early Interrupt Acknowledge. Defining unused opcodes.

[ElEctric_EyE]

Excellent info, thank you for the quick response! What you have mentioned is alot for my small brain to absorb...

Right now I (think?) I understand a few things:

1) It is possible to address up to 16MB contiguously, and have a JSR/RTS cross page boundaries automatically? (I DO need to re-re-re-read your webpage about this issue to fully understand how it works.). But, is this what you were referring to in your last paragraph of your last post?

2) I do know the signals I will need for the FPGA.

3) I will have to check the details between Rockwell's and Western Design Center's 6502's NOP's.

Thanks for your help. I am at the "if we see, we remember" stage. Soon I will be at the "if we do, we understand stage" and be able to add positive feedback hopefully.

[Dr Jefyll]

ElEctric_EyE wrote:

is this what you were referring to in your last paragraph of your last post?

No, I think you've raised a different topic. There's nothing automatic about what I was describing, just the simple question of how best to let the program explicitly control what appears in the 16K "window" you were talking about.

I'd better double-check that we're talking about the same thing! I visualize your window as occupying (say) addresses 8000 to C000 hex. By altering the contents of your circuit's "Window Select Register" -- I just made up that name, btw -- you could cause various 16K chunks of your humongous EEPROM/ram/whatever to appear at 8000 - C000. If the WSR is 8 bits wide then that would allow you to select between as many as 256 different different chunks, for a grand total well beyond the 65C02's usual 64K addressing limit. (The same trick was popular at one time for PCs. Known by the Marketing term "Expanded Memory," it allowed 8088 CPUs to transcend their usual 1 MB addressing limit.)

As for a means to update the WSR (so that different chunks may be selected), the conventional technique is to simply assign it an address that memory-address decoding circuitry will recognize. Then the update becomes a simple matter of the program writing from a CPU register to the WSR's address. In other words, just an ordinary output port. (The updated value is stored immediately in the WSR latch/flip-flop. But the stored value isn't used except when 8000 - C000 addresses appears on the address bus. C000 - 8000 addresses are what cue your hardware, indicating cycle-by-cycle whether or not the current access should be treated as part of the window.)

If what I've just described is comfortably within your capabilities then you can progress to the fancier approach of defining a new opcode capable of performing the update. But the fancy approach doesn't offer any earth-shattering advantage; writing to an I/O address is still a perfectly acceptable solution. (What is a big advantage is being able to juggle windows which are a full 64K in size. KK manages this, but at a cost in terms of complexity. Since the window is 64K in size, that means the CPU address bus is now useless for telling you whether or not to treat any given bus cycle as part of the window! Instead, microcode or other means are needed to make that window/not-window decision every cycle.)

Remember that your circuit must allow the regions not in the window (the regions 0 - 8000 and C000 - FFFF) to remain fixed, so that stuff like your stack, Zero Page and the interrupt vectors doesn't disappear! Reiterating, the WSR comes into play only on cycles during which the CPU produces addresses that fall within the window. BTW you'll notice that none of this requires that your CPLD be clocked faster than the CPU clock rate. Have fun, & bye for now
[Edit: clarity; remarks re cycle-by-cycle timing]

[ElEctric_EyE]

OK, didn't mean to keep raising new topics... For now my goal will be to do a simple data bus "snoop" for one of the undefined NOP's, and then for the following data to be written into the 7 bit bank register, which controls A14-A20. (maybe 8 bits for 2x(4Mx8), since decoding will require just one more inverter)

I still plan to incorporate the EEPROM to SRAM COPY so the 65C02 can run as fast as possible. I was hoping for 5V operation. The FMax vs VDD says at 5V to expect around 20MHz. Unfortunately for me, large/fast SRAM's do not run at these voltages. So, everything will be running at 3.3V, limiting speed of the CPU to 14MHz.

Off for a couple weeks to wire-up a new design using...:

1 5nS Spartan 2 FPGA
1 WDC65C02, ??>10MHz
1 2Mx8, 10nS Asynchronous SRAM
1 512Kx8, 150nS EEPROM
1 80MHz Crystal
1 640x480 pixel, 65K color TFT display


After a little experimentation, I'll pose some valid questions. Thanks again Dr. Jefyll!

[ElEctric_EyE]

Dr Jefyll wrote:

... I visualize your window as occupying (say) addresses 8000 to C000 hex.

Yes, $8000-$BFFF, the largest chunk of memory to decode easily, after zero-page and stack occupy $0000-$01FF, and my OS will be in $E000-$FFFF.... Although after looking at my circuit that was designed for a 44-pin CPLD, and the capabilities of a more adept 100-pin FPGA, I'm looking into using a larger chunk of memory...

ElEctric_EyE wrote:

...Right now I (think?) I understand a few things:

1) It is possible to address up to 16MB contiguously, and have a JSR/RTS cross page boundaries automatically?

Dr Jefyll wrote:

ElEctric_EyE wrote:

is this what you were referring to in your last paragraph of your last post?

No, I think you've raised a different topic.

