-2008 Joseph Koshy
2.\" Copyright (c) 2007 The FreeBSD Foundation
3.\" All rights reserved.
4.\"
5.\" Redistribution and use in source and binary forms, with or without
6.\" modification, are permitted provided that the following conditions
7.\" are met:
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.
13.\"
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
24.\" such damage.
25.\"
26.\" $FreeBSD$
27.\"
28.Dd September 19, 2008
29.Dt PMCSTAT 8
30.Os
31.Sh NAME
32.Nm pmcstat
33.Nd "performance measurement with performance monitoring hardware"
34.Sh SYNOPSIS
35.Nm
36.Op Fl C
37.Op Fl D Ar pathname
38.Op Fl E
39.Op Fl F Ar pathname
40.Op Fl G Ar pathname
41.Op Fl M Ar mapfilename
42.Op Fl N
43.Op Fl O Ar logfilename
44.Op Fl P Ar event-spec
45.Op Fl R Ar logfilename
46.Op Fl S Ar event-spec
47.Op Fl T
48.Op Fl W
49.Op Fl c Ar cpu-spec
50.Op Fl d
51.Op Fl f Ar pluginopt
52.Op Fl g
53.Op Fl k Ar kerneldir
54.Op Fl m Ar pathname
55.Op Fl n Ar rate
56.Op Fl o Ar outputfile
57.Op Fl p Ar event-spec
58.Op Fl q
59.Op Fl r Ar fsroot
60.Op Fl s Ar event-spec
61.Op Fl t Ar process-spec
62.Op Fl v
63.Op Fl w Ar secs
64.Op Fl z Ar graphdepth
65.Op Ar command Op Ar args
66.Sh DESCRIPTION
67The
68.Nm
69utility measures system performance using the facilities provided by
70.Xr hwpmc 4 .
71.Pp
72The
73.Nm
74utility can measure both hardware events seen by the system as a
75whole, and those seen when a specified set of processes are executing
76on the system's CPUs.
77If a specific set of processes is being targeted (for example,
78if the
79.Fl t Ar process-spec
80option is specified, or if a command line is specified using
81.Ar command ) ,
82then measurement occurs till
83.Ar command
84exits, or till all target processes specified by the
85.Fl t Ar process-spec
86options exit, or till the
87.Nm
88utility is interrupted by the user.
89If a specific set of processes is not targeted for measurement, then
90.Nm
91will perform system-wide measurements till interrupted by the
92user.
93.Pp
94A given invocation of
95.Nm
96can mix allocations of system-mode and process-mode PMCs, of both
97counting and sampling flavors.
98The values of all counting PMCs are printed in human readable form
99at regular intervals by
100.Nm .
101The output of sampling PMCs may be configured to go to a log file for
102subsequent offline analysis, or, at the expense of greater
103overhead, may be configured to be printed in text form on the fly.
104.Pp
105Hardware events to measure are specified to
106.Nm
107using event specifier strings
108.Ar event-spec .
109The syntax of these event specifiers is machine dependent and is
110documented in
111.Xr pmc 3 .
112.Pp
113A process-mode PMC may be configured to be inheritable by the target
114process' current and future children.
115.Sh OPTIONS
116The following options are available:
117.Bl -tag -width indent
118.It Fl C
119Toggle between showing cumulative or incremental counts for
120subsequent counting mode PMCs specified on the command line.
121The default is to show incremental counts.
122.It Fl D Ar pathname
123Create files with per-program samples in the directory named
124by
125.Ar pathname .
126The default is to create these files in the current directory.
127.It Fl E
128Toggle showing per-process counts at the time a tracked process
129exits for subsequent process-mode PMCs specified on the command line.
130This option is useful for mapping the performance characteristics of a
131complex pipeline of processes when used in conjunction with the
132.Fl d
133option.
134The default is to not to enable per-process tracking.
135.It Fl F Ar pathname
136Print calltree (Kcachegrind) information to file
137.Ar pathname .
138If argument
139.Ar pathname
140is a
141.Dq Li -
142this information is sent to the output file specified by the
143.Fl o
144option.
145.It Fl G Ar pathname
146Print callchain information to file
147.Ar pathname .
148If argument
149.Ar pathname
150is a
151.Dq Li -
152this information is sent to the output file specified by the
153.Fl o
154option.
155.It Fl M Ar mapfilename
156Write the mapping between executable objects encountered in the event
157log and the abbreviated pathnames used for
158.Xr gprof 1
159profiles to file
160.Ar mapfilename .
161If this option is not specified, mapping information is not written.
162Argument
163.Ar mapfilename
164may be a
165.Dq Li -
166in which case this mapping information is sent to the output
167file configured by the
168.Fl o
169option.
