[systemd-devel] Scheme: Linux startup process with systemd

[Jan Engelhardt]

On Tuesday 2013-11-05 00:28, ScotXW wrote:
>>
>> Seems rather strange to find GRUB and Linux in "BIOS".
>> And if you talk about Uboot and Redboot and whatever else, then
>> marking the orange section as "GRUB" is sorta outta place.
>
> Linux kernel and GNU GRUB can be payloads to coreboot.

That may be so, but you have a separate bootloader layer already.

> However, the y-axis is the time from the connection to power. Does it make more
> sense now?

No, it makes even less. Time is on x, if at all. What you have there
is a timeless diagram, just like
http://en.wikipedia.org/wiki/File:Netfilter-components.svg .

>> It's also wrong on so many accounts.
>> -initramfs is not mounted, because it is not a filesystem.
>
> AFAIK ramfs = ram file system, so I assumed it is being mounted.

If you want to be exact, yes. "ramfs" is mounted, implicitly.
And it's empty by default.

But any standard reader would assume "initramfs" to mean "initramfs
image". And that's what is not mounted but extracted. Unless it's an
initrd, in which case there is something (with contents) to mount.

>> -kernel does not execute /usr/bin/systemd but /init from the initramfs,
>> or in case none was used, /sbin/init.
>
> Hö? You lost me.

init/main.c: if (!run_init_process("/sbin/init") ||
...
init/main.c:         ramdisk_execute_command = "/init";


There is no mention of "systemd" in the linux kernel source *at all*.

>> -/usr/lib/systemd/system is a directory, not an executable entity
> So "system processes the declarative files in /../"?

For brevity, "units in /../"

>> -you already have systemd, so having systemd again in "user interfaces"
>> is sort of inside out
>
> Yeah, there is alink missing between them, because systemd gets startet and
> then remains available as a daemon.

Then you should write systemctl. The way it looks now is as if you
are referring to the systemd --user instances (which are pretty
much unimportant as of now)