BACK_END_BUBBLE bug on itanium 2?

[Bjarke B. M. <ro...@di...>]

Hi List,

I seem to have problems getting reasonable results (any at all,
really) out of the BACK_END_BUBBLE event on an itanium2 machine.

This is my sample run (the real setting is somewhat more complicated
but that doesn't seem to be the problem):

moose:/home/rodaz/mysql-files# opcontrol --start
--event=BACK_END_BUBBLE:10000:0:0:1 --separate=none --no-vmlinux
Using 2.4 OProfile kernel interface.
Using log file /var/lib/oprofile/oprofiled.log
Daemon started.
Profiler running.
moose:/home/rodaz/mysql-files# opcontrol --reset
Signalling daemon... done
moose:/home/rodaz/mysql-files# ls -R /tmp > /dev/null
moose:/home/rodaz/mysql-files# opcontrol --dump
moose:/home/rodaz/mysql-files# opcontrol --shutdown 
Killing daemon.
moose:/home/rodaz/mysql-files#  opreport  
opreport error: No sample file found: try running opcontrol --dump
or specify a session containing sample files

I can choose another event, such as BE_EXE_BUBBLE (which should always
count less than BACK_END_BUBBLE). Then I get:

moose:/home/rodaz/mysql-files# opcontrol --start
--event=BE_EXE_BUBBLE:10000:0:0:1 --separate=none --no-vmlinux
Using 2.4 OProfile kernel interface.
Using log file /var/lib/oprofile/oprofiled.log
Daemon started.
Profiler running.
moose:/home/rodaz/mysql-files# opcontrol --reset
Signalling daemon... done
moose:/home/rodaz/mysql-files# ls -R /tmp > /dev/null
moose:/home/rodaz/mysql-files# opcontrol --dump
moose:/home/rodaz/mysql-files#  opcontrol --shutdown 
Killing daemon.
moose:/home/rodaz/mysql-files# 
moose:/home/rodaz/mysql-files#  opreport  
CPU: Itanium 2, speed 900 MHz (estimated)
Counted BE_EXE_BUBBLE events (Full pipe bubbles in main pipe due to
Execution unit stalls) with a unit mask of 0x00 (ALL) count 10000
BE_EXE_BUBBLE:...|
  samples|      %|
------------------
     1151 42.5823 ld-2.3.2.so
      573 21.1987 libc-2.3.2.so
      498 18.4240 bash
      209  7.7321 insmod.modutils
      121  4.4765 no-vmlinux
       46  1.7018 emacs21-x
       30  1.1099 oprofiled
       18  0.6659 libcrypto.so.0.9.7
       12  0.4440 ISO8859-1.so
       12  0.4440 sshd
        9  0.3330 ls
        9  0.3330 gawk
        4  0.1480 grep
        3  0.1110 libX11.so.6.2
        3  0.1110 op_help
        2  0.0740 libXt.so.6.0
        1  0.0370 rm
        1  0.0370 libpthread-0.10.so
        1  0.0370 cut


This is as expected. 

I can also compare to a simple run of pfmon:
pfmon -e BACK_END_BUBBLE_ALL ls -R /tmp 
which gives me:
3692388 BACK_END_BUBBLE_ALL

so clearly BACK_END_BUBBLE produces some samples.

So why am I not seeing any samples for BACK_END_BUBBLE? I might be
doing something wrong, but the two runs of oprofile above are
identical except for the event monitored, so I can't realize my
mistake. Is is a bug in OProfile?

btw: I have also tried with a different unit mask for BACK_END_BUBBLE,
but it gives the same result.

btw2: 
<wild guess>
The event code for BACK_END_BUBBLE is 0x00
which seems to be correct according to the intel documentation, but
might it cause problems with the way oprofile sets up event
monitoring??
</wild guess>

Thanks for any input,
/Bjarke

-- 
Bjarke <ro...@di...>
http://www.diku.dk/hjemmesider/studerende/rodaz

And the pirates saw about a billion ninjas with guitars standing on
top this huge hill.  And they started to wail...