Create Linux kernel configuration script

[jodh-intel]

As @sameo noted in kata-containers/linux#5 (comment), we should have a script that can generate a valid kernel configuration file.

The script will generate a valid .config kernel configuration file to be used by the kernel's build system.

However, the script will allow individual options and groups of options to be documented as the kernel's config file does not support comments (well, it does, but only for "commented out" / disabled CONFIG_ options).

Each group of config options in the script should document:

• Why it has been enabled.
• If there is any known impact (positive or negative) on boot speed / general performance.
• If there is any known impact (positive or negative) on memory density.

See also: #7.

[jcvenegas]

@jodh-intel do you think is needed to do using Kconfig tools? Or you think is better an script that takes an exithing config and checks for the minimal set of configs ?

[jodh-intel]

I was imagining a script that would do something similar to https://github.com/kata-containers/packaging/blob/master/scripts/configure-hypervisor.sh. The script would read a YAML config file and would create a (partial) .config file. You could then run make olddefconfig so that Kconfig would add in any missing CONFIG_* options and (hopefully?) generate a usable kernel.

The config file would be something like the following (in pseudo-YAML):

config:
    CONFIG_FOO:
      type: [module|built-in|disabled]
      reasons: [boot-speed|memory-density|security]
      architectures: all
      notes: free-form multi-line text to describe the option goes here.

  CONFIG_BAR:
      ...

  CONFIG_BAZ:
      ...

The config file would not list all config options - it would only list those that we have to set or explicitly disable. However, I don't know whether such a plan would even work.

Any thoughts @grahamwhaley, @sboeuf, @bergwolf?

[sboeuf]

@jodh-intel I think it works technically as long as you apply the defaults for the rest of the config. But this partial .config will need to be explicit enough to get a kernel that boots eventually.
It sounds interesting to me, because this would help users of Kata to get their own kernel with the minimal config expected by Kata. This way they don't need to do the translation of what is in our specific .config for our current vmlinuz.

[jodh-intel]

Related: #56.

[grahamwhaley]

Right, let's start trying to add a little 'meat' to the whole config fragment stuff. I didn't find too much in any other Issue.

We have discussed seeing if we can use the 'config merge' script before. that lives in the linux kernel sources here afaict.
And then there is an interesting (recent) presentation on the ability for the kernel to merge fragments here that briefly shows how the kernel makefile can take multiple config args, and how to invoke merge_config.sh.

I'm going to have a little play with merge_config.sh and either the allnoconfig or tinyconfig. I suspect allnoconfig is the best base choice, as iirc tinyconfig may be an x86 only thing.

[egernst]

Stale -- I think we have forward progress now on this wrt config fragments. Please re-open if you disagree @jodh-intel

[devimc]

closing issue since kernel fragments support already landed