[arch-dev-public] mkinitcpio some improvements and question on how kernel modules should be treated

[Anatol Pomozov]

Hi

On Tue, Feb 1, 2022 at 6:54 AM Tobias Powalowski via arch-dev-public <
arch-dev-public at lists.archlinux.org> wrote:

> Hi guys,
> did some initcpio optimizing lately and implemented correct symlink
> handling. This saves already some space.
> https://bugs.archlinux.org/task/73439
>
> Now one major problem left in recent initcpio, uncompressed modules:
> https://github.com/archlinux/mkinitcpio/pull/25
> https://bugs.archlinux.org/task/72882
> At the moment modules are uncompressed during initcpio creation. This leads
> to bigger RAM usage during boot, makes it a bit smaller on storage, due to
> not recompress compressed modules. On small initrds for normal boot this is
> not a big matter.
>
> My measurement with all patches added:
> Ramdisk size uncompressed modules: 13261677
> Ramdisk size compressed modules: 13112083
> (Note the compressed one is smaller due to the symlink patches added)
> Unpacked initramfs uncompressed modules: 35 MB
> Unpacked initramfs compressed modules:  26 MB
>

This memory usage difference does not make much difference for the boot
experience.

Keep in mind that it is initramfs stage and all the memory used here
(except a tiny part) is reclaimed/freed at the end of initramfs. So this
memory difference is just a blip at the first 100ms of the boot process
when no OS/services are loaded and no memory pressure exists.

The drawback of this solution is that kernel modules are double-compressed
- the first time with initramfs and then module compression itself. And
decompressing requires extra time that is probably more valuable at
initramfs stage. What is the reported systemd-analyze time for compressed
vs non-compressed versions?


> Using this on the biggest "initramfs testsuite" archboot shows the most
> advantages.
>
> If you like to test it on your own initramfs setup you can use:
>
> https://gitlab.archlinux.org/tpowa/archboot/-/blob/master/usr/share/archboot/patches/31-initcpio.functions.fixed
> for /usr/lib/initcpio/functions and recreate your initramdisk.
>
> Now my question to the list, what should be the default how arch handles
> modules in initramfs?
> - Shall we leave the modules at compression level used during kernel build?
>  --> This means the uncompression should be optional.
> - Shall we still uncompress the modules always? (Status now)
> --> Then using modules as is needs to be optional.
>
> Thanks for you input.
> greetings
> tpowa
> --
> Tobias Powalowski
> Archlinux Developer & Package Maintainer (tpowa)
> https://www.archlinux.org <http://www.archlinux.org>
> tpowa at archlinux.org
>
> St. Martin-Apotheke
> Herzog-Georg-Str. 25
> 89415 Lauingen
> https://www.st-martin-apo.de
> info at st-martin-apo.de
>