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>
This commit is contained in:
MechaCat02
2026-06-05 07:19:08 +02:00
parent acd1656753
commit ef93a4fa14
620 changed files with 108303 additions and 1 deletions

View File

@@ -0,0 +1,135 @@
# Phase C+11.1 — re-validation (HARD GATE)
All gates listed in the original plan executed and passed.
## Gate 1 — Determinism (3× cold reproducible)
3 sequential `xrs-c11p1 check -n 50000000 --stable-digest` runs
with `find ~/.local/share/xenia-rs/cache -mindepth 1 -delete`
before each:
| run | digest md5 |
|---|---|
| 1 | `ad4f74ee324fdedb0bfdd4cc4c6468e9` |
| 2 | `ad4f74ee324fdedb0bfdd4cc4c6468e9` |
| 3 | `ad4f74ee324fdedb0bfdd4cc4c6468e9` |
PASS. New cold baseline. Differs from C+11's `b8fa0e0460359a4f660adb7605e053de`
because the disp=1 + opts=0x7 cold probes now miss cleanly instead
of producing spurious mkdir+SUCCESS — a behavior change is expected
when correcting a behavior bug.
Digest JSON:
```
{
"instructions": 50000002,
"imports": 40447,
"unimpl": 0,
"draws": 0,
"swaps": 1,
"unique_render_targets": 0,
"shader_blobs_live": 0,
"texture_cache_entries": 0
}
```
## Gate 2 — Warm digest still deterministic
`xrs-c11p1 check -n 50000000 --stable-digest` with the post-cold-run
cache state already in place: `ad4f74ee324fdedb0bfdd4cc4c6468e9`
identical to cold. The C+11 era's cold-vs-warm gap (`b8fa0e04…` vs
`cee0849a…`) has collapsed: with disp=1 returning NOT_FOUND
cleanly, the first cache probe always misses and the rest of the
boot path follows the same trajectory whether the cache is empty
or warm.
## Gate 3 — XENIA_CACHE_WIPE=1 opt-out still works
`XENIA_CACHE_WIPE=1 xrs-c11p1 check -n 50000000 --stable-digest`:
`ad4f74ee324fdedb0bfdd4cc4c6468e9` — identical. The AUDIT-038 wipe
mode is preserved; under the new fix it produces the same digest
as the default-persistent mode (because the boot path is identical
when the cache is empty in either case).
## Gate 4 — Phase B `image_loaded_sha256` unchanged
`xrs-c11p1 exec --phase-b-snapshot-dir <dir> --phase-b-snapshot-and-exit`:
```
phase-b-snap/ours/config.json:
"image_loaded_sha256": "ea8d160e9369328a5b922258a92113efb8d7ce3e1a5c12cc521e375985c91c18"
```
PASS — matches the prior Phase B baseline (`ea8d160e…`). The fix
does not perturb the XEX image or its load layout.
## Gate 5 — Cargo tests
`cargo test -p xenia-kernel --release`: **172 passed; 0 failed**.
Baseline pre-C+11.1 was 170 tests. Net +2 tests:
* `cache_open_directory_on_missing_path_returns_not_found` — pins
the cold-boot disp=1 + opts=0x7 behavior across `access`, `ignore`,
`recent`. Asserts NOT_FOUND status, handle=0, and no host
side-effect.
* `cache_disp5_after_disp1_miss_creates_file` — pins the
two-call sequence canary actually executes (disp=1 NOT_FOUND →
disp=5 FILE created).
## Gate 6 — Phase A cold-vs-cold matched-prefix table
See [cold-vs-cold-baseline.md](cold-vs-cold-baseline.md). Headline
main matched-prefix = **102404** (cold-vs-cold canonical; replaces
the deprecated `+1521 to 103925` warm-asymmetric value).
No regression on any of the 6 sister chains:
| canary_tid | ours_tid | matched | first_divergence | regressed? |
|---|---|---|---|---|
| 6 | 1 | 102404 | 102404 | no (= prior cold value) |
| 4 | 11 | 9 | — | no |
| 7 | 2 | 29 | — | no |
| 12 | 7 | 2 | 2 | no (pre-existing, unrelated) |
| 14 | 9 | 39 | 39 | no (pre-existing, unrelated) |
| 15 | 10 | 15 | — | no |
## Gate 7 — Cache layout matches canary structurally
Direct `stat` comparison:
```
~/.local/share/xenia-rs/cache/ ~/.local/share/Xenia/cache/
69d8e45c/ (dir) 69d8e45c/ (dir)
aab216c3/ (dir) aab216c3/ (dir)
d4ea4615/ (dir) d4ea4615/ (dir)
87719002/ (dir, canary-only — not
reached in 50M ours boot)
access (file, 72 B) access (file, 240 B)
recent (file, 48 B) recent (file, 48 B → 160 B over
many warm boots)
```
The `access`/`recent` dir-vs-file bug is **resolved**. ours's file
sizes are smaller because they reflect one boot's worth of manifest
records; canary's larger sizes reflect many warm boots. Structural
layout (file vs directory, no spurious `ignore/`) matches.
## Gate 8 — Build clean
`cargo build --release` from `xenia-rs/` succeeds with no errors.
One unrelated `dead_code` warning on `walk_committed_regions` in
`phase_b_snapshot.rs` is pre-existing and unaffected.
## Snapshots
* `canary-cache-pre-wipe.tar.gz` — canary's 23-file oracle, preserved
for any future warm-cache experiments.
* `canary-cache-post-cold.tar.gz` — identical to pre-wipe (canary
did not populate its cache in the 120 s cold-run window).
* `ours-cache-post-cold.tar.gz` — ours's post-cold cache state
showing the file-not-dir layout the fix produces.
## Summary
ALL 8 gates PASS. The fix is landable.