On another list, Warner Losh writes about the SOFTFLOAT kernel option
added to sys/arch/mips for the hpcmips port:

>In message <19990916201148.B464@jocelyn.rhein.de> Ignatios Souvatzis writes:
>: wait. is this "kernel doesn't do FP operations"?
>
>Yes.
>
>: If not: Shouldn't it detect no-FPU at boot time, and use a modified context
>: switch function in this case?
>
>Yes, but there is also some issues with setjmp/longjmp to deal with
>since you cannot save FP registers that don't exist...
>
>There may be other issues as well...

Like ptrace()/procfs access to userlevel fp registers?  (same deal, really).
Also kernel-debugger register-dump code.

How do other ports handle this (e.g., mac68k on FPU-less 68040LC)?
Is that a model the mips ports should follow?