xref: /freebsd/usr.sbin/pmcstat/pmcstat.8 (revision 325151a3)
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 May 27, 2015
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 a Ar pathname
50.Op Fl c Ar cpu-spec
51.Op Fl d
52.Op Fl f Ar pluginopt
53.Op Fl g
54.Op Fl k Ar kerneldir
55.Op Fl l Ar secs
56.Op Fl m Ar pathname
57.Op Fl n Ar rate
58.Op Fl o Ar outputfile
59.Op Fl p Ar event-spec
60.Op Fl q
61.Op Fl r Ar fsroot
62.Op Fl s Ar event-spec
63.Op Fl t Ar process-spec
64.Op Fl v
65.Op Fl w Ar secs
66.Op Fl z Ar graphdepth
67.Op Ar command Op Ar args
68.Sh DESCRIPTION
69The
70.Nm
71utility measures system performance using the facilities provided by
72.Xr hwpmc 4 .
73.Pp
74The
75.Nm
76utility can measure both hardware events seen by the system as a
77whole, and those seen when a specified set of processes are executing
78on the system's CPUs.
79If a specific set of processes is being targeted (for example,
80if the
81.Fl t Ar process-spec
82option is specified, or if a command line is specified using
83.Ar command ) ,
84then measurement occurs till
85.Ar command
86exits, or till all target processes specified by the
87.Fl t Ar process-spec
88options exit, or till the
89.Nm
90utility is interrupted by the user.
91If a specific set of processes is not targeted for measurement, then
92.Nm
93will perform system-wide measurements till interrupted by the
94user.
95.Pp
96A given invocation of
97.Nm
98can mix allocations of system-mode and process-mode PMCs, of both
99counting and sampling flavors.
100The values of all counting PMCs are printed in human readable form
101at regular intervals by
102.Nm .
103The output of sampling PMCs may be configured to go to a log file for
104subsequent offline analysis, or, at the expense of greater
105overhead, may be configured to be printed in text form on the fly.
106.Pp
107Hardware events to measure are specified to
108.Nm
109using event specifier strings
110.Ar event-spec .
111The syntax of these event specifiers is machine dependent and is
112documented in
113.Xr pmc 3 .
114.Pp
115A process-mode PMC may be configured to be inheritable by the target
116process' current and future children.
117.Sh OPTIONS
118The following options are available:
119.Bl -tag -width indent
120.It Fl C
121Toggle between showing cumulative or incremental counts for
122subsequent counting mode PMCs specified on the command line.
123The default is to show incremental counts.
124.It Fl D Ar pathname
125Create files with per-program samples in the directory named
126by
127.Ar pathname .
128The default is to create these files in the current directory.
129.It Fl E
130Toggle showing per-process counts at the time a tracked process
131exits for subsequent process-mode PMCs specified on the command line.
132This option is useful for mapping the performance characteristics of a
133complex pipeline of processes when used in conjunction with the
134.Fl d
135option.
136The default is to not to enable per-process tracking.
137.It Fl F Ar pathname
138Print calltree (Kcachegrind) information to file
139.Ar pathname .
140If argument
141.Ar pathname
142is a
143.Dq Li -
144this information is sent to the output file specified by the
145.Fl o
146option.
147.It Fl G Ar pathname
148Print callchain information to file
149.Ar pathname .
150If argument
151.Ar pathname
152is a
153.Dq Li -
154this information is sent to the output file specified by the
155.Fl o
156option.
157.It Fl M Ar mapfilename
158Write the mapping between executable objects encountered in the event
159log and the abbreviated pathnames used for
160.Xr gprof 1
161profiles to file
162.Ar mapfilename .
163If this option is not specified, mapping information is not written.
164Argument
165.Ar mapfilename
166may be a
167.Dq Li -
168in which case this mapping information is sent to the output
169file configured by the
170.Fl o
171option.
172.It Fl N
173Toggle capturing callchain information for subsequent sampling PMCs.
174The default is for sampling PMCs to capture callchain information.
175.It Fl O Ar logfilename
176Send logging output to file
177.Ar logfilename .
178If
179.Ar logfilename
180is of the form
181.Ar hostname Ns : Ns Ar port ,
182where
183.Ar hostname
184does not start with a
185.Ql \&.
186or a
187.Ql / ,
188then
189.Nm
190will open a network socket to host
191.Ar hostname
192on port
193.Ar port .
194.Pp
195If the
196.Fl O
197option is not specified and one of the logging options is requested,
198then
199.Nm
200will print a textual form of the logged events to the configured
201output file.
202.It Fl P Ar event-spec
203Allocate a process mode sampling PMC measuring hardware events
204specified in
205.Ar event-spec .
