fix(bitnet): default local_we low before the DMA case (Closes #2006)

[gHashTag]

Closes #2006

The defect was live on master verbatim

#2006 reports itself as fixed. Its fixes landed on a branch that never merged (no PR, in any state), so the defect is live. Confirmed against freshly-fetched origin/master (ee494b990b795267ce1c955d6009819b57ac55e8) before writing anything — bootstrap/src/bitnet_dma.rs:165:

    s.push_str("        end else case (state)\n");

No default for local_we. The only clears are :193 (end else local_we <= 1'b0;, inside READ_DATA) and :218 (DONE_ST).

The change

Emit end else begin / local_we <= 1'b0; / case (state), closing with endcase + end. Any state that does not explicitly drive the strobe now leaves it low. The case arms are byte-identical — no port, state, or other signal changed.

Honest scope: this fix is behaviourally latent today, and that was measured

A differential Icarus simulation running the master emitter and the fixed emitter side by side in one testbench, over a 4-beat read with a well-behaved AXI slave:

--- OLD emitter (origin/master) ---        --- NEW emitter (fixed) ---
  local writes = 4                          local writes = 4
    write[0] addr=1 data=d0d0...0000          write[0] addr=1 data=d0d0...0000
    write[3] addr=4 data=d0d0...0003          write[3] addr=4 data=d0d0...0003
  mem[0] = ffffffffffffffff                 mem[0] = ffffffffffffffff
  local_we high outside READ_DATA = 1       local_we high outside READ_DATA = 1
CONTROL: old AXI w-beats = 3   new AXI w-beats = 3

Identical. The reason is structural: local_we is only ever raised in READ_DATA, and the sole exit from READ_DATA is DONE_ST, which clears it. There is no path back to READ_ADDR in the single-burst FSM, so the missing default is unreachable today. The READ_ADDR-between-bursts window named in the issue only opens once #1970 derives the burst loop.

Claiming a behavioural win here would have been false, so it is not claimed. This lands the defensive shape first so the multi-burst change cannot silently reintroduce a stuck strobe.

Mutant evidence

Guard 1 — local_we_defaults_low_before_the_case. Mutant: delete only the emitted default line (begin/end stay balanced). Fails with:

`local_we` must be defaulted low between the reset block and `case (state)`, so the
strobe is a one-cycle pulse instead of a level that persists through states which never
drive it (READ_ADDR between bursts, IDLE, WRITE_ADDR, WRITE_DATA). Text found between
`end else` and `case (state)` was:
"end else begin\n            "
test result: FAILED. 16 passed; 1 failed

The message prints the offending span — that is what shows the assertion read the right text. Reverting: 17 passed, 0 failed.

The anchor is load-bearing, and that was measured. Adding a naive v.contains("local_we <= 1'b0;") to the defective master emitter:

test tests::naive_unanchored_guard_ON_DEFECTIVE_EMITTER ... ok
test tests::anchored_guard_ON_DEFECTIVE_EMITTER ... FAILED

The naive check passes on the defective emitter because DONE_ST already emits that exact string with single spacing (and the reset block emits a padded variant). An unanchored guard here would have been vacuous — the same trap that had to be repaired hours after landing elsewhere in this campaign.

Guard 2 — always_block_begin_end_balanced, planted separately. One mutant per guard, not one per file. Mutant: delete the closing end. Fails with:

unbalanced begin/end in emitted module: 15 `begin` vs 14 `end` (excluding `endcase`/`endmodule`)
test result: FAILED. 16 passed; 1 failed

Guard 1 still passes under this mutant, so neither guard masks the other.

The emitted Verilog parses: iverilog -g2005 accepts the fixed module.

Pre-existing gap, recorded not papered over

These tests are not executed by CI. No workflow runs cargo test -p t27c; corpus-ratchet.yml records that the step was removed by #2292 after going red on master. They were run with rustc --test on the module, which is self-contained. That gap is pre-existing and out of scope here.

