BYTEC/16 rev. 2.1

Ten years is a long time even by homebrew-computer standards. Last time I wrote anything here, in December 2016, I had a box of rev. 2.0 boards sitting in a drawer and a list of modest goals, one of which was moving the sources to GitHub. I then actually did some hardware work, forgot most of it, and disappeared.

This is the catch-up post. BYTEC/16 now has a CPU rev. 2.1, the downloads page is updated, and the project finally has a public git repository.

Why another hardware revision

Rev. 2.0 added the MMU and, with it, a better fault mechanism. To restart a faulted instruction at the microinstruction that actually failed, I added FRET, sacrificed HALT for the opcode, and dropped the old PPC register from the microcode assembler. That was good enough to finish the 2.0 boards. It was less good as something an OS would have to live with: two return paths (IRET vs FRET), no previous-PC, and a halt instruction thrown away for no real reason.

Rev. 2.1 folds interrupt, fault and syscall return into a single IRET. The stacked frame now includes IR, MAR and MDR, and IRET restores the microcode step from that stack instead of going through a separate FRET opcode. PPC is back. HALT is back at opcode $01, where it used to live. The extra microcode bits that used to be FRET/MEMOP are now IRET and SSP – SSP being a supervisor-stack qualifier, so the return sequence can walk the stacked frame without fighting the user SP.

That is more than a microcode tweak. Boards #1 (MMU), #2 and #3 (control, sequencer, microcode ROM) and #6 and #7 (register files, including PPC) have 2.1 layouts and gerbers from Satland. ucasm grew the new fields, and the assembler knows HALT and IRET again. The MMU itself is unchanged from 2016: 4kB pages, 22-bit physical addresses, PRESENT / WRITABLE / IO flags. 2.1 is about making faults and interrupts something you can actually return from.

Most of this was done in 2024 and then sat in an uncommitted tree. I only sorted it out now.

GitHub

I promised a git repo nine years ago and then never mentioned it again. Here it is:

https://github.com/dpilawa/bytec

Eagle schematics and boards, gerbers, microcode, ucasm, binutils, the LCC 4.2 backend, monitor, demos, and the historic clutter that accumulates over sixteen years of a project like this. The zip snapshots for the website are not in git; those still live on the downloads page.

I am sharing updated schematics, Eagle files, gerbers, microcode, ucasm, the instruction set spreadsheet and binutils there today. The LCC patch is still the 2013 one – I never rebuilt it against a fresh lcc-4.2 tree, so there is nothing new to publish on that front.

What is not done

Honesty seems appropriate after a ten-year gap. The monitor still talks to the old 1.x page registers. The C runtime bank-switching is from before the MMU. The simulator and emulator do not know about 2.1. I have no bring-up write-up for this revision. Minix is as far away as it was in 2016.

If I pick this up again, the useful next steps are the same short list as last time, minus GitHub, which is done: bring-up of the 2.1 boards, a monitor that understands the updated MMU, and loading a program over the serial cable. Stay tuned, or don’t – I have learned not to put dates on these things.

Leave a Reply

  

  

  

Time limit is exhausted. Please reload the CAPTCHA.