Reading over your webpage (yet again, sill absorbing), I realized why I assumed this. On page 3 of your website: http://laughtonelectronics.com/arcana/B ... onPg3.html , you say "The programmer sees memory as a 16MB space...". From what you just said though in your previous post, I now understand you are bank switching a full 64K.

[ElEctric_EyE]

Dr Jefyll wrote:

... BTW you'll notice that none of this requires that your CPLD be clocked faster than the CPU clock rate...

Does this have something to do with the 2 "special" 2-cycle NOP's? $42 and $C2?

[Dr Jefyll]

ElEctric_EyE wrote:

Yes, $8000-$BFFF, the largest chunk of memory to decode easily...

Large and easy? You might consider a 32K window of 4000 to C000-1, decoded with simply A15 XOR A14. But as long as your CPLD has enough inputs, you can probably achieve any window you desire.

Quote:

your webpage [...] why I assumed this

Thank you; I'm grateful for any feedback regarding the web page. It's dangerously easy for me to accidentally imply something incorrect, or to overlook the forest and describe the trees instead, if you know what I mean! Please hit your browser's Reload button anytime you sit down to have a read, since you never know when I may be making revisions. Page 4 and 5 in particular have received a lot of attention lately -- including an explicit list of the undefined ops that you might wish to use!

[Dr Jefyll]

I wrote:

none of this requires that your CPLD be clocked faster than the CPU clock rate

Except for its video circuits, the KimKlone has pretty simple timing. One 65C02 bus cycle equals one fetch/execute from the microcode. It's pipelined, but still the clocks are only 5 MHz. There aren't any cogs and gears spinning inside at some insane rate; everything happens step by step at the same pace as what the 65C02 is doing. That applies to all ops, 42 and C2 included.

So, your CPLD doesn't need to be clocked faster than the CPU clock rate in order to use undefined opcodes. Of course, for your particular project there may be other signals/issues that call for faster clocking.

[Edit: other KK-like features you don't need include op-code substitution (because 1 new op is easier to map than 44) and microcode (because, with a less-than-64K window, you can rely on the address bus to tell you which bus accesses to treat as "window").]

I finally finished reading your (rather lengthy) Pulse Width Analyzer thread, and it all seems solid. I expect you'll soon have this windowing thing adapted to use an undefined opcode and humming like a top

[ElEctric_EyE]

I've finally wired everything up... The memory map is here:
viewtopic.php?t=1370&start=190 , I divided the 64K into 16K blocks to keep it simple.

The address decoding scheme I'm using is logical but isn't working. I'm failing to understand why, but I think I know, in part because I am bad with timing diag's. So let me thank anyone ahead of time for any help they can offer!

The first piece of software that runs, after a /Reset, copies $C000-$FFFF from EEPROM to SRAM using indirect indexed w/2 variables in zero page, so the multiplexers I'm using for A14-A20 at the 2M SRAM are switching from all zero's (for $0000-$3FFF) and all ones (for $C0000-$FFFF) when the program is running, and I'm thinking this is where my problem lies, but am unsure why.

Something is running, because I see the address and data lines switching on the 'scope...

Some facts:
The CPU is running at a conservative 2.5MHz since the access time for the EEPROM is 200ns.

Noise levels measure at different grounds are ~30mV, with everything running @3.3V. I don't think this is a problem. A concern though, since the
SRAM and display each take close to 300mA and everything is wirewrapped. Volts at each IC are in spec and clean looking.

I would like to know if I should pursue this idea further. Conceptually is it sound? or should I go on to another idea?

[8BIT]

Without seeing the actual decoding logic, or the init code, I will try to describe what I understand should be happening.

You say when the cpu addresses $C000-$FFFF, that SRAM is addressed to $1FC000-$1FFFFF.

Also, your ROM is addressed so that cput addresses $C000-$FFFF map to $7C000-$7FFFF in ROM.

Addresses in $0000-$3FFF always map to $000000-$003FFF in SRAM (for reads and writes presumably).

Do you have a control register in your CPLD that defines where reads come from vs. where writes go to, or are you using some other method?

After restart, your ROM must be mapped into Reads to address $C000-$FFFF or else your startup code will never run. If Reads are coming from SRAM, then you are most likely executing invalid code from uninitialized SRAM.

I would guess that if you can map writes into RAM while reading from ROM, that a simple

Code: Select all

     lda $c000
     sta $c000
     lda $c001
     sta $c001

sequence would move ROM to RAM, then writing to a control register to set Reads from RAM vs. ROM, would do what you want.

Can you provide a little more info?

Daryl

[Dr Jefyll]

ElEctric_EyE wrote:

The first piece of software that runs, after a /Reset, copies $C000-$FFFF from EEPROM to SRAM

There are different ways you could proceed, but if I were in your shoes I'd temporarily alter this block move code so it moves the first byte... then continues to move it again and again, forever. The cycle-by-cycle execution of the loop will be exactly the same every time, and that means your 'scope can produce a stable pattern and show you what's going on. Sync the 'scope to the R/W line (which goes low for one cycle only during the loop). Adjust the timing so you can see the entire loop displayed, 15 or 20 cycles -- whatever. Then go poking 'round on the other address and control lines, and verify they're doing what you THINK they're doing, particularly on the cycles that do the data read and write.

