[Coco] Tandy Assembly Exhibit: CocoMEM

[William Astle]

I suspect a lot of "advanced" OS9 programs will fail under proper memory 
protection. Some of the descriptions of "clever" techniques I've heard 
over the years lead me to suspect all manner of things that really 
aren't kosher being (ab)used.

Speaking with my OS development hat on, I would much prefer an NMI on a 
"segfault" over FIRQ or IRQ.[1] That way a program can't mask it, even 
temporarily, by disabling interrupts. That will make it a *lot* easier 
to identify the source of the fault in most poorly written programs 
since it will always occur immediately after the instruction that 
triggered it finishes, at least if I have my facts on the 6809 straight. 
Also, the NMI is edge triggered so you wouldn't have to maintain an 
interrupt state anywhere once it fires.

To head off the primary objection: this doesn't need to conflict with 
the FDC's use of the NMI. One could simply decide that an NMI occurring 
in "system  mode" when an FDC read/write loop is active is an FDC 
interrupt and anything else is a protection fault. The coding for that 
isn't that difficult (software wise).[3]

[1] Actually, I would much prefer something more integrated into the CPU 
but that isn't going to happen without replacing the CPU with something 
completely redesigned. Definitely not practicable[2] for a memory board.

[2] Cool word, practicable. Let's call that the word of the day. :)

[3] Well, no more difficult than getting user<->system IRQ handling 
correct, anyway.

On 2017-10-12 08:14 AM, Brett Gordon wrote:
> Cool beans!  A mmu with read/write protection (when exploited) should make
> for some interesting times.  I wonder how many poorly coded nitros9 (and
> fuzix) proggies will seg fault :)  I imagine a segfault triggers a firq
> (cart) then?
> 
> brett
> 
> On Oct 11, 2017 10:38 PM, "RETRO Innovations" <go4retro at go4retro.com> wrote:
> 
>> On 10/6/2017 11:00 PM, camillus gmail wrote:
>>
>>> Hi Jim,
>>>
>>> Do you except preorders?
>>>
>>> I'm in for the 32MB @ $24 ( is this the total cost or just the mem chips?)
>>>
>> It's just memory chip cost, not total unit cost.
>>
>> Let me finish getting OS9 running (close, just need to verify >2MB
>> support), clean up the Verilog, and price out the unit
>>
>> Jim
>>
>>
>> --
>> Coco mailing list
>> Coco at maltedmedia.com
>> https://pairlist5.pair.net/mailman/listinfo/coco
>>
>