[coLinux-users] Re: using 2.6 kernels, moving to latest coLinux snapshot, etc.

[Joe W. <sl...@bl...>]

Okay, here is the story of me moving from coLinux 0.6.1 to a recent
coLinux snapshot.  There is a bug reported at the end of the story
where Firefox is dying with a floating point exception and I'm
wondering if it is due to something wrong in coLinux.

This is what I did:

* prepared for kernel 2.6 modules:
    emerge module-init-tools
      (This seems to be the kernel 2.6 replacement for the modutils
      package which is for kernel 2.4.  The new Gentoo package
      includes the old modutils so that if module-init-tools is
      installed, then both 2.4 and 2.6 kernels can be used.)

* prepared for coLinux snapshot's change in the location of the cobd
  devices (only needed if using devfs (which Gentoo does)):
  * created new cobd names:
      mkdir /dev/cobd
      for i in `seq 0 31`; do
          mknod /dev/cobd/$i b 117 $i
          chmod u=rw,g=r,o= /dev/cobd/$i 
      done
  * changed /dev/cobd{0,1} to /dev/cobd/{0,1} in /etc/fstab

* installed the new TAP driver from the 2004-07-10 coLinux snapshot
  (not sure if this is needed, I actually did this later when trying
  to get the TAP working):
    ./coLinux-20040710.exe
    Next -> "I agree" -> unselect(coLinux, coLinux Bridged Ethernet,
    Root Filesystem image Download) -> Next -> "C:\Program
    Files\coLinux\snapshot-2004-07-10" -> Next -> Next -> "Continue 
    Anyway" -> Finish

* unpack 2004-07-19 monotone build from http://firechief.dyndns.org/colinux/:
    mkdir '/c/Program Files/coLinux/0.6.2-gbp-2004-07-19'
    cp coLinux-0.6.2-gpb.tar.bz2 '/c/Program Files/coLinux/0.6.2-gbp-2004-07-19'
    cd '/c/Program Files/coLinux/0.6.2-gbp-2004-07-19'
    tar xfj coLinux-0.6.2-gpb.tar.bz2

* stop old coLinux

* switch to the new coLinux driver:
    cd ..
    ./colinux-daemon --remove-driver
    cd 0.6.2-gbp-2004-07-19
    ./colinux-daemon --install-driver

* fix the configuration file:
    cp ../gentoo-i686-ext3-4g.colinux.xml .
    emacs gentoo-i686-ext3-4g.colinux.xml:
      change "root=/dev/cobd0" to "root=/dev/cobd/0"
      added 'name="Local Area Connection 3"' to the network setting
        (actually, I only added the 'name' attribute later when I had
        problems with the TAP not being found)

* check that it boots okay:
    ./colinux-daemon -c gentoo-i686-ext3-4g.colinux.xml

* the first time I did this it didn't find my TAP device and I went
  back and did things as described above.

* did lots of screwing around with Windows to get ICF/ICS to work
  with "Local Area Connection 3" instead of "Local Area Connection
  2" as I had it set before.  painful.

* unpack and test the new kernel modules:
    smbmount //MYMACHINE/C$ /c -o username=jbw,uid=jbw,gid=users
    cd /
    tar xvfz '/c/Program Files/coLinux/0.6.2-gbp-2004-07-19/vmlinux-modules.tar.gz'
    modprobe ip_tables

* check that Linux kernel 2.6 detects my CPU correctly:
    cat /proc/cpuinfo

* observe that Firefox dies with a floating point exception:
    /usr/bin/firefox: line 87:  3285 Floating point exception$mozbin "$@"

The last point is the main annoyance.  Firefox dies with a repeatable
floating point exception.  I have run it like this to get a trace:

----------------------------------------------------------------------
#!/bin/sh
export MOZILLA_FIVE_HOME=/usr/lib/MozillaFirefox
export LD_LIBRARY_PATH=/usr/lib/MozillaFirefox:/usr/lib/MozillaFirefox/plugins
strace -o firefox-death.trace /usr/lib/MozillaFirefox/firefox-bin
----------------------------------------------------------------------

I have put the trace output at

  http://www.macs.hw.ac.uk/~jbw/firefox-death.trace

in case it can help anyone figure this out.

What could have changed to cause floating point exceptions?  I checked
that my configuration files could not be causing the problem by
temporarily moving them out of the way.  Is this a Linux kernel 2.6
change or a coLinux change?  Any thoughts or suggestions?  What
information is needed to debug this?

-- 
Joe