Adjacent problems seen and deliberately NOT fixed here

• Wave Loop 579 — the write-pairing shape, enumerated across every port #2003 — :189-190 register local_we <= 1'b1 and local_addr <= local_addr + 12'd1 in the same non-blocking group, so beat 0's data lands at address 1 and slot 0 is never written. Visible in the trace above (mem[0] = ffffffffffffffff on both columns). Filed separately, fixed in its own PR.
• Wave Loop 557 — dma_controller abandoned bursts and advanced on ready-without-valid #1970 — arlen/awlen hardwired to 8'hFF, and READ_DATA exits on m_axi_rlast || bytes_remaining <= 32'd8.

Nothing was weakened: no || true, no continue-on-error, no existing assertion relaxed, no case dropped.

[github-actions]

PR Dashboard

Generated at: 2026-08-21 16:06:37 UTC

• fix(bitnet): default local_we low before the DMA case (Closes #2006) #2344 fix(bitnet): default local_we low before the DMA case (Closes #2006) | Created: 2026-08-21 | Updated: 2026-08-21 | 10 failing / 26 total
• fix(bitnet): make the DMA local write strobe a pulse, not a level #2339 fix(bitnet): make the DMA local write strobe a pulse, not a level | Created: 2026-08-21 | Updated: 2026-08-21 | 5 failing / 33 total
• corpus: freeze the damage, patch one class at a time, and say where one rule stops #2161 corpus: freeze the damage, patch one class at a time, and say where one rule stops | Created: 2026-08-14 | Updated: 2026-08-21 | 0 failing / 9 total
• Unbreak FPGA CI and correct two unbacked hardware claims #2081 Unbreak FPGA CI and correct two unbacked hardware claims | Created: 2026-08-11 | Updated: 2026-08-20 | 5 failing / 27 total

Summary

Status	Count
Total Open PRs	4
PRs with Failing Checks	2
PRs with All Checks Green	2
READY	1
FAILING	2
PENDING	0

Seal Status

• ⚠️ STALE -- sha256(compiler.rs)=65f033d04125 != manifest seal=87e5cbd3ad94.
  The committed NMSE numbers were certified against an older compiler.rs.
  Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

[github-actions]

📓 NotebookLM Notebook linked to this PR

• Notebook ID: b83263109fb055dc
• URL: https://notebooklm.google.com/notebook/b83263109fb055dc

This notebook contains session context, decisions, and artifacts for this work.

[github-actions]

PR Dashboard

Generated at: 2026-08-21 16:10:23 UTC

• fix(bitnet): default local_we low before the DMA case (Closes #2006) #2344 fix(bitnet): default local_we low before the DMA case (Closes #2006) | Created: 2026-08-21 | Updated: 2026-08-21 | 18 failing / 24 total
• fix(bitnet): make the DMA local write strobe a pulse, not a level #2339 fix(bitnet): make the DMA local write strobe a pulse, not a level | Created: 2026-08-21 | Updated: 2026-08-21 | 5 failing / 33 total
• corpus: freeze the damage, patch one class at a time, and say where one rule stops #2161 corpus: freeze the damage, patch one class at a time, and say where one rule stops | Created: 2026-08-14 | Updated: 2026-08-21 | 0 failing / 9 total
• Unbreak FPGA CI and correct two unbacked hardware claims #2081 Unbreak FPGA CI and correct two unbacked hardware claims | Created: 2026-08-11 | Updated: 2026-08-20 | 5 failing / 27 total

Summary

Status	Count
Total Open PRs	4
PRs with Failing Checks	2
PRs with All Checks Green	2
READY	1
FAILING	2
PENDING	0

Seal Status

• ⚠️ STALE -- sha256(compiler.rs)=65f033d04125 != manifest seal=87e5cbd3ad94.
  The committed NMSE numbers were certified against an older compiler.rs.
  Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

[github-actions]

📓 NotebookLM Notebook linked to this PR

• Notebook ID: b83263109fb055dc
• URL: https://notebooklm.google.com/notebook/b83263109fb055dc

This notebook contains session context, decisions, and artifacts for this work.