Tuesday, December 26, 2023

Forty Years of Personal Computing - MC6809 V2

MC6809 CPU card, version 2.
By March 1988, the MC6809E V1 card I designed in 1983 needed updates. I built an entirely new card with new features intended to run OS-9 more effectively. 

CPU

A MC6809 chip simplified things with the on-chip clock oscillator. The chip handled M.RDY without extra logic, and the rising edge of the Q clock did not need delay.

Memory

The MC6809E V1 card had no on-board RAM. There wasn't room. By 1988, a number of manufacturers had 32 KB static RAMs in 28-pin packages. 64 KB of memory is realized with a couple of chips. 

For the V2 board, I allowed for eight chips, totaling 256 KB of memory. This was a good compromise between cost and the space available. The memory is logically separate from the rest of the card -- decoding from the physical address and data bus, using appropriate buffers. In this way, the memory can be accessed by a bus master other than the CPU. It responds to physical addresses C0000-FDFFF or FEFFF, jumper selectable. For years, it held two chips -- 64 KB on the board -- with only 56 KB accessible. The six remaining chips were added recently, making 248 KB or 252 KB accessible. 

Buffering

20-pin bus driver chips reduced the chip count, even with two sets of bus drivers, one for the CPU, and one for the memory array.

Program ROM

The design allows for a much larger ROM. The MC6809E V1 card originally had two 2KB 2716-compatible sockets -- one for a ROM and another for ROM or RAM. To make swapping OS-9 and BBUG easier, I changed this to a single 4 KB 2732-compatible ROM socket

For the MC6809 V2 board, the ROM can be a 2764, 27128 or 27256-compatible device, holding 8 KB, 16 KB or 32 KB, respectively. The larger ROM permitted more OS-9 modules to reside there, if desired. 

As built, a 2764-compatible EPROM is used, containing a BBUG image in one 4 KB half, and the OS-9 ROM image in the other 4 KB half. A jumper selects which half is active. This is much easier than swapping chips to go between BBUG and OS-9.

Accessing the correct amount of the ROM requires clever decoding. 

Decoder

A hard-wired decoder would limit the flexibility of the system, and it would be complex and difficult to change. Rather than discrete logic, the decoder consists of a Cypress Semiconductor CY2C291 2Kx8 EPROM. This is a fast device with a 70ns access time. The CPU address lines A5 to A15 are connected directly to A0 to A10 on the chip. The decoder is enabled with the logical OR of E and Q, which asserts during three quarters of the memory cycle. This way, the eight data output pins can be used as decoder selects programmable on every 32-byte segment of memory.

Three select lines are used: one for bus access (including the on-board memory array), one for the program ROM, and one for the DAT. Each select line is pulled up to +5v. Placing a 0 bit in the decoder ROM data array makes the select line active for that 32-byte memory segment. 

Modifying the memory map becomes a simple matter of programming the decoder ROM. I programmed the following logical memory map:
  • 0000-EFFF - Bus
  • F000-F77F - Program ROM
  • F780-F7FF - Bus
  • F800-FFFF - Program ROM
  • FFE0-FFFF - DAT (writes only)
This configuration is compatible with the existing ROMs for BBUG and OS-9, which require I/O at E000-E07F. It has 4KB of program ROM, except for the hole at F780-F7FF. This hole deserves a bit of explanation. 

I/O Port Address Migration

BBUG occupies the top 2 KB of ROM. The OS-9 ROMs take up nearly 4KB. However OS9p2 doesn't use the last 128 bytes of that space. This unused space became an alternate location for the I/O ports. If the I/O ports moved from E000-E07F to F780-F7FF, the MC6809 could use RAM in the logical E block (E000-EFFF), for a total of 60 KB of RAM, up from 56 KB. 

Moving the I/O address requires motherboard decoder changes and software changes to the BBUG and OS-9 ROMs, as well as revision to Flex09 and OS-9 I/O configurations. The V2 board decoder ROM would work with the existing motherboard, or with the motherboard and ROMs altered for the new I/O addresses.

