init observations

[Avi Kivity]

Chris Adams wrote:

>Once upon a time, Avi Kivity <avi at argo.co.il> said:
>  
>
>>We require /. We require initrd.
>>    
>>
>
>initrd is not currently required.  It is for out-of-the-box setups, but
>you can:
>
>a) not use LVM for root
>b) not use ext3 for root or rebuild kernel with ext3 included
>c) use hardware that doesn't require modules for root device or rebuild
>   kernel with drivers included
>
>  
>
seems very close to "required" to me.

>>I'm sure we can arrange initrd to look at the command line. For example:
>>
>> mount=/=server1:/path/to/root,/usr=server2:/path/to/usr
>>
>>IP can already be configured this way.
>>    
>>
>
>That's a pretty ugly hack just to get python support for startup.
>
It's not just for python; it's to avoid dependency on a very small 
subset of functionaly present in /bin and /sbin. For flexibility.

>  Also,
>that is limited; the kernel command line has a limit that isn't all that
>big.
>
This limit is going away.

>  You also then need to handle fsck and such in initrd for
>non-network filesystems.
>
>  
>
This can be handled in the same way that fsck / is done.

The point is that we have two (well, more, but in this context just two) 
bootstrap sequences: initrd -> / and / -> rest of world. the first 
bootstrap does not leave permanent effects, but the second does, and 
these hurt. Eliminating the second stage would simplify the system and 
open the door for many improvements.

-- 
Do not meddle in the internals of kernels, for they are subtle and quick to panic.