Re: Some more debug / behaviour from the perfmon problems

[Stephane E. <er...@hp...>]

John,

I spent some time trying to reproduce your problem here.
So I downloaded the latest Oprofile from your CVS tree. 
Then I hacked the Makefile in daemon to use opd_perfmon.c.
I managed to get a binary for the daemon.

I installed it on my 2-way Itanium2 machine running Debian.
I am using 2.6.0-test7 as my kernel base in which I have incorporated
the your oprofile patch.

I launched the daemon, then start measuring:
	# opcontrol --start-daemon
	# opcontrol --start
	wait 10 seconds
	# opcontrol --stop
	# cat /proc/perfmon	

perfmon version           : 2.0
model                     : Itanium 2
fastctxsw                 : No
ovfl_mask                 : 0x7fffffffffff
CPU0  overflow intrs      : 99764
CPU0  overflow cycles     : 51595015
CPU0  overflow min        : 489
CPU0  overflow max        : 6154
CPU0  smpl handler calls  : 99764
CPU0  smpl handler cycles : 11399161
CPU0  spurious intrs      : 0
CPU0  replay   intrs      : 0
CPU0  syst_wide           : 1
CPU0  dcr_pp              : 0
CPU0  exclude idle        : 0
CPU0  owner               : 32628
CPU0  context             : e00000405ebf6000
CPU0  activations         : 0
CPU1  overflow intrs      : 99776
CPU1  overflow cycles     : 50214102
CPU1  overflow min        : 477
CPU1  overflow max        : 4504
CPU1  smpl handler calls  : 99776
CPU1  smpl handler cycles : 10650890
CPU1  spurious intrs      : 0
CPU1  replay   intrs      : 0
CPU1  syst_wide           : 1
CPU1  dcr_pp              : 0
CPU1  exclude idle        : 0
CPU1  owner               : 32629
CPU1  context             : e00000405e942000
...

It is showing the two CPUs have received interrupts and more or less
the same count.

Next I tried the following:
#  opcontrol --start-deamon; opcontrol --start; sleep 5; opcontrol --stop && sleep 5 && opcontrol --shutdown

And both CPU would stop cleanly.  I tried several times and nothing bad happened.

It may be something different in your kernel. Could you try by downloading
David's kernel instead of Linus's kernel? Alternatively I could give you
the test kernel I am using now.

--
-Stephane