Larger ROM

Once the I/O addresses are moved, the decoder can be reprogrammed to allow for more ROM space. This opens the option of moving OS-9 modules into ROM. The decoder allows the lower limit of the ROM to be changed in 32-byte increments. This allows an OS-9 system to be entirely in ROM. OS-9 would start from the reset button without requiring a boot disk.

DAT

Back side of MC6809 V2 card.

The DAT configuration is similar to the MC6809E V1 board, with one important difference. In the SWTPc MP-09 board, as well as my V1 board, the outputs of the DAT are inverted on the lower four bits (A12-A15), but non-inverted on the higher four bits (S0-S3). 

This means that values programmed into the DAT must be one's complemented on the lower four bits (A12-A15), with the higher four bits (S0-S3) not complemented. 

For the V2 board, all eight bits of the DAT are inverted on the bus. Thus, the value programmed into the DAT is the one's compliment of the highest eight physical address bits (A12-A15, S0-S3). 

Which makes programming correct DAT values simpler, since the entire byte is complemented.

I introduced a hardware bug in the DAT decoder. More on this later.

Building

Rather than wirewrap, I opted to try something new. A technician from work gave me a couple of 3M Scotchflex Breadboarding kits. This breadboarding system was brilliant. Chip sockets connected to IDC pins. Wiring is accomplished by forcing wire-wrap wire between the IDC pins with a special tool. 

It is way  easier than wire-wrap, because there's no tedious cutting, stripping, threading and winding of wire. One lays the wire down and pushes it on to the pins. Wiring several connections in succession, such as with a bus, is a breeze. The results also look great. The IDC pins are low profile, so there's less chance of shorting a connection than with wire-wrap.

It's sad 3M discontinued this product. It was great. 3M has since re-used the Scotchflex brand on three other products.

Fixing the Bug

The MC6809 V2 board worked great. There were no wiring errors. I did find a problem with the DAT.

In the default BBUG and OS-9 configuration, the DAT is written once during reset and never touched. And that seemed to work just fine.

Then I started playing with an OS-9 driver called VDisk. It created a virtual disk from selected extended memory blocks. At the time, I had 56 KB of memory from the MC6809 V2 card, plus another 60 KB from the Digital Research Computers / Tanner card. That made possible a 60 KB virtual disk.

Every time I tried to access the virtual disk, the computer would crash. This took a while to track down. 

I eventually realized the new decoder did not take into account the clock cycle when accessing the DAT. Transients on the R/W* line early in the clock cycle could cause bad data to be written to the DAT. After I added the missing gate, the Disk driver worked perfectly. 

Usage

Like the MC6809E V1 board, this V2 board was exactly how I wanted it. There are only two jumpers. 

The jumper at the top edge of the board selects the 4KB portion of the EPROM. This makes it easy to switch between OS-9 and BBUG. No more hassle of changing out chips - just move a jumper.

The jumper in the middle of the board, just above the decoder ROM enables the FE000-FEFFF block of on-board memory. This would be installed once the motherboard I/O addresses are moved out of the E-block of memory and would allow 60 KB of RAM to be used.

Future

Moving the I/O addresses out of the E-block gains 4KB more usable memory for OS-9. Perhaps I'll try that in my retirement.

Another fun project would be to put a full OS-9 Level I system into ROM. Unfortunately, all of the essential modules take up just over 16 KB of memory, so the division doesn't fall on a natural 4 KB boundary. This might cause a conflict accessing extended memory with the DAT.  I'd also have to figure out how to program the decoder ROM. There are not many EPROM programmers that can program the Cypress Semiconductor CY2C291 devices, and I no longer have access to the ones I originally used. 

OS-9 Level II

This design works well for OS-9 Level I. To run OS-9 Level II, which allows each process to have a full 64 KB address space, requires more hardware. First, a second set of DAT memory chips allows the user and supervisors states to have separate memory maps. Second, a means of switching between those maps automatically -- like when servicing and returning from interrupts. Third, would require ROM to be accessible from an extended memory address, and then mapped into the supervisor space. 

Those requirements go beyond the scope of this design. Perhaps there's room for a V3 board. All of this assumes access to a copy of OS-9 Level II, which may be difficult to find.