Question

I’m trying to use oprofile to record cache misses in a large realtime app:

$ sudo opcontrol --no-vmlinux --event=LLC_MISSES:100000 --session-dir=/var/tmp/oprofile -c=5 --start

But when I look at the reports, it doesn’t mention the cache misses. It only samples CPU_CLK_UNHALTED:

$ sudo opreport -l --session-dir=/var/tmp/oprofile 

CPU: Intel Architectural Perfmon, speed 1596 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               app name                 symbol name
63243    92.2946  no-vmlinux               no-vmlinux               /no-vmlinux
564       0.8231  libc-2.13.so             libc-2.13.so             /lib32/libc-2.13.so
 (etc)

But –status claims that oprofile is sampling L2 misses:

$ sudo opcontrol --status

Daemon running: pid 3220
Event 0: LLC_MISSES:500000:65:1:1
Separate options: library
vmlinux file: none
Image filter: none
Call-graph depth: 5

What am I doing wrong? I can’t get it to sample any of the other counters listed in ophelp either.

This is with oprofile 0.9.6 on Ubuntu, kernel version 2.6.38.

Answer 1

Turns out you need to actually kill and restart the oprofile daemon with

sudo opcontrol --stop
sudo opcontrol --reset
sudo opcontrol --shutdown
sudo opcontrol --start-daemon
sudo opcontrol --start

when changing sampled events. Simply stopping and starting the profile isn’t enough. Not that this is documented anywhere.