170.It Fl N
171Toggle capturing callchain information for subsequent sampling PMCs.
172The default is for sampling PMCs to capture callchain information.
173.It Fl O Ar logfilename
174Send logging output to file
175.Ar logfilename .
176If
177.Ar logfilename
178is of the form
179.Ar hostname Ns : Ns Ar port ,
180where
181.Ar hostname
182does not start with a
183.Ql \&.
184or a
185.Ql / ,
186then
187.Nm
188will open a network socket to host
189.Ar hostname
190on port
191.Ar port .
192.Pp
193If the
194.Fl O
195option is not specified and one of the logging options is requested,
196then
197.Nm
198will print a textual form of the logged events to the configured
199output file.
200.It Fl P Ar event-spec
201Allocate a process mode sampling PMC measuring hardware events
202specified in
203.Ar event-spec .
204.It Fl R Ar logfilename
205Perform offline analysis using sampling data in file
206.Ar logfilename .
207.It Fl S Ar event-spec
208Allocate a system mode sampling PMC measuring hardware events
209specified in
210.Ar event-spec .
211.It Fl T
212Use a top like mode for sampling PMCs. The following hotkeys
213can be used: 'c+a' switch to accumulative mode, 'c+d' switch
214to delta mode, 'm' merge PMCs, 'n' change view, 'p' show next
215PMC, ' ' pause, 'q' quit. calltree only: 'f' cost under threshold
216is seen as a dot.
217.It Fl W
218Toggle logging the incremental counts seen by the threads of a
219tracked process each time they are scheduled on a CPU.
220This is an experimental feature intended to help analyse the
221dynamic behaviour of processes in the system.
222It may incur substantial overhead if enabled.
223The default is for this feature to be disabled.
224.It Fl c Ar cpu-spec
225Set the cpus for subsequent system mode PMCs specified on the
226command line to
227.Ar cpu-spec .
228Argument
229.Ar cpu-spec
230is a comma separated list of CPU numbers, or the literal
231.Sq *
232denoting all unhalted CPUs.
233The default is to allocate system mode PMCs on all unhalted
234CPUs.
235.It Fl d
236Toggle between process mode PMCs measuring events for the target
237process' current and future children or only measuring events for
238the target process.
239The default is to measure events for the target process alone.
240.It Fl f Ar pluginopt
241Pass option string to the active plugin.
242.br
243threshold=<float> do not display cost under specified value (Top).
244.br
245skiplink=0|1 replace node with cost under threshold by a dot (Top).
246.It Fl g
247Produce profiles in a format compatible with
248.Xr gprof 1 .
249A separate profile file is generated for each executable object
250encountered.
251Profile files are placed in sub-directories named by their PMC
252event name.
253.It Fl k Ar kerneldir
254Set the pathname of the kernel directory to argument
255.Ar kerneldir .
256This directory specifies where
257.Nm
258should look for the kernel and its modules.
259The default is
260.Pa /boot/kernel .
261.It Fl m Ar pathname
262Print the sampled PCs with the name, the start and ending addresses
263of the function within they live.
264The
265.Ar pathname
266argument is mandatory and indicates where the information will be stored.
267If argument
268.Ar pathname
269is a
270.Dq Li -
271this information is sent to the output file specified by the
272.Fl o
273option.
274.It Fl n Ar rate
275Set the default sampling rate for subsequent sampling mode
276PMCs specified on the command line.
277The default is to configure PMCs to sample the CPU's instruction
278pointer every 65536 events.
279.It Fl o Ar outputfile
280Send counter readings and textual representations of logged data
281to file
282.Ar outputfile .
283The default is to send output to
284.Pa stderr
285when collecting live data and to
286.Pa stdout
287when processing a pre-existing logfile.
288.It Fl p Ar event-spec
289Allocate a process mode counting PMC measuring hardware events
290specified in
291.Ar event-spec .
292.It Fl q
293Decrease verbosity.
294.It Fl r Ar fsroot
295Set the top of the filesystem hierarchy under which executables
296are located to argument
297.Ar fsroot .
298The default is
299.Pa / .
300.It Fl s Ar event-spec
301Allocate a system mode counting PMC measuring hardware events
302specified in
303.Ar event-spec .
304.It Fl t Ar process-spec
305Attach process mode PMCs to the processes named by argument
306.Ar process-spec .
307Argument
308.Ar process-spec
309may be a non-negative integer denoting a specific process id, or a
310regular expression for selecting processes based on their command names.
311.It Fl v
312Increase verbosity.
