How to get xen whole-machine dump image and analyse it prepair

enable kexec

• apply kexec patches to xen source tree. latest patches are available in xen-devel list. (It maybe merged into xen 3.0.3)

install dom0cut

• prerequirement: libelf, python, binutils

  get dom0cut script and libelf wrapper from http://people.valinux.co.jp/~moriwaka/dumpread/ install libelf wrapper

   $ wget http://people.valinux.co.jp/~moriwaka/dumpread/libelf-0.3.0.tar.gz
   $ tar zxf libelf-0.3.0.tar.gz 
   $ cd libelf-0.3.0
   $ sudo python ./setup.py install
  

  install dom0cut script

   $ wget http://people.valinux.co.jp/~moriwaka/dumpread/dumpread-20060927.tar.gz
   $ tar zxvf dumpread-20060927.tar.gz
  

prepair xen-syms, vmlinux

• copy xen executable binary with symbols (when you compile, it is made at xen/xen-syms) and vmlinux used for domain 0

get coredump

load kdump kernel

1. append followings to xen boot args

• kdump_megabytes=64 kdump_megabytes_base=32
  

2. on dom0, load kdump kernel, take a dump.

• This is same as linux kdump usage. http://lse.sourceforge.net/kdump/

get a dump image

•  echo c > /proc/sysrq-trigger
  

after reboot, copy /proc/vmcore to core file.

• cp /proc/vmcore /vmcore
  

analysis

Now dom0cut script eats large memory(about 100-200MB). You should reboot machine before analysis.

extract domain 0 image

• $ cd dumpread
  $ ./dom0cut_x86.py -d/vmcore --target=0 -odom0image -xxen-syms -txen
  

you maybe need LD_PRELOAD=/usr/local/lib/libelf.so to resolve conflict for distribution's libelf.(same-name different library is contained in Fedora Core or Redhat)

use crash

• $ crash vmlinux dom0image 
  

extract xen hypervisor image

• $ cd dumpread
  $ ./dom0cut_x86.py -d/vmcore --target=x -oxenimage -xxen-syms -telf
  

use gdb

$ gdb xen-syms xenimage