206.It Fl R Ar logfilename
207Perform offline analysis using sampling data in file
208.Ar logfilename .
209.It Fl S Ar event-spec
210Allocate a system mode sampling PMC measuring hardware events
211specified in
212.Ar event-spec .
213.It Fl T
214Use a top like mode for sampling PMCs. The following hotkeys
215can be used: 'c+a' switch to accumulative mode, 'c+d' switch
216to delta mode, 'm' merge PMCs, 'n' change view, 'p' show next
217PMC, ' ' pause, 'q' quit. calltree only: 'f' cost under threshold
218is seen as a dot.
219.It Fl W
220Toggle logging the incremental counts seen by the threads of a
221tracked process each time they are scheduled on a CPU.
222This is an experimental feature intended to help analyse the
223dynamic behaviour of processes in the system.
224It may incur substantial overhead if enabled.
225The default is for this feature to be disabled.
226.It Fl a Ar pathname
227Perform a symbol and file:line lookup for each address in each
228callgraph and save the output to
229.Ar pathname .
230Unlike
231.Fl m
232that only resolves the first symbol in the graph, this resolves
233every node in the callgraph, or prints out addresses if no
234lookup information is available.
235This option requires the
236.Fl R
237option to read in samples that were previously collected and
238saved with the
239.Fl O
240option.
241.It Fl c Ar cpu-spec
242Set the cpus for subsequent system mode PMCs specified on the
243command line to
244.Ar cpu-spec .
245Argument
246.Ar cpu-spec
247is a comma separated list of CPU numbers, or the literal
248.Sq *
249denoting all available CPUs.
250The default is to allocate system mode PMCs on all available
251CPUs.
252.It Fl d
253Toggle between process mode PMCs measuring events for the target
254process' current and future children or only measuring events for
255the target process.
256The default is to measure events for the target process alone.
257(it has to be passed in the command line prior to
258.Fl p ,
259.Fl s ,
260.Fl P ,
261or
262.Fl S ) .
263.It Fl f Ar pluginopt
264Pass option string to the active plugin.
265.br
266threshold=<float> do not display cost under specified value (Top).
267.br
268skiplink=0|1 replace node with cost under threshold by a dot (Top).
269.It Fl g
270Produce profiles in a format compatible with
271.Xr gprof 1 .
272A separate profile file is generated for each executable object
273encountered.
274Profile files are placed in sub-directories named by their PMC
275event name.
276.It Fl k Ar kerneldir
277Set the pathname of the kernel directory to argument
278.Ar kerneldir .
279This directory specifies where
280.Nm
281should look for the kernel and its modules.
282The default is to use the path of the running kernel obtained from the
283.Va kern.bootfile
284sysctl.
285.It Fl l Ar secs
286Set system-wide performance measurement duration for
287.Ar secs
288seconds.
289The argument
290.Ar secs
291may be a fractional value.
292.It Fl m Ar pathname
293Print the sampled PCs with the name, the start and ending addresses
294of the function within they live.
295The
296.Ar pathname
297argument is mandatory and indicates where the information will be stored.
298If argument
299.Ar pathname
300is a
301.Dq Li -
302this information is sent to the output file specified by the
303.Fl o
304option.
305This option requires the
306.Fl R
307option to read in samples that were previously collected and
308saved with the
309.Fl O
310option.
311.It Fl n Ar rate
312Set the default sampling rate for subsequent sampling mode
313PMCs specified on the command line.
314The default is to configure PMCs to sample the CPU's instruction
315pointer every 65536 events.
316.It Fl o Ar outputfile
317Send counter readings and textual representations of logged data
318to file
319.Ar outputfile .
320The default is to send output to
321.Pa stderr
322when collecting live data and to
323.Pa stdout
324when processing a pre-existing logfile.
325.It Fl p Ar event-spec
326Allocate a process mode counting PMC measuring hardware events
327specified in
328.Ar event-spec .
329.It Fl q
330Decrease verbosity.
331.It Fl r Ar fsroot
332Set the top of the filesystem hierarchy under which executables
333are located to argument
334.Ar fsroot .
335The default is
336.Pa / .
337.It Fl s Ar event-spec
338Allocate a system mode counting PMC measuring hardware events
339specified in
340.Ar event-spec .
341.It Fl t Ar process-spec
342Attach process mode PMCs to the processes named by argument
343.Ar process-spec .
344Argument
345.Ar process-spec
346may be a non-negative integer denoting a specific process id, or a
347regular expression for selecting processes based on their command names.
348.It Fl v
349Increase verbosity.
