Introduction

The Ubuntu Kernel Crash Dump is a mechanism that enable enterprise style post-mortem crash analysis in Linux operating systems. It uses a special mode of kexec which allows to automatically boot a secondary kernel whenever a crash (Oops/panic) occurs. This secondary kernel will then save the state and memory of the primary kernel to a certain location of the filesystem (/var/crash on newer releases). This file can then be used by crash to gather detailed information about the problem.

Installation

For convenience, the kernel crash dump utility has been packaged in Ubuntu. It can be installed with the following command:

sudo apt-get install linux-crashdump 

Newer versions of the package will automatically add an entry crashkernel=384M-2G:64M,2G-:128M to the kernel commandline in grub. However this may cause problems on systems with less than 2G of memory (see troubleshooting).

ppc64el installation

For those running ppc64el please follow the crash kernel recommendations found on the ppc64el Recommendations page.

Verifying linux-crashdump installation

How to verify the linux-crashdump installation was once documented for Ubuntu 14.04. This may or may not apply to later releases.

Inspecting the crash dump using crash

In order to use the generated crash dump with crash one needs the vmlinux file which has the debugging information. This is part of the kernel ddeb package which can be found at:

http://ddebs.ubuntu.com/pool/main/l/linux/

sudo tee /etc/apt/sources.list.d/ddebs.list << EOF
deb http://ddebs.ubuntu.com/ $(lsb_release -cs)          main restricted universe multiverse
deb http://ddebs.ubuntu.com/ $(lsb_release -cs)-security main restricted universe multiverse
deb http://ddebs.ubuntu.com/ $(lsb_release -cs)-updates  main restricted universe multiverse
deb http://ddebs.ubuntu.com/ $(lsb_release -cs)-proposed main restricted universe multiverse
EOF

sudo apt-key adv --keyserver keyserver.ubuntu.com --recv-keys ECDCAD72428D7C01
sudo apt-get update
sudo apt-get install linux-image-$(uname -r)-dbgsym

Be aware that those packages are huge! (~600 MB)

When installed, the debug kernel can be found under /usr/lib/debug/boot/ and crash is started by:

crash <debug kernel> <crash dump> 

Unfortunately the tool does not allow to look at a 32bit dump on a 64bit system and the other way round. Also it tends to be quite picky about matching up kernel and dump.

Inspecting the crash dump using apport-retrace

To get a local retrace, you need apport-retrace and then run:

apport-retrace --stdout --rebuild-package-info /var/crash/linux-image*.crash 

Again, this can take a while because it needs to download the kernel debug package.

Enabling various types of panics

To make Linux kernel to panic on different situations please use:

echo 1 > /proc/sys/kernel/hung_task_panic          # panic when hung task is detected
echo 1 > /proc/sys/kernel/panic_on_io_nmi          # panic on NMIs from I/O
echo 1 > /proc/sys/kernel/panic_on_oops            # panic on oops or kernel bug detection
echo 1 > /proc/sys/kernel/panic_on_unrecovered_nmi # panic on NMIs from memory or unknown
echo 1 > /proc/sys/kernel/softlockup_panic         # panic when soft lockups are detected
echo 1 > /proc/sys/vm/panic_on_oom                 # panic when out-of-memory happens

Troubleshooting

Allocated memory for the crash kernel

When testing crash dump sometimes the system just seems to lock up. The main issue there is how much memory was assigned for the crash kernel. When kexec starts the crash kernel it requires enough memory to fit the unpacked kernel, the compressed initrd and the uncompressed initrd (at least while unpacking). If there is not enough memory allocated, things usually go wrong without any hint. To solve this there are the following options:

1. Increase the allocation by changing crashkernel= on the grub command line or in /boot/grub/grub.cfg (for grub2) or /boot/grub/menu.lst (for old grub). To avoid loosing the settings when running update-grub the change can be made in /etc/default/grub.d/kexec-tools.cfg.

2. Reduce the size of the initrd. By default this is set to include all the modules and firmware ever needed. This allows using the same initrd on any system but increases its size a lot. In order to limit it to the modules really required to boot on the current hardware, change the following in /etc/initramfs-tools/initramfs.conf:

    ...
    MODULES=dep
    ... 

Crash kernel fails to load: Hang

This can be frustrating to debug, especially if you're unable to record the console messages from the new kexec kernel. A serial console attached to the system is best here to continue debugging. An easy troubleshooting step is to systematically eliminate the additional kernel parameters passed to the crash kernel and retrying. These arguments are kept in /etc/init.d/kdump:

...
        # Append kdump_needed for initramfs to know what to do, and add
        # maxcpus=1 to keep things sane.
        APPEND="$APPEND kdump_needed maxcpus=1 irqpoll reset_devices"

        # --elf32-core-headers is needed for 32-bit systems (ok
        # for 64-bit ones too).
        log_action_begin_msg "Loading crashkernel"
        kexec -p "$KERNEL_IMAGE" --initrd="$INITRD" --append="$APPEND"
        log_action_end_msg $?
... 

Leave $APPEND and kdump_needed. Start by removing reset_devices and then install the new kexec crash kernel configuration:

sudo service kdump start 

Then retest; if that doesn't work, remove the next argument, rinse and repeat.

ACPI memory hotplug issues

If you see the following call trace from your serial console after kexecing into the crash kernel you may need to append 'acpi_no_memhotplug' to the crashdump kernel cmdline.

Call Trace:
 dump_stack+0x45/0x57
 warn_alloc_failed+0xf2/0x140
 __alloc_pages_nodemask+0x2e4/0xa10
 vmemmap_alloc_block+0xb5/0xbf
 vmemmap_alloc_block_buf+0x15/0x3b
 vmemmap_populate+0xb3/0x20c
 sparse_mem_map_populate+0x29/0x38
 sparse_add_one_section+0x71/0x16e
 __add_pages+0xb9/0x280
 arch_add_memory+0x71/0xf0
 add_memory+0xdf/0x210
 acpi_memory_device_add+0x1ab/0x282
 acpi_bus_attach+0xe3/0x196
 acpi_bus_scan+0x70/0x8f
 acpi_scan_init+0x89/0x1d3
 acpi_init+0x272/0x28f
 do_one_initcall+0xb3/0x200
 kernel_init_freeable+0x17b/0x21a
 kernel_init+0xe/0xe0
 ret_from_fork+0x3f/0x70

Edit KDUMP_CMDLINE_APPEND in /etc/default/kdump-tools such that it is un-commented and contains 'acpi_no_memhotplug' as well. Then restart the kdump service.

Release specific notes

Ubuntu 12.04 "Precise Pangolin"

• Bug 785394: Hard-coded crashkernel=... memory reservation in /etc/grub.d/10_linux is insufficient
  The default allocation for systems below 2G is not enough for the current initrd size. Manually adapting the size allows to use the crash kernel.

• The current (1.3.7-2) version of makedumpfile reports to be incompatible with the 3.2 kernel. The dumps created seem to be ok.

Ubuntu 15.10 "Wily Werewolf" and later

• Bug 1496317: kexec fails with OOM killer with the current crashkernel=128 value
  

The current allocation for the crashkernel value is too low to correctly load the default initrd.img. This means that the OOM killer will break the crash dump capture procedure. While the bug is being worked on, you can increase the value of crashkernel to something more than 150Mb to work around the bug.