From: <job...@ao...> - 2009-11-06 20:01:08
|
I must confess that I had previously altered check_version() to avoid a "Kernel support not available" error I was getting initially because my kernel should support oprofile. So, I went back to the original version and this is what happens when I try to start oprofile after rebooting my board: Eclipse # insmod oprofile.ko Eclipse # /usr/bin/opcontrol --init Kernel support not available, missing opcontrol --init as root ? /etc/mtab is empty. I am doing everything as root. I apologize for having bypassed this error. Should "oprofilefs" appear in /etc/mtab after inmod'ing oprofile.ko and doing opcontrol -- init? Sent from my iPhone 3G On Nov 6, 2009, at 10:19 AM, Maynard Johnson <may...@us...> wrote: > Robert Richter wrote: >> On 03.11.09 10:21:40, job...@ao... wrote: >>> I tried that and am still seeing the "bad magic number" error >>> when running opreport and "Invalid argument" entries in >>> oprofiled.log. >>> >>> >>> The commands I am using and part of oprofiled.log follow. Any >>> suggestions? >>> >>> Eclipse # rm /var/lib/oprofile/samples/oprofiled.log >>> Eclipse # rm -rf /var/lib/oprofile/samples/current/ >>> Eclipse # cd /usr/bin/ >>> Eclipse # ./opcontrol --init >>> mount: mounting nodev on /dev/oprofile failed: Device or resource >>> busy >> There is an error mounting oprofilefs, try the following steps to >> debug this: > You need to figure out why the init is failing. Any oprofile > commands you do afterwards are useless until you fix this. It > sounds to me like you still have an oprofile daemon running. Do 'ps > awx |grep oprofiled' to see if the daemon is running. If so, do: > 1. opcontrol --deinit > 2. rm /root/.oprofile/daemonrc > 3. opcontrol --init > 4. Execute your other opcontrol setup commands to specify profiling > parameters > 5. 'opcontrol --start' and run your app. > 6. generate your report(s) with opreport > > If the 'ps' command shows oprofiled is not running, you need to > determine why you're getting "Device or resource busy". Look at > dmesg. Are there any other perf tools on the system that might be > using the performance monitoring hardware? > > > -Maynard >> First, unmount /dev/oprofile if it is there. >> The script tries to mount the oprofile filesystem that the kernel >> should provide. It checks /proc/filesystems, you should get something >> like this on your system: >> # cat /proc/filesystems | grep oprofile >> nodev oprofilefs >> If it is not there, the script tries then to load the oprofile module >> and checks the filesystem again. oprofile should then be listed in >> /proc/modules and also oprofilefs in /proc/filesystems. If not, the >> script fails. Maybe /proc is disabled or not available? Not sure if >> this is a config option. >> You can also try to mount oprofile directly: >> # mkdir /dev/oprofile >> # mount -t oprofilefs nodev /dev/oprofile >> Also, it seems you are using an own build of oprofile, which version >> are you using. Did you use configure/make/make install to install the >> package on your system? Which config did you use to compile your >> kernel? >> -Robert > |