How to use kdump to debug kernel crashes

From FedoraProject

Jump to: navigation, search

Contents

Kernel and kdump

Kdump is a kernel crash dumping mechanism and is very reliable because the crash dump is captured from the context of a freshly booted kernel and not from the context of the crashed kernel. Kdump uses kexec to boot into a second kernel whenever system crashes. This second kernel, often called the crash kernel, boots with very little memory and captures the dump image.

The first kernel reserves a section of memory that the second kernel uses to boot. Kexec enables booting the capture kernel without going through the BIOS, so contents of the first kernel's memory are preserved, which is essentially the kernel crash dump.

How to Use Kdump

Step 1: Configuring Kdump

  1. First, install the kexec-tools, crash and kernel-debuginfo packages. Use following command line to install the packages.
    dnf install --enablerepo=fedora-debuginfo --enablerepo=updates-debuginfo kexec-tools crash kernel-debuginfo
    NOTE: The `crash` and `kernel-debuginfo` packages are only required if you are planning on looking at the resulting kernel vmcore yourself. Most often this is the case, however if you are setting up kdump on a machine simply to capture a vmcore that will be analyzed by someone else or on a different machine, you can skip those packages.
  2. Next, edit /boot/grub/grub.conf or /boot/grub2/grub.cfg and add the "crashkernel=128M" command line option. An example command line might look like this (for GRUB Legacy, "linux" is replaced by "kernel"):
    linux /vmlinuz-3.1.4-1.fc16.x86_64 ro root=/dev/VolGroup00/LogVol00 rhgb LANG=en_US.UTF-8 crashkernel=128M
  3. Next, consider editing the kdump configuration file /etc/kdump.conf. This will allow you to write the dump over the network or to some other location on the local system, rather than to the default location of /var/crash. For additional information, consult the mkdumprd man page and the comments in /etc/kdump.conf.
  4. Next, reboot your system
  5. Finally, active the kdump system service
    systemctl start kdump.service

Considerations:

  1. Above shown parameter reserves 128MB of physical memory. This reserved memory is used to preload and run the capture kernel.
  2. Init scripts take care of pre-loading the capture kernel at system boot time.
  3. It is recommended to either set up a serial console or switch to run level 3 (init 3) for testing purposes. The reason being that kdump does not reset the console if you are in X or framebuffer mode, and no message might be visible on console after system crash. You may also see screen corruption in graphics mode during capture.
  4. Capturing a crash dump can take a long time, especially if the system has a lot of memory. Be patient. The system will reboot after the dump is captured.

Step 2: Capturing the Dump

Normally kernel panic() will trigger booting into capture kernel but for testing purposes one can simulate the trigger in one of the following ways.

  1. Enable SysRq then trigger a panic through /proc interface
    echo 1 > /proc/sys/kernel/sysrq
    echo c > /proc/sysrq-trigger
  2. Trigger by inserting a module which calls panic().

The system will boot into the capture kernel. A kernel dump will be automatically saved in /var/crash/<dumpdir> and the system will boot back into the regular kernel. The name of the dump directory will depend on date and time of crash. For example, /var/crash/2006-02-17-17:02/vmcore.

Step 3: Dump Analysis

Once the system has returned from recovering the crash, you may wish to analyse the kernel dump file using the crash tool.

  1. First, locate the recent vmcore dump file:
    find /var/crash -type f -mtime -1
  2. One you have located a vmcore dump file, call crash:
    crash /var/crash/2009-07-17-10\:36/vmcore /usr/lib/debug/lib/modules/`uname -r`/vmlinux
Note.png
Missing debuginfo?
Cannot find any files under /usr/lib/debug? Make sure you have the kernel-debuginfo package installed.

For more information on using the crash tool, see #More Documentation.

On versions

The versions of kexec-tools and crash can be very reliant on the version of kernel running. On Fedora, from time-to-time the package versions can get out of sync and can lead to partially working crash dumps. This may manifest as warning messages from crash such as

page excluded: kernel virtual address: ffff.........9d28  type: "..."

If you want to know specifically what versions are supported, you can examine the srpm for the version of kexec-tools you are running, in particular makedumpfile.h will have something like

 #define OLDEST_VERSION          KERNEL_VERSION(2, 6, 15)/* linux-2.6.15 */
 #define LATEST_VERSION          KERNEL_VERSION(4, 5, 3)/* linux-4.5.3 */

If you run makedumpfile against an unsupported kernel version it will probably still mostly work. It will output an error message to the console, but it can be easy to miss in the kexec output.

If the dump is behaving unexpectedly you can modify kdump.conf to not filter any pages (perhaps except zero-filled pages with -d1) and only use it to compress the vmcore with -c. This might result in a more useful vmcore. If that fails, you could take makedumpfile out of the picture entirely by change the core_collector in kdump.conf to scp, which will simply copy /proc/vmcore to a permanent location.

If having further issues, you may also try building the latest crash tool from source. If you are at the point of debugging kernel crash dumps you can probably figure it out :) You might want to try something like:

$ sudo dnf builddep crash # quick way to get the right libraries
$ git clone https://github.com/crash-utility/crash.git
$ cd crash
$ make lzo # don't forget the lzo if you're using compressed dumps

More Documentation