Files

MechaCat02 ef93a4fa14 handoff: VSync/event-wedge fixes + iterate 2.A–2.BC research notes

Source changes (dormant parity infra, retained from iterate 2.AI/2.AO):
- xenia-kernel/exports.rs: nt_create_event manual_reset polarity +
  related event wiring
- xenia-gpu/mmio_region.rs: D1MODE_VBLANK_VLINE_STATUS hardcode parity

Also lands the audit-runs/ analysis notes (.md/.txt/.json digests) for the
iterate 2.x VSync/0x10e8/0x1004 wedge investigation. Raw trace dumps
(.jsonl/.gz/.csv/.stdout) and agent worktrees (.claude/) are gitignored as
regenerable local artifacts — see memory + HANDOFF for the running findings.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

2026-06-05 07:19:08 +02:00

2.6 KiB

Raw Blame History

Phase C+23 broad-impact assessment (2026-05-18)

Resolved

D-NEW-2 (C+17 catalog) — KeWait timeout wait.begin.timeout_ns on canary tid=12 → ours tid=7 idx=3. Was 429466729600 (broken positive sign-extension); now -30000000 matching canary exactly. Sister chain advances 3 → 4.

Advanced

addis opcode semantics across the entire CPU interpreter. Any place in any guest binary using lis/addis with a negative immediate that flows into a 64-bit std, mr, orx, or anything else that propagates the full GPR value will now produce canary-equivalent behavior. The KeWait timeout was just the first observable symptom in Sylpheed's cold-vs-cold trajectory.

Persisted (no change)

Main matched-prefix tid=6→1 = 104,607 (the C+22 scheduler- determinism asymmetry is unaffected by this fix).
Sister chains tid=4→11 (11), tid=7→2 (32), tid=14→9 (41), tid=15→10 (16).
Phase B image_canonical_sha256 = ea8d160e9369328a5b922258a92113ef b8d7ce3e1a5c12cc521e375985c91c18.
Kernel tests 204 (unchanged — no kernel-side code change).

NEW

D-NEW-2.1 (idx=4 sister chain canary tid=12 → ours tid=7): KeWaitForSingleObject return value canary=258 (TIMEOUT) vs ours=0 (SUCCESS). Same C+20/C+22 family — scheduler-determinism; ours's monolithic-thread runner allows the wait to return SUCCESS where canary's contended scheduler lets the 30 ms timeout elapse with no signaler. Out of scope for this phase; same parallel scheduler-determinism track recommended in C+22.
CPU correctness probe: opportunity for a low-key audit of every "32-bit ABI defensive truncation" in xenia-cpu/src/interpreter.rs for other ops that legitimately need 64-bit sign-extension at the producer rather than truncation. Quick mental scan suggests addis was likely the only such case; addi already extends via simm16 as i64 as u64, ori/oris are pure 64-bit OR. Not a blocking concern.

Test count delta

crate	pre-C+23	post-C+23	delta
xenia-cpu	288	291	+3
xenia-kernel	204	204	0

Cold-stable digest delta

baseline	pre-C+23	post-C+23
ours-cold det-fields	e1dfcb1559f987b35012a7f2dc6d93f5	23cf4c4cbf61a577caa4118ab2308ba6
Phase B image_canonical	ea8d160e9369…	ea8d160e9369… (UNCHANGED)

3× ours-cold runs all yield the new digest. Determinism preserved.

2.6 KiB Raw Blame History Unescape Escape

Phase C+23 broad-impact assessment (2026-05-18)

Resolved

Advanced

Persisted (no change)

NEW

Test count delta

Cold-stable digest delta

2.6 KiB

Raw Blame History