On 10.02.2011 22:23, David Laight wrote:
> On Thu, Feb 10, 2011 at 04:49:19PM +0000, Jean-Yves Migeon wrote:
>> Module Name: src
>> Committed By:        jym
>> Date:                Thu Feb 10 16:49:19 UTC 2011
>>
>> Modified Files:
>>      src/sys/arch/i386/conf: INSTALL
>>
>> Log Message:
>> For i386, include MONOLITHIC for INSTALL rather than GENERIC. While here,
>> remove drm drivers, we don't need them for install.
>>
>> i386 GENERIC has FFS and ELF support compiled as modules, so we hit
>> an interesting "chicken-egg" situation when the kernel attempts to mount
>> a ffs ramdisk, while the module might be contained inside... the ramdisk.
> 
> I'm not 100% sure it is worth having FFS and ELF support as modules
> in GENERIC.
> It may be nice that they CAN be modules, but I suspect 99.99% of systems
> will need them.
> Anyone who wants them as modules can build a kernel without them.

It's something I mentioned privately with Jared. I think we could come
up with a golden mean for GENERIC kernels: leave most "third party"
drivers/systems as modules, while keeping critical ones included by
default. FFS and ELF come to mind, but there are others too. amd64
GENERIC is close to this.

This would open up the possibility to provide a modular kernel, without
going the "all or nothing" MONOLITHIC way (and avoid many complaints
like "I just replaced my kernel for testing, and it returns an error
when attempting to mount / or exec init").

BTW, I wonder whether modules shouldn't be part of the kern-GENERIC.tgz
set. But this is an orthogonal issue.

-- 
Jean-Yves Migeon
jeanyves.mig...@free.fr

Reply via email to