[Coco] Memory Expansion Beyond 64k for the Coco 2

[Juan Castro]

Doc: In the 80s, I worked on a memory expansion that banked the top 32KB
just like you said. They pulled the SELECT pins of the mobo RAMs out of the
socket and wired them to the output of a 74LS138. Eight outputs times 32KB
meant the expansion went up to 256 KB. All using the same kind of DRAM as
the main memory. Hardware project was not mine, I just did firmware
routines so the company's business, made in DECB (!!!) could use the memory.

Aaaron, it's nice to know that would be not good for NitrOS-9. I imagine
another possibility: what about a cartridge that can, internally, swap 16
KB blocks? All RAM could be accessed in addresses C000-FEFF. Maybe divide
into 2 8KB blocks. It would work much like the EMS of ancient PCs. And it
would require no bodily violation of a CoCo 2. (You might recall a question
I asked recently, it was with that purpose.) Should be a simple circuit,
right? Especially if it used SRAM?

Juan

On Thu, Jun 28, 2012 at 8:00 AM, Aaron Wolfe <aawolfe at gmail.com> wrote:

> On Thu, Jun 28, 2012 at 5:50 AM, Computer Doc <computerdoc at sc.rr.com>
> wrote:
> > Hello Everybody,
> >
> > (snip)
> > Is there a memory expansion upgrade available for the 64KB Coco 2?  I
> want
> > to upgrade my 64KB Coco 2, and also add Memory Management (MM) of some
> type
> > to Grant Searle's 6-Chip 6809 Computer.  I say for the Coco 2 because it
> > doesn't use a GIME chip, which except for the graphics portion and the
> > MC68B21 PIA chips, it's close (not exactly of course) to Grant Searle's
> > 6-chip 6809 computer which is also a 64KB design using 32KB of RAM and
> 32KB
> > of EPROM.  I've thought about bank switching 32KB of RAM at a time for a
> > total of two banks.  I've drawn up some circuits with this design idea
> using
> > a 4Mb chip yielding 512KB of RAM, but I'll need to write a driver for
> > NitrOS-9  eventually.
>
> (snip)
> I think NitrOS9 or OS9L2 would take an awful lot of adaptation if you
> could only swap 32k banks, if supporting such a scheme is even
> possible.  I'm not intimately familiar with the kernel, but what I've
> seen leads me to believe that one page dedicated to the system has to
> stay mapped in pretty much all the time, and the OS9 calls that copy
> memory blocks between address space in two different pages do so by
> mapping both pages in.  That means 3 pages minimum without changing
> the plumbing in some pretty dramatic ways, unless I misunderstand how
> it works (which is certainly possible).   You might want to consider
> doing 8k blocks like the GIME just to save that kind of work, if it
> doesn't make your design too much more complex.
>