313.It Fl w Ar secs
314Print the values of all counting mode PMCs or sampling mode PMCs
315for top mode every
316.Ar secs
317seconds.
318The argument
319.Ar secs
320may be a fractional value.
321The default interval is 5 seconds.
322.It Fl z Ar graphdepth
323When printing system-wide callgraphs, limit callgraphs to the depth
324specified by argument
325.Ar graphdepth .
326.El
327.Pp
328If
329.Ar command
330is specified, it is executed using
331.Xr execvp 3 .
332.Sh EXAMPLES
333To perform system-wide statistical sampling on an AMD Athlon CPU with
334samples taken every 32768 instruction retirals and data being sampled
335to file
336.Pa sample.stat ,
337use:
338.Dl "pmcstat -O sample.stat -n 32768 -S k7-retired-instructions"
339.Pp
340To execute
341.Nm firefox
342and measure the number of data cache misses suffered
343by it and its children every 12 seconds on an AMD Athlon, use:
344.Dl "pmcstat -d -w 12 -p k7-dc-misses firefox"
345.Pp
346To measure instructions retired for all processes named
347.Dq emacs
348use:
349.Dl "pmcstat -t '^emacs$' -p instructions"
350.Pp
351To measure instructions retired for processes named
352.Dq emacs
353for a period of 10 seconds use:
354.Dl "pmcstat -t '^emacs$' -p instructions sleep 10"
355.Pp
356To count instruction tlb-misses on CPUs 0 and 2 on a Intel
357Pentium Pro/Pentium III SMP system use:
358.Dl "pmcstat -c 0,2 -s p6-itlb-miss"
359.Pp
360To collect profiling information for a specific process with pid 1234
361based on instruction cache misses seen by it use:
362.Dl "pmcstat -P ic-misses -t 1234 -O /tmp/sample.out"
363.Pp
364To perform system-wide sampling on all configured processors
365based on processor instructions retired use:
366.Dl "pmcstat -S instructions -O /tmp/sample.out"
367If callgraph capture is not desired use:
368.Dl "pmcstat -N -S instructions -O /tmp/sample.out"
369.Pp
370To send the generated event log to a remote machine use:
371.Dl "pmcstat -S instructions -O remotehost:port"
372On the remote machine, the sample log can be collected using
373.Xr nc 1 :
374.Dl "nc -l remotehost port > /tmp/sample.out"
375.Pp
376To generate
377.Xr gprof 1
378compatible profiles from a sample file use:
379.Dl "pmcstat -R /tmp/sample.out -g"
380.Pp
381To print a system-wide profile with callgraphs to file
382.Pa "foo.graph"
383use:
384.Dl "pmcstat -R /tmp/sample.out -G foo.graph"
385.Sh DIAGNOSTICS
386If option
387.Fl v
388is specified,
389.Nm
390may issue the following diagnostic messages:
391.Bl -diag
392.It "#callchain/dubious-frames"
393The number of callchain records that had an
394.Dq impossible
395value for a return address.
396.It "#exec handling errors"
397The number of
398.Xr exec 2
399events in the log file that named executables that could not be
400analyzed.
401.It "#exec/elf"
402The number of
403.Xr exec 2
404events that named ELF executables.
405.It "#exec/unknown"
406The number of
407.Xr exec 2
408events that named executables with unrecognized formats.
409.It "#samples/total"
410The total number of samples in the log file.
411.It "#samples/unclaimed"
412The number of samples that could not be correlated to a known
413executable object (i.e., to an executable, shared library, the
414kernel or the runtime loader).
415.It "#samples/unknown-object"
416The number of samples that were associated with an executable
417with an unrecognized object format.
418.El
419.Pp
420.Ex -std
421.Sh COMPATIBILITY
422Due to the limitations of the
423.Pa gmon.out
424file format,
425.Xr gprof 1
426compatible profiles generated by the
427.Fl g
428option do not contain information about calls that cross executable
429boundaries.
430The generated
431.Pa gmon.out
432files are also only meaningful for native executables.
433.Sh SEE ALSO
434.Xr gprof 1 ,
435.Xr nc 1 ,
436.Xr execvp 3 ,
437.Xr pmc 3 ,
438.Xr pmclog 3 ,
439.Xr hwpmc 4 ,
440.Xr pmccontrol 8 ,
441.Xr sysctl 8
442.Sh HISTORY
443The
444.Nm
445utility first appeared in
446.Fx 6.0 .
447It is
448.Ud
449.Sh AUTHORS
450.An Joseph Koshy Aq jkoshy@FreeBSD.org
451.Sh BUGS
452The
453.Nm
454utility cannot yet analyse
455.Xr hwpmc 4
456logs generated by non-native architectures.
457