Re: core's decision on modular kernels

[David Laight <david%l8s.co.uk@localhost>]

On Fri, Sep 23, 2011 at 07:13:16PM +0300, Jukka Ruohonen wrote:
> On Thu, Sep 22, 2011 at 08:35:02PM +0100, David Laight wrote:
> > > I think that by MODULAR with built-in modules, you mean a barebones
> > > kernel linked with some .kmod's?  I would love to see that.  What has to
> > > happen to make it so?
> > 
> > Probably just some 'round tuits'.
> > Mostly in the area of config() and the kernel makefile.
> > 
> > First stage would be linking an existing kmod into the kernel and sorting
> > out the required data area linkage to get it initialised.
> 
> Wouldn't this provide an answer also the difficult question of autoloading
> driver modules? Assuming a robust mechanism, link everything and then
> selectively unload what did not attach during autoconfiguration?

Not really, linking in a module will do all its code/data fixups, so
you wouldn't want to unload. Any, in any case, the data wouldn't really
be appropraitely aligned...

Of course, you could put a load of config stuff and modules in a separate
filesystem that is 'magically' loaded at the same time as the kernel
and load stuff from there - then free the entire space :-)

        David

-- 
David Laight: david%l8s.co.uk@localhost