Speed up kernel loading time at boot.

[HystericalLich]

So I have been trying to speed up boot times (really more for show than any functional purpose) and I have hit a point where I can't seem to get it a lot lower. However, running systemd-analyze (I am a gnome user) I noted what feels to be a strangely long kernel load time:

Startup finished in 2.975s (firmware) + 296ms (loader) + 8.273s (kernel) + 3.110s (userspace) = 14.656s

Facts to know about this:
1) A lot of things are compiled into my kernel rather than added as modules. (Like the GPU/DRM driver)
2) I am using an SSD
3) This is a UEFI setup
4) I have previously gotten around 4 seconds with this drive in a slower computer with an arch linux install, given that was a lighter desktop environment (and lightdm vs gdm), I don't expect that here, but I feel like I should be able to get down to 10 seconds total.

I have a very simple setup, with a efi partition (vfat), my rootfs as ext4 with 'noatime' mount option, and a swap partition.
The kernel is stored in the efi partition, and I have heard that the fat filesystem can be rather slow. Is there a way to speed up fat?

[Keruskerfuerst]

You can add relatime to the mont options.
Rest of the setup?
Like Systemd
And emerge --info?

[HystericalLich]

So my system setup uses the default gnome/systemd profile. I use the open source radeon driver for my gpu (fglrx causes a lot of issues with gnome), and a lot of things, including it, are compiled in my kernel rather than included as modules. There isn't a compelling reason for this. The computer is an HP pavilion g6-2235 us with a 2.7 GHz AMD a-6 APU and 12GB ram. I run apparmor, I know that my main source of the userspace boot time component is a bash script that loads a couple of apparmor profiles and launches preload. (A lot of gentoo packages don't come with systemd compatible initscripts). I also have sys-apps/systemd-readahead-216 installed.

I will try the relatime option.

My fstab:

[Keruskerfuerst]

The content of /boot ?

In my opinion the vfat partition is not nessecary. It should be ext4.

[szatox]

For UEFI it is necessary, but you only need kernel and optional initramfs there. It's a total of ~10MB. Doesn't matter how slow vfat is, it's still only 10MB split into only 2 files.

Whatever you do to speedup boot, you're not gonna beat resuming from suspend. If you want to do something really for show, you might consider hibernation. With SSD loading RAM from disk will probably be much faster than going through the boot process.
So, hibernate your system once, do some magic to protect the resulting file, and then you can power on almost instantly. You'd have to rebuild the hibernation file after updating your system though.

[HystericalLich]

I suppose hibernation probably is my best bet actually.
If anyone thinks up anything though, I would be interested to hear.

[HystericalLich]

After some tests, it turns out, I am wrong there. Contrary to what we might expect, my system boots faster from a clean state than it does from hibernate.
Hibernate appears to significantly increase the loading time.

[Ant P.]

`dmesg -H` might show you what's taking a long time in the kernel. Look for the numbers on the left.

[HystericalLich]

[Ant P.]

Seems to spend a pretty long time doing plug-and-play device setup... is this a genkernel config?

[C5ace]

Had a similar problem 2 days ago after installing Gentoo on a new HP laptop with genkernel, openrc , grub-static, Xfce.

Very long boot time compared to Gentoo installed with identiacal configuartion on an identical laptop in November 2015. The only difference was using install-amd64-minimal-20160204.iso and stage3-amd64-20160204.tar.bz2 instead of the November 2015 versions and that the resulting compressed initramfs is about 32Mbyte versus about 3Mbyte for the November 2015 version.

After various rebuilds and a test install in VirtualBox also resulting in a 32Mbyte compressed initramfs, I copied the kernel, initramfs to /boot/*.bak, deleted kernel and initramfs files, renamed the *.bak files and rebooted. The result was nearly the same boot speed as the November 2015 laptop.

I belive the slow booting was caused by fragmentation of kernel and/or initramfs files.

EDIT:
Corrected "kernel" to "initramfs".

[HystericalLich]

@Ant P. This is not a genkernel config, the config is linked to pastbin in my second post. I don't know a lot about plug and play, it was enabled by default, and I left it that way.
@C5ace This is not a file fragmentation issue. I have an SSD so it should have no effect if present. My kernel is 5.6M.

[HystericalLich]

So I was able to cut 4 seconds by switching my kernel compression mode to LZ4.
Also did some userspace optimization to reduce a small overhead there. (Mostly just rewriting blocking scripts so that they forked)

[dmpogo]

[HystericalLich]

I did some research, and messed around with kernel compression as well as initramfs compression, trying all supported options.
What I found was that on my ssd, LZ4 got the best kernel loading time.
As for the initramfs, it seemed to load fastest with .xz for some reason (disk I/O?), but then I realized by accident my system doesn't even need one. Although there isn't a notable boot time increase between an .xz compressed initramfs and no initramfs, strangely. Note that an LZ4 initramfs was not supported by genkernel, so I did not test it.

[C5ace]