Who would be interrested in flash replacement kickstart pcb?

[FrenchShark]

Hello,

I am about to order some small PCBs the size of a DIL-40 that will hold a 29F400 flash. They will be a drop-in replacement for Kickstart ROMs with two wires to connect : _WE and _RESET.
On the 32-bit machines, the flash rom will be 1MB total (2 x 29f400) with two PCBs.

The only problem I see : how do we start the Amiga with a blank flash the first time ?
Either, I provide the PCB pre-flashed or the user has to piggy-back the Kickstart ROM on the PCB and put a switch to select ROM or flash.

What do you think ?

Regards,

Frederic

[alexh]

Drop in on which Amiga? I fear the form factor is different on different Amiga's. You'd have to have two PCB's connected via a ribbon cable to make it universally compatible.

[Toni Wilen]

Suggestions:

42 pin DIL + PCB solder pads for 2 "extra" address lines (+pulldown resistors) -> simple kickstart switcher and/or makes 1M/2M ROM experiments easy (I still have some plans, eprom erasing is sooo slow) and also it would be CD32 compatible.

At least 8Mbit flash chip support.

Socket for original ROM + switch is imho the easiest and best choice.

[Photon]

Yeah, make a kickstart switch of it Should be dead easy for backup and trying various (existing backups of ;P) Kickstarts. Legal department should love it, you've made it for the sole purpose of providing a backup function to preserve legacy firmware!

[yaqube]

I have almost completed flash kickrom board for my A4000D (this project is currently on hold). It uses two 29F400 flash chips and has two sockets for original ROMs (required for initial flashrom programming).

I have found some issues with fitting flash rom into Amiga:
1. _ROMEN (rom chip enable) is generated by GARY for read access only.
For writes you have to:
- generate _CE for flashrom chips which would be active during CPU write cycles.
- generate _DSACK0/1 cycle termination signals with appropriate timing (flashrom access speed), otherwise the cycle will be terminated by bus timeout counter which is very slow.

2. Gary doesn't support 1MB ROMs. _ROMEN is generated for the $F80000-$FFFFFF range only. In A4000T for 1MB ROM support there is additional logic which maps $E00000 space into $F80000 so Gary response with _ROMEN. This address lines logic only affects Gary.

3. I don't know exactly how auto vectoring of interrupts are solved on big Amigas (I guess they use _AVEC signal during interrupt acknowledge cycle) but A500/A600 reads the interrupt vector from ROM (last 7 words). So if you use FC0-2 lines for _ROMEN generation remember about activating it during interrupt acknowledge cycle.

[Toni Wilen]

Originally Posted by yaqube

Gary doesn't support 1MB ROMs. _ROMEN in generated for the $F80000-$FFFFFF range only.

Gary surely does. ROM appears at 0xE0000-0xE7FFFF (at least on my real A500) Of course extra address line is missing but it is easy to add.

Also on my A1200 ROM (Gayle) appears at 0xA80000-0xB7FFFF and 0xE0000-0xE7FFFF (2M supported). Unfortunately this does not work with Blizzard Map ROM.

In A4000T for 1MB ROM support there is additional logic which maps $E00000 space into $F80000 so Gary response with _ROMEN. This address lines logic only affects Gary.

A4000 does not have Gary chip
(neither does A600 or A1200)

[yaqube]

Originally Posted by Toni Wilen

Gary surely does. ROM appears at 0xE0000-0xE7FFFF (at least on my real A500) Of course extra address line is missing but it is easy to add.

On my A4000 there is no ROM alias at $E00000. Look for P155 PAL in the A4000T schematics. It modifies A20 and A19 address lines going to the Fat Gary.

Also on my A1200 ROM (Gayle) appears at 0xA80000-0xB7FFFF and 0xE0000-0xE7FFFF (2M supported). Unfortunately this does not work with Blizzard Map ROM.

No ROM here in my A4000.

A4000 does not have Gary chip
(neither does A600 or A1200)

Sorry, my bad. A4000 (and A4000T) has Fat Gary.

[DJBase]

Theres already a project in final stage...

http://www.a1k.org/forum/showthread.php?t=13078

[Toni Wilen]

Originally Posted by yaqube

On my A4000 there is no ROM alias at $E00000. Look for P155 PAL in the A4000T schematics. It modifies A20 and A19 address lines going to the Fat Gary.

