1OPERF(1) General Commands Manual OPERF(1)
2
6 operf - Performance profiler tool for Linux
7
10 operf [ options ] [ --system-wide | --pid <pid> | [ command [ args ] ]
11 ]
12
15 Operf is the profiler tool provided with OProfile. Operf uses the Linux
16 Performance Events Subsystem and, thus, does not require the obsolete
17 oprofile kernel driver.
18 By default, operf uses <current_dir>/oprofile_data as the session-dir
20 and stores profiling data there. You can change this by way of the
21 --session-dir option. The usual post-profiling analysis tools such as
22 opreport(1) and opannotate(1) can be used to generate profile reports.
23 Unless a session-dir is specified, the post-processing analysis tools
24 will search for samples in <current_dir>/oprofile_data first. If that
25 directory does not exist, the post-processing tools use the standard
26 session-dir of /var/lib/oprofile.
27 Statistics, such as total samples received and lost samples, are writ‐
29 ten to the operf.log file that can be found in the <session_dir>/sam‐
30 ples directory.
31
34 One (and only one) of the following run modes must be specified:
35 command[args]
37 The command or application to be profiled. args are the input
38 arguments that the command or application requires.
39 --pid / -p PID
41 This option enables operf to profile a running application. PID
42 should be the process ID of the process you wish to profile.
43 When finished profiling (e.g., when the profiled process ends),
44 press Ctrl-c to stop operf. If you run operf --pid as a back‐
45 ground job (i.e., with the &), you must stop it in a controlled
46 manner in order for it to process the profile data it has col‐
47 lected. Use kill -SIGINT <operf-PID> for this purpose.
48 Limitation: When using this option to profile a multi-threaded
50 application that also forks new processes, be aware that samples
51 for processes that are forked before profiling is started may
52 not be recorded (depending on timing of thread creation and when
53 operf is started).
54 --system-wide / -s
56 This option is for performing a system-wide profile. You must
57 have root authority to run operf in this mode. When finished
58 profiling, Ctrl-c to stop operf. If you run operf --system-wide
59 as a background job (i.e., with the &), you must stop it in a
60 controlled manner in order for it to process the profile data it
61 has collected. Use kill -SIGINT <operf-PID> for this purpose.
62 It is recommended that when running operf with this option, the
63 user's current working directory should be /root or a subdirec‐
64 tory of /root to avoid storing sample data files in locations
65 accessible by regular users.
66
68 --vmlinux / -k vmlinux_path
69 A vmlinux file that matches the running kernel that has symbol
70 and/or debuginfo. Kernel samples will be attributed to this
71 binary, allowing post-processing tools (like opreport) to
72 attribute samples to the appropriate kernel symbols.
73 The kernel symbol information may be obtained from /proc/kall‐
75 syms if the user does not specify a vmlinux file. The symbol
76 addresses are given in /proc/kallsyms if permitted by the set‐
77 ting of /proc/sys/kernel/kptr_restrict.
78 If the --vmlinux option is not used and kernel symbols cannot be
80 obtained from /proc/kallsyms, then all kernel samples are at‐
81 tributed to "no-vmlinux", which is simply a bucket to hold the
82 samples and not an actual file.
83 --events / -e event1[,event2[,...]]
85 This option is for passing a comma-separated list of event spec‐
86 ifications for profiling. Each event spec is of the form:
87 name:count[:unitmask[:kernel[:user]]]
88 The count value is used to control the sampling rate for profil‐
90 ing; it is the number of events to occur between samples. The
91 rate is lowered by specifying a higher count value — i.e., a
92 higher number of events to occur between samples.
93 You can specify unitmask values using either a numerical value
95 (hex values must begin with "0x") or a symbolic name (if the
96 name=<um_name> field is shown in the ophelp output). For some
97 named unit masks, the hex value is not unique; thus, OProfile
98 tools enforce specifying such unit masks value by name. If no
99 unit mask is specified, the default unit mask value for the
100 event is used.
101 The kernel and user parts of the event specification are binary
103 values ('1' or '0') indicating whether or not to collect samples
104 for kernel space and user space.
105 Note: In order to specify the kernel/user bits, you must also
106 specify a unitmask value, even if the processor type (or the
107 specified event) does not use unit masks — in which case, use
108 the value '0' to signify a null unit mask; for example:
109 -e INST_RETIRED_ANY_P:100000:0:1:0
110 ^ ^ ^ ^
111 | | | |--- '0': do not record user
112 space samples
113 | | |-- '1': record kernel space
114 samples
115 | |-- '0': the null unit mask
116 |--count value
117 Event names for some IBM PowerPC systems include a _GRP<n>
119 (group number) suffix. You can pass either the full event name
120 or the base event name (i.e., without the suffix) to operf. If
121 the base event name is passed, operf will automatically choose
122 an appropriate group number suffix for the event; thus, OProfile
123 post-processing tools will always show real event names that
124 include the group number suffix.
125 When no event specification is given, the default event for the
127 running processor type will be used for profiling. Use ophelp
128 to list the available events for your processor type.
129 --callgraph / -g
131 This option enables the callgraph to be saved during profiling.
132 NOTE: The full callchain is recorded, so there is no depth
133 limit.
134 --separate-thread / -t
136 This option categorizes samples by thread group ID (tgid) and
137 thread ID (tid). The '--separate-thread' option is useful for
138 seeing per-thread samples in multi-threaded applications. When
139 used in conjunction with the '--system-wide' option, the '--sep‐
140 arate-thread' option is also useful for seeing per-process
141 (i.e., per-thread group) samples for the case where multiple
142 processes are executing the same program during a profiling run.
143 --separate-cpu / -c
145 This option categorizes samples by cpu.
146 --session-dir / -d path
148 This option specifies the session path to hold the sample data.
149 If not specified, the data is saved in the oprofile_data direc‐
150 tory on the current path.
151 --lazy-conversion / -l
153 Use this option to reduce the overhead of operf during profil‐
154 ing. Normally, profile data received from the kernel is con‐
155 verted to OProfile format during profiling time. This is typi‐
156 cally not an issue when profiling a single application. But when
157 using the --system-wide option, this on-the-fly conversion
158 process can cause noticeable overhead, particularly on busy
159 multi-processor systems. The --lazy-conversion option directs
160 operf to wait until profiling is completed to do the conversion
161 of profile data.
162 Note: This option is not recommended to be used in conjunction
164 with the --pid option for profiling multi-threaded processes.
165 Depending on the order of thread creation (or forking of new
166 processes), you may not get any samples for the new threads/pro‐
167 cesses.
168 --append / -a
170 By default, operf moves old profile data from <session_dir>/sam‐
171 ples/current to <session_dir>/samples/previous. If a 'previous'
172 profile already existed, it will be replaced. If the --append
173 option is passed, old profile data is left in place and new pro‐
174 file data will be added to it, and the 'previous' profile (if
175 one existed) will remain untouched. To access the 'previous'
176 profile, simply add a session specification to the normal invo‐
177 cation of oprofile post-processing tools. For example:
178 opreport session:previous
179 --verbose / -V level
181 A comma-separated list of debugging control values, used to
182 increase the verbosity of the output. Valid values are: debug,
183 record, convert, misc, sfile, arcs, or the special value, 'all'.
184 --version / -v
186 Show operf version.
187 --help / -h
189 Display brief usage message.
190 --usage / -u
192 Display brief usage message.
193
196 $ operf make
197
200 This man page is current for oprofile-1.3.0.
201
204 opreport(1), opannotate(1).
205oprofile 1.3.0 Thu 15 November 2018 OPERF(1)