Kernel debugging essentials workshop

Topics

We'll be holding a workshop on basics of finding and debugging (and eventually fixing) common issues with Linux kernel.

Topics I'd initially cover would include leveraging in-kernel debugging facilities, techniques for capturing debug output from various crashes, understanding the captured data, and modifying, building and loading the altered kernel code with more debugging code or an actual fix.

Original mailing list reference: [1]

Requirements

Some elementary knowledge of C
A laptop with Linux (Fedora 20 preferred, Debian 7 fine, we'll try to cover older versions and other distribution too) and root access, 2G + size of RAM free space
- If you don't have any available, please let me know in advance

Language

I'm fine with holding the workshop in English language if there's anyone who doesn't speak Czech/Slovak.

Setup instructions

Please prepare your laptop in advance so that we don't waste too much time at the workshop.

Extra hardware will be available for ones who won't be able to bring their own laptops.

Install required packages

Debian Fedora

apt-get -y install kdump-tools crash git linux-headers linux-image-$(uname -r)-dbg linux-source ncurses-dbg

pkcon install kernel{-debug,}{-devel,} kexec-tools crash git gcc ncurses-devel

debuginfo-install -y kernel{-debug,}

Get the kernel source tree

git clone git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git

Cheatsheet

Here are some distro-dependent details. I'll be using Fedora, follow these to translate to Debian equivalents if you're running Debian.

These are here just for reference, no need to follow these prior to the workshop.

Enable kdump

Debian Fedora

Edit /etc/default/kdump-tools
Add crashkernel=256M to /etc/default/grub
Run update-grub2
Reboot
Check kdump-config status

system-config-kdump

or

Add crashkernel=256M to /etc/default/grub
Run grub2-mkconfig >/etc/grub.conf
systemctl enable kdump.service
Reboot
Check systemctl status kdump.service

Enable netconsole

Debian	Fedora
`modprobe netconsole \` `netconsole=@/,514@$(getent hosts odvarok.local \|` `awk '{print $1}')/`	Edit `/etc/sysconfig/netconsole` `systemctl enable netconsole.service` `systemctl start netconsole.service`

Running crash

The location of the raw vmlinux binary from the debugging packages differs in distros.

Debian	Fedora
`crash /usr/lib/debug/boot/vmlinux-$(uname -r) /var/crash/201402090936/dump.201402090936`	`crash /usr/lib/debug/lib/modules/$(uname -r)/vmlinux /var/crash/127.0.0.1-2013.10.24-02:48:40/vmcore`

Materials

Work-in-progress slides. Content will be extended and polished (as I learn beamer...), but outline will stay roughly the same: File:Kernel-workshop-draft.pdf.gz

https://github.com/lkundrak/sad-ugly-and-broken This is a repository for variously broken kernel code, for demonstration
https://github.com/lkundrak/kexec-tools Patched kexec-tools, for >= 3.11 kernels
http://v3.sk/~lkundrak/firescope/ Firewire debugging tool