Kernel package and modules

[Werner Almesberger]

Graeme Gregory wrote:
> I changed GTA02 yesterday to install all modules built.
> 
> This uses 5.6M of storage (uncompressed no idea how to get the
> compressed size out of jffs2)

Ah, I only looked at the modules we build, and they took 10x as much
space. 5.6MB isn't too bad a starting point. Still 10% of poor GTA01,
though.

> I guess Werner made a valid point, do we need them all to be modules. I
> know now I stopped debugging it I'm happy for sound to join monolithic
> kernel. I guess some of the file-systems can as well.

Great ! Not having stuff in modules also makes it easier to debug it
through JTAG. Also, having things we're inevitably going to load
anyway not as modules may also shave a few milliseconds off total
boot time ;-)

However, the problem we have at hand is a different one: some kernel
features rely on modules as a selection mechanism. I.e., the USB
function a device assumes is determined by what module is loaded,
and there is no mechanism in the monolithic kernel to allow multiple
such drivers to coexist.

So, unless someone steps in and provides a mechanism for dynamically
switching such drivers in a monolithic kernel, we need to solve the
issues resulting from selecting them through the loading of modules.

Andy, by the way, how urgent is that functionality actually ? It
doesn't look like anything ASU or FSO would need. If it's only a
fringe interest, it may be better to just stick with having only
EoUSB in the regular config, and solve this properly at a later time.

- Werner