Re: x86 release builds are slow

[Andrew Doran <ad%netbsd.org@localhost>]

On Thu, May 01, 2008 at 10:39:37AM +0900, Masao Uebayashi wrote:

> > For i386 and amd64, we could change the process so that the GENERIC kernel
> > is used for installs, and the root file system objcopy'd into a seperate ELF
> > binary that is loaded by the boot loader. We could modify the in-kernel ELF
> > linker to set the root file system image up. Any comments on that?
> 
> This sounds nice. :)
> 
> But how is the filesystem image passed from bootloader to kernel?  I
> think the memory area is out of kernel's control first, and then
> kernel loads the area later, right?

On x86 the _boot loader_ would load the module into memory and pass a
bootinfo record describing it. So the kernel would be told of a list
of ELF objects that it should try to link with kobj_load().

Thinking about it further, it could be like any other kernel module. We
could have the md_root code do a ksyms_lookup() for its boot image, or
have simple module init code in the root disk image call up to say hello.
So you would do mdsetroot on the kernel module, not the kernel.

Andrew