My A4000T schematics have really crappy quality, haven't noticed this before.. Weird hack, I don't see the point. Just remove the chip and reconnect address lines normally

Theres already a project in final stage...

Maybe but german-only projects can be considered non-existing, unfortunately

[DJBase]

Well once its finished I am sure there will be also an english description.

[Toni Wilen]

Originally Posted by DJBase

Well once its finished I am sure there will be also an english description.

Perhaps but then it is too late to improve the design with something I'd like to have

[DJBase]

What would you like to have? Maybe I can bring in your ideas.

[Doobrey]

Originally Posted by Toni Wilen

My A4000T schematics have really crappy quality, haven't noticed this before.. Weird hack, I don't see the point. Just remove the chip and reconnect address lines normally

It simply fools Fat Gary into seeing any reads to 0xE0000-0xE7FFFF as being in the 'normal' 0xF8000-0xFFFFFF range, so it asserts /ROMEN and handles the cycle termination etc.

No idea why C= removed the 0xE0000-0xE7FFFF ROM mapping from Fat Gary, I did some tests last year and the A3000 suffers from it too

[Toni Wilen]

Originally Posted by Doobrey

It simply fools Fat Gary into seeing any reads to 0xE0000-0xE7FFFF as being in the 'normal' 0xF8000-0xFFFFFF range, so it asserts /ROMEN and handles the cycle termination etc.

No idea why C= removed the 0xE0000-0xE7FFFF ROM mapping from Fat Gary, I did some tests last year and the A3000 suffers from it too

Ah. I thought that circuit disabled Fat Gary's 0xe0 mapping but it "fake enables" it..

What do A3000 and A4000 (non-T) return at 0xA80000-0xB7FFFF and 0xE00000-0xE7FFFF? "nothing" and/or "bus noise"? (I really need to find/buy/steal A3000 soon..)

So this means only "standard" Gary Amigas (A500, A500+, A2000, CDTV), Gayle-based Amigas (A600, A1200) and CD32 have 1M+ ROM support. Not sure about A1000 (It does not have Gary but separate logic chips, will test soon, A1000 soonish here..)

[Doobrey]

Originally Posted by Toni Wilen

What do A3000 and A4000 (non-T) return at 0xA80000-0xB7FFFF and 0xE00000-0xE7FFFF? "nothing" and/or "bus noise"?

IIRC, The A4000T has the same problem too, that PAL mentioned above is marked as 'optional' on the schematics I've seen, and I've not seen it fitted on any of the photos on amiga-hardware.com or amiga.resource.cx

From what I remember about the tests I tried last year.. (fuzzy memory!)
68060.library and 68040.library remap the 0xe00000-0xe7ffff, so I loaded a fake 'cdstrap' module which makes them think it's a CD32 and stops the remapping.

After that, any reads to 0xe00000 gave a 8000 0002 "bus error" guru.
(Also tested by booting without startup-sequence, so no '040.lib loaded)

That's changeable by flipping a bit in Ramsey or Fat Gary(damn my memory!), but then the default cycle termination kicks in, which meant I was only getting around 300KB/s reads. But it was 300KB/s of reading bugger all, just returning 0xFFFFFFFF.

There was some code in expansion.library that sets it to 'bus error mode' instead of 'timeout mode'

edit: found it, it was Fat Gary's TOENB register set to 'bus error mode'.. MOVE.B #$80,$DE0000

[FrenchShark]

@ all,

thanks you guys for the replies.
Since the address decoding is not done during write, I think I am gonna drop the project .
Too bad, I have a bunch of 29F400 laying around. I thought turning them into kickstarts was a good choice.
I am going back to my FPGA projects .

Regards,

Frederic

[coze]

Hasn't this been done before ?

KF500

[Toni Wilen]

Originally Posted by Doobrey

After that, any reads to 0xe00000 gave a 8000 0002 "bus error" guru.

Ok, so this area is not mapped at all.

UAE does handles this incorrectly but practically nobody cares and it is probably safer because MMU isn't emulated..

edit: found it, it was Fat Gary's TOENB register set to 'bus error mode'.. MOVE.B #$80,$DE0000

This is also "emulated" but does nothing, see above

Hasn't this been done before ?

More than once I guess, but so far every one appears to be very basic without any extra features, like jumperless design

Seriously, problem is building one, I am too lazy and most people are not interested in soldering. Ready-made flash chip to 16-bit 40/42pin ROM adapter would be nice.