How do kernel arguments work?

[cgwalters]

(Writing this for someone else)

[cgwalters]

The important thing to understand is that today, bootc is backed by ostree, and in general most things work how they work with ostree. Today, kernel arguments with ostree are part of "BLS type 1" config files in /boot/loader that any tool can edit - they are not explicitly managed state. rpm-ostree offers a high level rpm-ostree kargs CLI.

Typically, kernel arguments are injected at the time the "deployment" is made; this can be done as part of making a disk image (e.g. via image builder or Anaconda). A new thing with bootc is support for bootc install which offers a --karg argument that allows injecting kernel arguments at the time of installation.

(Now, some people making derived operating systems may just bake kernel arguments they want into the kernel binary; this makes a ton of sense when making "sealed" systems; UKI style. That is obviously supported. However a tricky thing here is that one can't change the kernel binary from a distribution kernel without breaking Secure Boot)

In OpenShift we ended up creating a CRD MachineConfig that allows setting kernel arguments - a crucial aspect of this is that it also implements "day 2" management; if you delete the CRD, the kargs go away. In other words, they become managed state.

The way I'd like to do this in bootc is support kernel arguments also via #22 - this would be a generalization of the MachineConfig.

[cgwalters]

Forgot to mention I'd also really like to make

FROM fedora-boot
RUN grubby --add nosmt

also work (which might just end up writing a file into /usr/lib/bootc/kargs.d or so), going to container image as source of truth.