Getting started with oprofile on Ubuntu

Oprofile is a powerful system wide profiler for Linux.  It can profile all running code on a system with minimal overhead.   Running oprofile requires the uncompressed vmlinux image, so one has to also install the kernel .ddeb images.

To install oprofile:

 sudo apt-get update && sudo apt-get install oprofile

..and then install the kernel .ddebs:

 echo "deb http://ddebs.ubuntu.com $(lsb_release -cs) main restricted universe multiverse" | \  
 sudo tee -a /etc/apt/sources.list.d/ddebs.list  
 sudo apt-key adv --keyserver keyserver.ubuntu.com --recv-keys 428D7C01  
 sudo apt-get update  
 sudo apt-get install linux-image-$(uname -r)-dbgsym  

 ..the installed vmlinux image can be found in /usr/lib/debug/boot/vmlinux-$(uname-r)

Oprofile is now ready to be used.  Let's assume we want to profile the following command:

 dd if=/dev/urandom of=/dev/null bs=4K  

First, before running opcontrol, one may have to stop the NMI watchdog to free up counter 0 using the following:

 echo "0" | sudo tee /proc/sys/kernel/watchdog  

Next, we tell opcontrol the location of vmlinux, separate out kernel samples, initialize, reset profiling and start profiling:

 sudo opcontrol --vmlinux=/usr/lib/debug/boot/vmlinux-$(uname -r)  
 sudo opcontrol --separate=kernel  
 sudo opcontrol --init  
 sudo opcontrol --reset  
 sudo opcontrol --start  

 ..and run the command we want to profile for the desired duration. Next we stop profiling, generate a report for the executable we are interested in and de-initialize oprofile using:

 sudo opcontrol --stop  
 sudo opreport image:/bin/dd -gl  
 sudo opcontrol --deinit  

The resulting output from opreport is as follows:

 Using /var/lib/oprofile/samples/ for samples directory.  
 warning: /kvm could not be found.  
 CPU: Intel Ivy Bridge microarchitecture, speed 2.501e+06 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  
 55868  59.8973 vmlinux-3.9.0-0-generic sha_transform  
 14942  16.0196 vmlinux-3.9.0-0-generic random_poll  
 10971  11.7622 vmlinux-3.9.0-0-generic ftrace_define_fields_random__mix_pool_bytes  
 3977   4.2638 vmlinux-3.9.0-0-generic extract_buf  
 1905   2.0424 vmlinux-3.9.0-0-generic __mix_pool_bytes  
 1596   1.7111 vmlinux-3.9.0-0-generic _mix_pool_bytes  
 900    0.9649 vmlinux-3.9.0-0-generic __ticket_spin_lock  
 737    0.7902 vmlinux-3.9.0-0-generic copy_user_enhanced_fast_string  
 574    0.6154 vmlinux-3.9.0-0-generic perf_trace_random__extract_entropy  
 419    0.4492 vmlinux-3.9.0-0-generic extract_entropy_user  
 336    0.3602 vmlinux-3.9.0-0-generic random_fasync  
 146    0.1565 vmlinux-3.9.0-0-generic sha_init  
 133    0.1426 vmlinux-3.9.0-0-generic wait_for_completion  
 129    0.1383 vmlinux-3.9.0-0-generic __ticket_spin_unlock  
 72     0.0772 vmlinux-3.9.0-0-generic default_spin_lock_flags  
 69     0.0740 vmlinux-3.9.0-0-generic _copy_to_user  
 35     0.0375 dd            /bin/dd  
 23     0.0247 vmlinux-3.9.0-0-generic __srcu_read_lock  
 22     0.0236 vmlinux-3.9.0-0-generic account  
 15     0.0161 vmlinux-3.9.0-0-generic fsnotify
...  

This example just scratches the surface of the capabilities of oprofile. For further reading I recommend reading the oprofile manual as it contains some excellent examples.