The first thing to do is to price ColdFire chips of the type you might wish to use. The problem, in my mind, with the 68060 accelerator idea is that the 68060 chip costs (last time I checked) well over $100 per unit. So even if you designed and built such an upgrade, it would cost in the neighborhood of $200 each even if you gave your time away for free and assumed the risk of not getting your money back after building some number of units.
But if fast Coldfire chips are, say, $30 each, then that would be an affordable upgrade.
I haven't looked at Coldfire in a while but are they 32 bit wide processors? How much does the 300 MHz version cost per chip? Which version of the chip is most 68K compatible?
If the processors are down under $30 each I'd say that it is very doable project, although I'm unclear on how you get the board to run 68K code without losing too much performance.
Ideally, you'd want the Coldfire to execute instructions without any other processing, when an instruction is compatible so that you don't lose any CPU cycles. Yet, you need some mechanism for intercepting non-compatible 68K instructions and translating them, otherwise. Perhaps a little FPGA or second microprocessor could sit between the Coldfire and the datastream from the Mac and monitor the instructions as they go by and provide translation as needed. The only problem with this approach is how does the monitor system distinguish between data and instruction transactions?
Ah, I read the link above. If I'm reading it correctly, 68K instructions which are not supported on the Coldfire, should generate an exception, which can then be handled by executing the proper code on the Coldfire to emulate the excepted 68K instruction. Very neat. It's the same mechanism which the Mac Toolbox uses to execute its routines. Now, would that be completely compatible or only mostly compatible? And again, how much to purchase each Coldfire CPU?
Another approach would be to design a 68030 chip into an FPGA. Chips which can do the deed are available for under $20 each now. But I think the magnitude of completely designing a 68030 compatible CPU chip, even with modern Verilog or VHDL would be enormous. Heck, just reading the User's Manual for the 68030 such that you understand what needs to be emulated is a daunting task. There's a ton of stuff going on behind the scenes on that chip.
Beyond properly emulating each instruction, one must make sure that all the control registers, interrupt functionality and supervisory systems operate as expected. That's a lot of stuff.
