[klibc] The future of klibc in the kernel

[Jeff Bailey]

Le mardi 03 octobre 2006 à 14:02 -0700, H. Peter Anvin a écrit :
> Well, as of right now, both Linus and Andrew have rejected klibc in the 
> kernel based on "it doesn't add anything new" (Linus) and "lack of 
> interest" (Andrew).  It seems that there is little hope of getting it 
> into the kernel any time soon, unless there is additional features, 
> and/or people start actively asking for it.
> 
> I would appreciate hearing suggestions.  The suggestions so far include:
> 
> - nbd boot in kinit
> - Mount by label support
> - nfsroot over IPv6 (easy in klibc, but it requires the kernel to
>    support NFS over IPv6, which it currently doesn't.)

I think we're suffering from a slight dogma difference.

The kernel folks want everything involved in bringing up the system to
user init to be included.

The distro folks don't care.  We get packages from everywhere anyway.
One more is hardly a problem.  The tradeoff in this case is decently
modular kernels and a single initramfs that does NFS booting, Harddrive
booting, and other crazy thing we've cooked up.

Having it in the kernel hasn't been a pre-requisite for using it in
Ubuntu and Debian.  We depend on udev, module-init-tools, klibc, and
busybox to get the system up and running.  

It would be really nice to make that not also draw in glibc at some
point.  Right now I think we pull in glibc for udev and busybox.

I guess what might be a nice patch to the kernel is to make the
in-kernel bits that klibc replaces configurable.  That way the distros
could just disable them and not worry whether they're there or not.
Anyone using our kernels is using a setup that doesn't have any device
drivers loaded in anyway, so they're already forced to use this.

-- 
Jeff Bailey
Manager, Technical Support
Canonical Ltd. - Sales, Service, and Support.
+1 514 940-8910
http://www.ubuntu.com/