Re: GENERIC and INSTALL kernels (was: Re: CVS commit: src/sys/arch/i386/conf)

[Thor Lancelot Simon <tls%panix.com@localhost>]

On Fri, Feb 11, 2011 at 05:12:47PM +0100, Jean-Yves Migeon wrote:
> 
> Problem 2:
> Being able to mountroot the ramdisk that contains the modules is
> sufficient in itself. However, for install ramdisks, you would have to
> include modules + sysinst + some tools needed for install. You end up
> having a ramdisk with a size close to 25MiB (count 10MiB for /stand, and
> some more MiB for tools/sysinst).

And what's exactly wrong with that?  The installation sets are that
size already; what difference is it how they're packaged?

I have a script that effectively produces NetBSD USB key images using
only in-tree tools (it actually produces disk images for a different
purpose entirely, but they can be used for that).  It creates a FAT16
filesystem and puts four things in it: bootxx_fat16 (as the FAT
boot loader), /boot, /netbsd.gz, and /miniroot.gz

/miniroot.gz has a *complete runnable* NetBSD system in it, plus all
the install sets.  Our install sets are somewhat smaller than the
standard NetBSD ones (because we subset each setlist with an additional
setlist) but the whole thing easily fits on a 128MB flash module
and requires only about 256MB of RAM to run.  It would not take much
more space for a fairly standard "Live NetBSD" image that put back
most or all of what our runtime takes out.  This ends up being a very
useful thing and can be built with only in-tree tools and can even be
manipulated and pieced together on *other* operating systems, so long
as they can read/write FAT.

Suffice to say I do not think "there's a giant miniroot!" is much
reason to reject an approach to installation or upgrade.

Thor