Skip to content

TASKS.md

Latest commit

 

History

History
230 lines (175 loc) · 5.45 KB

TASKS.md

File metadata and controls

230 lines (175 loc) · 5.45 KB

Tasks

Current Stable Baseline

  • ./runtests.sh must stay green.
  • cd test && node runner.js must stay 231/233.
  • CALLM and RTM remain explicitly deferred and untouched.

Current CPU Completion Block

The next CPU-facing block is now explicit.

Immediate next CPU blockers:

  1. MOVEM
    • replace the current stub with real register-mask semantics
    • cover register-to-memory and memory-to-register paths properly
  2. NBCD
    • replace the current stub with real decimal negate semantics
    • verify X, C, Z, and result behavior
  3. ABCD
    • replace the current stub with real decimal add semantics
    • cover register and predecrement-memory forms
  4. memory rotates
    • ROL memory
    • ROR memory
    • ROXL memory
    • ROXR memory

Acceptance for this CPU block:

  • no semantic stubs remain for the above instructions
  • targeted CPU tests are added or strengthened for each one
  • ./runtests.sh stays green
  • cd test && node runner.js stays 231/233
  • CALLM / RTM remain deferred

Current Machine / Monitor Checkpoint

The small j68 software environment has reached a clean checkpoint:

  • ROM monitor is stable
  • source/ + loadasm workflow is stable
  • host-built binaries in work/programs/ build into save/
  • reusable guest-side output helpers now exist
  • guest-side polling input helpers now exist
  • a minimal cooperative guest runtime now exists
  • visible formatted output works from real 68k binaries
  • interactive guest input works from the real launcher

Current reusable guest-side helper layer:

  • work/programs/lib/console.inc
    • putc
    • puts
    • newline
    • puthex32
    • puts_hex32
  • work/programs/lib/input.inc
    • getc
    • put_prompt
    • readline
  • work/programs/lib/task.inc
    • scheduler_start
    • yield
    • task_exit

Current visible proof binaries:

  • hello_uart.bin
  • puthex.bin
  • print_pi16.bin
  • print_status.bin
  • ram_checksum.bin
  • pi16_nilakantha.bin
  • echo_line.bin
  • coop_demo.bin

This is now a proper little software environment, not just isolated probes.

Next Subsystem Boundary

The next clean subsystem is input polish and then interrupt-driven input, not CALLM/RTM.

Current completed order:

  1. guest-side input helpers
    • getc
    • prompt helper
    • simple line read
    • polling against UART status/data
  2. one interactive demo binary
    • echo_line.bin
  3. minimal cooperative multitasking
    • fixed two-task setup
    • explicit yield
    • coop_demo.bin

Recommended next order:

  1. input niceties
    • backspace handling
    • bounded line editing polish
  2. interrupt-driven input
    • cleaner IPL exposure first
    • no JS-side scheduler
  3. richer cooperative demos
    • still no preemption

Immediate acceptance for the next block:

  • guest input remains polling-clean in the launcher
  • a richer interactive demo works end to end
  • cooperative runtime remains guest-side only

Current Architectural Gap

The machine has a useful memory API and device map, but it is still growing from a memory map into a 68k-style external bus model.

The most important missing semantics are:

  • FC[2:0]
  • IPL[2:0]
  • BERR

Near-term impact:

  • FC[2:0]
    • useful for fuller machine fidelity
    • not blocking the first polling-based input path
  • IPL[2:0]
    • should be exposed more cleanly before interrupt-driven input work
  • BERR
    • important for machine fault semantics
    • not blocking the first happy-path input helpers

Related lower-priority concepts that should remain deferred until needed:

  • AS
  • UDS/LDS
  • DTACK
  • VPA/VMA
  • RESET/HALT
  • bus arbitration (BR/BG/BGACK)

Bus Transaction Model

Keep the public API:

  • read8/read16/read32
  • write8/write16/write32

Under that, model transactions with:

  • addr
  • size
  • write
  • value
  • fc
  • ipl
  • kind
  • result fields such as:
    • value
    • berr

The bus must distinguish at least:

  • user data
  • user program
  • supervisor data
  • supervisor program
  • instruction fetch vs data access

Priority Order

  1. Add transaction metadata under the current bus API.
  2. Expose resolved current IPL cleanly.
  3. Add machine-visible BERR result path.
  4. Leave timing and handshake details for later.

For the next implementation block, input may start in polling mode without waiting for more IPL work. The IPL exposure work becomes more important when moving from polling input to interrupt-driven input.

Acceptance For This Slice

  • existing tests still pass
  • add focused system tests for:
    • transaction carries fc
    • unresolved access produces berr
    • highest active IPL wins

Explicitly Not In Scope

  • DTACK timing
  • AS/UDS/LDS pin modeling
  • DMA
  • storage/sound expansion
  • more device families
  • CALLM/RTM
  • FPU work

Guest Monitor Integration Checkpoint

68K-Monitor.X68 guest bring-up is paused at the correct boundary:

  • machine map and UART remap are no longer the first blocker
  • guest RAM/layout rebase is no longer the first blocker
  • startup reaches real guest code

Current exposed blocker:

FAULT PC=000917FC OP=C240

This is AND.W D0,D1 reached from the adapted guest monitor.

Required next CPU-facing work:

  1. build a tiny standalone repro around AND.W D0,D1
  2. determine whether the problem is:
    • decode missing
    • execute missing
    • PC advance corruption
    • flags or register writeback
  3. only return to the guest monitor after that path is understood

Guest adaptation must stay paused here to avoid turning porting into repeated workarounds for core gaps.