Re: [D-I] Supporting 2.6.14 kernels in base-installer

[Jeff Bailey <jbailey@ubuntu.com>]

On dim, 2005-11-13 at 20:34 +0100, Frans Pop wrote:
> On Sunday 13 November 2005 18:32, Jeff Bailey wrote:
> > On dim, 2005-11-13 at 12:46 +0100, Bastian Blank wrote:
> > > I think it is possible at least for initramfs-tools to run without
> > > sysfs on the building system.
> >
> > Right.  The caveat to this is "dep" mode.  The default mode for
> > initramfs-tools is to include all of the modules that you're likely
> > interested in booting with in the initramfs and detecting which ones to
> > use at boot time.  If you ask it to detect which modules are needed, it
> > needs a valid sysfs tree to scan (although it is resiliant in that case
> > against module name changes)
> 
> Thanks for the explanation.
> 
> I think the "dep" mode is to be preferred over the "most" mode in 
> principle, so this would need to be configured by d-i.
> I very much doubt that mounting /sys in the chroot will be a problem. It 
> has worked for me so far when installing 2.6.14 kernels in a chroot.

FWIW, Ubuntu's install defaults to 'most' mode on the grounds that for
most systems (not lowmem, not oldworld ppc, not netboot) there's no harm
in having a larger initramfs (approx 5 meg on disk, 40meg in memory).
The upside is that the initramfs created should be more or less
identical for every system and is resilient against people moving the
drive from one machine to the other, doing perfect copies (using ghost,
dd, or whatnot), or using an already generated initramfs to recover
broken systems on other machines.

I'd argue for keeping that mode as default if possible because there
isn't any benefit to the smaller initramfs in 95% of cases, and it
increases the risk of a non-booting system.

Tks,
Jeff Bailey