Chapter 23. OProfile
OProfile is a low overhead, system-wide performance monitoring tool provided by the oprofile package. It uses performance monitoring hardware on the system’s processor to retrieve information about the kernel and executables on the system, such as when memory is referenced, the number of second-level cache requests, and the number of hardware interrupts received. OProfile is also able to profile applications that run in a Java Virtual Machine (JVM).
The following is a selection of the tools provided by OProfile:
- Displays available events for the system’s processor along with a brief description of each event.
The main profiling tool. The
operftool uses the Linux Performance Events subsystem, which allows OProfile to operate alongside other tools using performance monitoring hardware of your system.
Unlike the previously used
opcontroltool, no initial setup is required, and it can be used without root privileges unless the
--system-wideoption is used.
- A tool for counting the absolute number of event occurrences. It can count events on the whole system, per process, per CPU, or per thread.
- Converts sample database files from a foreign binary format to the native format for the system. Only use this option when analyzing a sample database from a different architecture.
- Creates an annotated source for an executable if the application was compiled with debugging symbols.
- Reads the recorded performance data and generates a summary as specified by the profile specification. It is possible to generate different reports from the same profile data using different profile specifications.
23.1. Using OProfile
operf is the recommended tool for collecting profiling data. The tool does not require any initial configuration, and all options are passed to it on the command line. Unlike the legacy
operf can run without
root privileges. See the Using operf chapter in the System Administrator’s Guide for detailed instructions on how to use the
Example 23.1. Using ocount
The following example shows counting the amount of events with
ocount during execution of the
$ ocount -e INST_RETIRED -- sleep 1 Events were actively counted for 1.0 seconds. Event counts (actual) for /bin/sleep: Event Count % time counted INST_RETIRED 683,011 100.00
The events are processor implementation specific. It might be necessary to set the option
perf_event_paranoid or limit the counts to only user-space events.
Example 23.2. Basic operf usage
In the following example, the
operf tool is used to collect profiling data from the
ls -l ~ command.
Install debugging information for the
# debuginfo-install -y coreutils
Run the profiling:
$ operf ls -l ~ Profiling done.
Analyze the collected data:
$ opreport --symbols CPU: Intel Skylake microarchitecture, speed 3.4e+06 MHz (estimated) Counted cpu_clk_unhalted events () with a unit mask of 0x00 (Core cycles when at least one thread on the physical core is not in halt state) count 100000 samples % image name symbol name 161 81.3131 no-vmlinux /no-vmlinux 3 1.5152 libc-2.17.so get_next_seq 3 1.5152 libc-2.17.so strcoll_l 2 1.0101 ld-2.17.so _dl_fixup 2 1.0101 ld-2.17.so _dl_lookup_symbol_x [...]
Example 23.3. Using operf to Profile a Java Program
In the following example, the
operf tool is used to collect profiling data from a Java (JIT) program, and the
opreport tool is then used to output per-symbol data.
Install the demonstration Java program used in this example. It is a part of the java-1.8.0-openjdk-demo package, which is included in the Optional channel. See Adding the Optional and Supplementary Repositories for instructions on how to use the Optional channel. When the Optional channel is enabled, install the package:
# yum install java-1.8.0-openjdk-demo
Install the oprofile-jit package for OProfile to be able to collect profiling data from Java programs:
# yum install oprofile-jit
Create a directory for OProfile data:
$ mkdir ~/oprofile_data
Change into the directory with the demonstration program:
Start the profiling:
$ operf -d ~/oprofile_data appletviewer \ -J"-agentpath:/usr/lib64/oprofile/libjvmti_oprofile.so" example2.html
Change into the home directory and analyze the collected data:
opreport --symbols --threshold 0.5
A sample output may look like the following:
$ opreport --symbols --threshold 0.5 Using /home/rkratky/oprofile_data/samples/ for samples directory. CPU: Intel Ivy Bridge microarchitecture, speed 3600 MHz (estimated) Counted CPU_CLK_UNHALTED events (Clock cycles when not halted) with a unit mask of 0x00 (No unit mask) count 100000 samples % image name symbol name 14270 57.1257 libjvm.so /usr/lib/jvm/java-1.8.0-openjdk-126.96.36.199-1.b16.el7_1.x86_64/jre/lib/amd64/server/libjvm.so 3537 14.1593 23719.jo Interpreter 690 2.7622 libc-2.17.so fgetc 581 2.3259 libX11.so.6.3.0 /usr/lib64/libX11.so.6.3.0 364 1.4572 libpthread-2.17.so pthread_getspecific 130 0.5204 libfreetype.so.6.10.0 /usr/lib64/libfreetype.so.6.10.0 128 0.5124 libc-2.17.so __memset_sse2
23.2. OProfile Documentation
For more extensive information on OProfile, see the oprofile(1) manual page. Red Hat Enterprise Linux also provides two comprehensive guides to OProfile in
- OProfile Manual
A comprehensive manual with detailed instructions on the setup and use of OProfile is available at
- OProfile Internals
Documentation on the internal workings of OProfile, useful for programmers interested in contributing to the OProfile upstream, can be found at