Linux for non-CMOV processors

[jack-rp9/bkPP+cDYtjvyW6yDsg at public.gmane.org (Jack Coats)]

The idea of using an invalid instruction to talk with the OS so it can
interpret the instruction is a great
technology.  That is what IBM mainframes used to let their Virtual
Machine software work.  They named
their invalid instruction a 'diagnose', and it's roots came from SE
code (IBM Systems Engineers that wrote
diagnostics used it, since all invalid instructions caused
'interrupts'.  It was really a pretty efficient way to
implement a VM system.

Yes, keeping 'legacy' hardware running does cause a certain amount of
continuing overhead in newer systems.
But it seems reasonable to have a fork and run parallel
implementations at least for a while.

This would be similar to supporting i386, ARM, PowerPC, 68000, or any
other architecture.

Having the 'invalid instruction' interpreter isn't needed on machines
that can run it natively, or at least
not for THAT instruction.

Just a few thoughts. ... But hey, I'm not a hard core developer, and
they are the ones that implement
it for those of us that can't/won't/don't have the time.