Configuring kdump to capture a RHEL kernel crash

Goal

Reserve crash-kernel memory and enable kdump so a kernel panic leaves a vmcore instead of just a reboot.

Prerequisites

Root access; enough free disk on the configured dump target to hold a full vmcore; a maintenance window if you intend to test it end to end.

Steps

1. Check whether crash-kernel memory is already reserved:

cat /sys/kernel/kexec_crash_size

A 0 means the reservation isn't active yet. 2. Reserve it via the crashkernel boot parameter, sized for the installed RAM, and apply it through the bootloader:

grubby --update-kernel=ALL --args="crashkernel=256M"

This is a kernel boot parameter, so it needs a reboot to take effect. On RHEL 9 and later, kdumpctl reset-crashkernel --kernel=ALL writes the distribution's recommended default instead of a hand-picked size. 3. After reboot, confirm the reservation:

cat /proc/iomem | grep -i crash

4. Enable and start the service:

systemctl enable --now kdump.service
kdumpctl status

5. Set the dump target and collector in /etc/kdump.conf, then rebuild the kdump initramfs so the new settings take effect:

kdumpctl restart

Expected result

After a real or deliberately triggered crash, a vmcore file appears under the configured dump path (by default under /var/crash/) on the next successful boot.

Limits and test basis

The kernel's own kdump documentation frames the reservation as what lets a second, minimal kernel boot inside memory set aside ahead of time; sizing it too small causes that second kernel to fail to boot, which silently defeats capture with no error on the running system. The only way to validate the whole path end to end is to deliberately crash the kernel, which reboots the host immediately and loses any in-flight work — do this only in a scheduled maintenance window on a host carrying no live traffic, never to "just check" on a production system. Undo the reservation with grubby --update-kernel=ALL --remove-args="crashkernel" and systemctl disable --now kdump.service, then reboot.