How can I enable kernel crash dumps in Debian?

Question

I want to enable kernel crash dumps for my Debian 5.0.7 machine. The kernel version is 2.2.26 amd64.

How can I configure this?

Solution 1

You may also want to look at Debian's kdump-tools package to automate some of the necessary boot-time steps.

The steps are roughly,

1. sudo apt-get install kdump-tools
2. Set USE_KDUMP=1 in /etc/default/kdump-tools
3. Add crashkernel=128M to the kernel command-line given in bootloader configuration (e.g. /etc/default/grub). It also doesn't hurt to pass nmi_watchdog=1 as well to ensure that hard hangs are caught.
   • Note that 128MB is merely a ballpark figure. It needs to be large enough to accomodate the kernel image and the associated init ramdisk.
   • If your initram disk is large, you might be able to shrink it by tweaking /etc/initramfs-tools/initramfs.conf
4. Ensure that your boot loader configuration is updated (e.g. sudo update-grub)
5. Ensure your kernel is built with,
   • CONFIG_RELOCATABLE=y
   • CONFIG_KEXEC=y
   • CONFIG_CRASH_DUMP=y
   • CONFIG_DEBUG_INFO=y
6. Reboot
7. Verify that the crash kernel is loaded, cat /sys/kernel/kexec_crash_loaded
8. Optional: Test that all of this worked,
   1. sudo sync; echo c | sudo tee /proc/sysrq-trigger
   2. Use the crash tool to look at the resulting crash dump
9. Find a handle of good whiskey to ease the pain of your future in kernel debugging.

Solution 2

A short answer, but...

Go to your kernel source (E.g. cd /usr/src/linux/ ) and configure the options for the next kernel (make menuconfig). Go to "Processor type and features". Enable "kernel crash dumps". (CONFIG_CRASH_DUMP=y)

Build new kernel, install.

Then read these for more background information: Linux-Crash-HOWTO.pdf and lkcd utils

Comments

[Matthew Flaschen]

And to check if it's already enabled: grep CONFIG_CRASH /boot/config-$(uname -r)

[clearlight]

Upvote for the whisky reference. (And everything else).

[Michael Firth]

It looks like manually adding "crashkernel" to the Grub configuration is no longer needed (at least in Debian 10). On doing so, I got a resulting commandline ending "crashkernel=256M quiet crashkernel=384M-:128M" - the second crashkernel parameter was added by something automatically