350.It Fl w Ar secs
351Print the values of all counting mode PMCs or sampling mode PMCs
352for top mode every
353.Ar secs
354seconds.
355The argument
356.Ar secs
357may be a fractional value.
358The default interval is 5 seconds.
359.It Fl z Ar graphdepth
360When printing system-wide callgraphs, limit callgraphs to the depth
361specified by argument
362.Ar graphdepth .
363.El
364.Pp
365If
366.Ar command
367is specified, it is executed using
368.Xr execvp 3 .
369.Sh EXAMPLES
370To perform system-wide statistical sampling on an AMD Athlon CPU with
371samples taken every 32768 instruction retirals and data being sampled
372to file
373.Pa sample.stat ,
374use:
375.Dl "pmcstat -O sample.stat -n 32768 -S k7-retired-instructions"
376.Pp
377To execute
378.Nm firefox
379and measure the number of data cache misses suffered
380by it and its children every 12 seconds on an AMD Athlon, use:
381.Dl "pmcstat -d -w 12 -p k7-dc-misses firefox"
382.Pp
383To measure instructions retired for all processes named
384.Dq emacs
385use:
386.Dl "pmcstat -t '^emacs$' -p instructions"
387.Pp
388To measure instructions retired for processes named
389.Dq emacs
390for a period of 10 seconds use:
391.Dl "pmcstat -t '^emacs$' -p instructions sleep 10"
392.Pp
393To count instruction tlb-misses on CPUs 0 and 2 on a Intel
394Pentium Pro/Pentium III SMP system use:
395.Dl "pmcstat -c 0,2 -s p6-itlb-miss"
396.Pp
397To collect profiling information for a specific process with pid 1234
398based on instruction cache misses seen by it use:
399.Dl "pmcstat -P ic-misses -t 1234 -O /tmp/sample.out"
400.Pp
401To perform system-wide sampling on all configured processors
402based on processor instructions retired use:
403.Dl "pmcstat -S instructions -O /tmp/sample.out"
404If callgraph capture is not desired use:
405.Dl "pmcstat -N -S instructions -O /tmp/sample.out"
406.Pp
407To send the generated event log to a remote machine use:
408.Dl "pmcstat -S instructions -O remotehost:port"
409On the remote machine, the sample log can be collected using
410.Xr nc 1 :
411.Dl "nc -l remotehost port > /tmp/sample.out"
412.Pp
413To generate
414.Xr gprof 1
415compatible profiles from a sample file use:
416.Dl "pmcstat -R /tmp/sample.out -g"
417.Pp
418To print a system-wide profile with callgraphs to file
419.Pa "foo.graph"
420use:
421.Dl "pmcstat -R /tmp/sample.out -G foo.graph"
422.Sh DIAGNOSTICS
423If option
424.Fl v
425is specified,
426.Nm
427may issue the following diagnostic messages:
428.Bl -diag
429.It "#callchain/dubious-frames"
430The number of callchain records that had an
431.Dq impossible
432value for a return address.
433.It "#exec handling errors"
434The number of
435.Xr exec 2
436events in the log file that named executables that could not be
437analyzed.
438.It "#exec/elf"
439The number of
440.Xr exec 2
441events that named ELF executables.
442.It "#exec/unknown"
443The number of
444.Xr exec 2
445events that named executables with unrecognized formats.
446.It "#samples/total"
447The total number of samples in the log file.
448.It "#samples/unclaimed"
449The number of samples that could not be correlated to a known
450executable object (i.e., to an executable, shared library, the
451kernel or the runtime loader).
452.It "#samples/unknown-object"
453The number of samples that were associated with an executable
454with an unrecognized object format.
455.El
456.Pp
457.Ex -std
458.Sh COMPATIBILITY
459Due to the limitations of the
460.Pa gmon.out
461file format,
462.Xr gprof 1
463compatible profiles generated by the
464.Fl g
465option do not contain information about calls that cross executable
466boundaries.
467The generated
468.Pa gmon.out
469files are also only meaningful for native executables.
470.Sh SEE ALSO
471.Xr gprof 1 ,
472.Xr nc 1 ,
473.Xr execvp 3 ,
474.Xr pmc 3 ,
475.Xr pmclog 3 ,
476.Xr hwpmc 4 ,
477.Xr pmccontrol 8 ,
478.Xr sysctl 8
479.Sh HISTORY
480The
481.Nm
482utility first appeared in
483.Fx 6.0 .
484It is
485.Ud
486.Sh AUTHORS
487.An Joseph Koshy Aq Mt jkoshy@FreeBSD.org
488.Sh BUGS
489The
490.Nm
491utility cannot yet analyse
492.Xr hwpmc 4
493logs generated by non-native architectures.
494