[BUG] CPU hotplug: system hang on CPU hot remove during oprofile

[BUG] CPU hotplug: system hang on CPU hot remove during oprofile

[Satoru Takeuchi]

Hi,

I found a bug on CPU hotplug. If oprofile is running, CPU hot remove causes
system hang. I only confirm that this problem occurs on my ia64 box. I'm glad
if someone report about other arch.

How to reproduce
================

1) start oprofile

  # opcontrol --start

2) offline a CPU.

  # echo 0 >/sys/devices/system/cpu/cpu1/online 

Expected Result
===============

Succeed to remove CPU

Actual Result
=============

See the following console log.

  # opcontrol --start
  Detected stale lock file. Removing.
  Using 2.6+ OProfile kernel interface.
  Reading module info.
  Running perfmon child on CPU0.
  Waiting on CPU0
  Perfmon child up on CPU0
  Running perfmon child on CPU1.
  Waiting on CPU1
  Perfmon child up on CPU1
  Running perfmon child on CPU2.
  Waiting on CPU2
  Perfmon child up on CPU2
  Running perfmon child on CPU3.
  Waiting on CPU3
  Perfmon child up on CPU3
  Running perfmon child on CPU4.
  Waiting on CPU4
  Perfmon child up on CPU4
  Running perfmon child on CPU5.
  Waiting on CPU5
  Perfmon child up on CPU5
  Running perfmon child on CPU6.
  Waiting on CPU6
  Perfmon child up on CPU6
  Running perfmon child on CPU7.
  Waiting on CPU7
  Perfmon child up on CPU7
  Using log file /var/lib/oprofile/oprofiled.log
  Daemon started.
  PFM_START on CPU1
  PFM_START on CPU4
  PFM_START on CPU0
  Profiler running.
  PFM_START on CPU3
  PFM_START on CPU5
  PFM_START on CPU2
  PFM_START on CPU7
  PFM_START on CPU6
  
  # echo 0 >/sys/devices/system/cpu/cpu1/online 
  
  <SYSTEM HANG UP here!>


Thanks,

Satoru

Re: [BUG] CPU hotplug: system hang on CPU hot remove during oprofile

[Russell King]

On Mon, May 28, 2007 at 01:53:30PM +0900, Satoru Takeuchi wrote:
> I found a bug on CPU hotplug. If oprofile is running, CPU hot remove causes
> system hang. I only confirm that this problem occurs on my ia64 box. I'm glad
> if someone report about other arch.

oprofile is not CPU hotplug safe.  (It should probably be made so,
but I think that would require a redesign, or at least a rewrite of
the backends.)

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of: