Formalize the use of linker scripts for kernel modules.

[markj]

Summary

Automatically use ldscript.kmod.${MACHINE_CPUARCH} if it exists.
We already have an i386-specific linker script; rename it accordingly.
I plan to add an amd64 linker script to insert padding after .text
and .rodata*, so that permission boundaries are page-aligned. (It turns
out that my plan to use DSOs on amd64 will not work because of issues
that arise when combining -mcmodel=kernel and -fPIC.)

Note that the linker script is applied during the partial link. For
amd64 (which uses .o's), this is also the final link. I think this is
probably fine, though maybe it will be useful someday to be able to
script the final link instead.

[imp]

sys/conf/kmod.mk

252 ↗

(On Diff #62875)

So this is 'mips' for all mips things and 'powerpc' for all powerpc things.
History has shown that we really want to have both the ability to check MACHINE_ARCH as well as the more generic MACHINE_CPUARCH.
Would it be possible to update to do that?

[jhb]

sys/conf/kmod.mk

252 ↗

(On Diff #62875)

We can probably wait to fix that until an actual use case turns up. My inclination would be to actually use MACHINE_ARCH instead of MACHINE_CPUARCH for now (a no-op on i386 and amd64), though it really doesn't matter. I'm somewhat surprised the original code used CPUARCH for this.

[markj]

Use MACHINE_ARCH instead of MACHINE_CPUARCH.

[imp]

I'm cool with jhb's reasoning, so this looks ready to land.