For me, this technique is Standard Operating Procedure with a new board I'm just "breathing life into" for the first time. (It's exciting -- I envy you!!) Anyway, the point is to take the (complex) boot process (which is hard to test), and isolate something within it that is SIMPLE to test.

It's true you may have a noise issue due to all the outputs switching on that multiplexer, but before you worry about slippery problems like that, make sure there's not anything basic wrong, like a logic error.

-- Jeff

[GARTHWILSON]

Quote:

The CPU is running at a conservative 2.5MHz since the access time for the EEPROM is 200ns.

Noise levels measure at different grounds are ~30mV, with everything running @3.3V. I don't think this is a problem. A concern though, since the
SRAM and display each take close to 300mA

Is the 200ns access time for the EEPROM spec'ed at 5V, even if it can run at 3V? It will definitely be slower at the lower voltage.

The 300mA does seem very, very high. Is it possible there's bus contention? My workbench computer which runs at 5MHz and the smaller one beside it running at 1MHz together take about 165mA, and some of that goes to RS-232 line drivers, op amps, and annunciator LEDs; and the clock oscillator for one of them is one of those very power-hungry crystal can oscillators that takes about 70mA.

[ElEctric_EyE]

8BIT wrote:

Without seeing the actual decoding logic, or the init code, I will try to describe what I understand should be happening.

You say when the cpu addresses $C000-$FFFF, that SRAM is addressed to $1FC000-$1FFFFF.

Also, your ROM is addressed so that cput addresses $C000-$FFFF map to $7C000-$7FFFF in ROM.

Addresses in $0000-$3FFF always map to $000000-$003FFF in SRAM (for reads and writes presumably)....
Daryl

Exactly right.

8BIT wrote:

...Do you have a control register in your CPLD that defines where reads come from vs. where writes go to, or are you using some other method?...
Daryl

For the EEPROM to SRAM copy, both are /CS'd and read simultaneously. The /OE controls which one the CPU sees. When writing, the EEPROM is unaffected. This portion I've completed successfully in the last stage... I can provide a small schematic for that portion as well, but it's pretty much like this:( viewtopic.php?t=1370&start=152 ), focus on the upper left. The 4 bit counter controls CPU /Res, SRAM /OE, & EEPROM /OE. I've slowed the count WAY down to make sure the CPU has plenty of time to copy 16K.

Garth, I'm using this EEPROM ( http://www.atmel.com/dyn/resources/prod ... oc0383.pdf ), everything running @3.3V right now, except the oscillator@5v. But that has p-pV of 2.5V. And this SRAM ( http://www.cypress.com/?docID=22950 ). Yes, 275mA!

Jeff, I'll try your idea, soon I think. Definately before I give up this attempt.

Here's the address decoding, only 1 MUX is shown for RBA14 (SRAM bank addr). According to the FPGA report the delay is <20ns from Ax to RBAx:


Copy software, pretty simple, it stays within boundaries. It should take about 2-3 sec? to complete @2.5Mhz:


Edit: added software pic & 4bit counter is upper left (correction), not upper right.

[Dr Jefyll]

Hmmm. "O2OUT" is Phase Two? If so, it looks to me as if the RBA14 signal (SRAM bank addr) can change state when phase 2 goes true. That ought not to be the case; address lines on the SRAM should already be stable by the start of phase 2 (ie, when CS activates). This might not be what's causing your difficulty, though.

Garth's suspicion about bus contention rings true... FWIW, my oscilloscope suggestion can reveal stuff like that, whether due to wiring errors or design issues. But I haven't described it in proper detail I'm afraid.
-- Jeff

[Dr Jefyll]

PS: Is there anything else on the data bus? (Maybe a 6522 or other IO device?) If so, I hope it only gets read-enabled within addresses 8000h-BFFFh. Looks as if you planned it that way.

[ElEctric_EyE]

Dr Jefyll wrote:

Hmmm. "O2OUT" is Phase Two? If so, it looks to me as if the RBA14 signal (SRAM bank addr) can change state when phase 2 goes true. That ought not to be the case; address lines on the SRAM should already be stable by the start of phase 2 (ie, when CS activates). This might not be what's causing your difficulty, though...

This type of address decoding I've used successfully, in my last stage. It ran 6MHz. I'm all ears for a better type of decoding though.

The only I/O device I have at this point is the display @$8000,$8001. As an experiment I took phase 2 out of the decoding for it, and it was receiving data, erroneously, but the display was changing.
I'm pretty sure everything is wired correctly, I've doube, triple checked my wiring, rechecked the QFP to BGA adapter pinouts, and ohmed out a few pins direct from each of the IC's (54-pin TSOPII SRAM & 100-pin VQFP FPGA) to the wirewrap pins. To make sure my soldering is 100%, I guess I should check all pins, but before I spent that much time...

So the idea of "Locking" 16K blocks, which is really bank switching, is not a problem? Even in the middle of a program?