In message <20041025183949.EE13B49224C@s102-n054.tele2.cz>,
Jaromir Dolecek writes:

>Yeah, in-kernel linker and thus ability to checksum 'safe'
>LKMs would be great.

Yes, exactly.

But once you go the route of having an in-kernel ELF linker, we should
really ask how much of the SunOS-4-inspired lkm(4) is worth keeping.

With linksets, and well-defined per-system hooks for modules to link
themselves into per-subsystem kernel infrastructure, the lkm(4) module
types serve no purpose, not that I can see.  But I think per-subsystem
hooks are strictly orthogonal: we want them regardless.

I just dont' have enough details about per-CPU ELF variants
(GOTs, all that stuff) at my fingertips to know how feasible a FreeBSD-style
kld(4) really is, for all the CPUs we support.  

(Flames about the FreeBSD kern/kern_linker.c implementation  to /dev/null.)