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
91
audit-runs/phase-c7-keSetEvent/re-validation.md
Normal file
91
audit-runs/phase-c7-keSetEvent/re-validation.md
Normal file
@@ -0,0 +1,91 @@
|
||||
# Phase C+7 — re-validation
|
||||
|
||||
## Gate table
|
||||
|
||||
| gate | result |
|
||||
|---|---|
|
||||
| 1. cvar-OFF determinism (3 runs, 50M, `--stable-digest`) | **PASS** all 3 = `c6d895829b4611964978990ae1cb8a6a` (UNCHANGED from C+6½) |
|
||||
| 1b. cvar-OFF determinism (2 runs, 200M, `--stable-digest`) | **PASS** both = `8186841b38737f3cd1b3d2e91a35d108` (new 200M baseline) |
|
||||
| 2. Phase B `image_loaded_sha256` | **PASS** `ea8d160e9369328a5b922258a92113efb8d7ce3e1a5c12cc521e375985c91c18` matches baseline |
|
||||
| 3. Phase A main matched prefix ≥ 102158 | **PASS** 102158 (preserved — no main-chain regression, no #23 redux) |
|
||||
| 3b. Phase A sister chain tid=4→11 advances | **PASS** 5 → **9 (full alignment, no divergence in ours's 9-event window)** |
|
||||
| 3c. Phase A sister chain tid=7→2 advances | **PASS** 26 → **29 (full alignment, no divergence in canary's 29-event window)** |
|
||||
| 4. Both engines build clean | **PASS** ours: `cargo build --release` ok (pre-existing `walk_committed_regions` dead-code warning) |
|
||||
| 5. Phase A emitter determinism (det fields only) | **PASS** 2 runs `90fb28202b70cb43a63def7a2f8b470d` byte-identical |
|
||||
| 6. Unit tests | **PASS** 146 → **152** (6 new, 0 modified, 0 regressed) |
|
||||
|
||||
## Gate 1 — 50M determinism
|
||||
|
||||
```
|
||||
c6d895829b4611964978990ae1cb8a6a digest-cvaroff-1.json
|
||||
c6d895829b4611964978990ae1cb8a6a digest-cvaroff-2.json
|
||||
c6d895829b4611964978990ae1cb8a6a digest-cvaroff-3.json
|
||||
```
|
||||
|
||||
Digest is **unchanged** from C+6½ baseline. This is expected:
|
||||
`--stable-digest` reports `(instructions, imports, draws, swaps,
|
||||
unique_render_targets, shader_blobs_live, texture_cache_entries)`,
|
||||
none of which are sensitive to KeSetEvent return-value within the 50M
|
||||
horizon. Imports stay at 40470 (the KeSetEvent change moves no new
|
||||
import call; only the return value of existing calls flips from 0 to
|
||||
1). Draws stay at 0, swaps stay at 1.
|
||||
|
||||
## Gate 1b — 200M stability
|
||||
|
||||
```
|
||||
8186841b38737f3cd1b3d2e91a35d108 digest-cvaroff-200M-1.json
|
||||
8186841b38737f3cd1b3d2e91a35d108 digest-cvaroff-200M-2.json
|
||||
```
|
||||
|
||||
200M run is byte-deterministic across 2 runs. No 200M C+6½ baseline
|
||||
to compare against (none was captured), but the new digest is itself
|
||||
a regression marker for any future fix. Field values at 200M:
|
||||
imports=40470, draws=0, swaps=1 — same plateau as 50M.
|
||||
|
||||
## Gate 3 — Phase A diff (all chains)
|
||||
|
||||
| chain | C+6½ XAM | C+7 | Δ |
|
||||
|---|---|---|---|
|
||||
| canary tid=6 → ours tid=1 (main) | 102158 | 102158 | 0 (preserved) |
|
||||
| canary tid=4 → ours tid=11 | 5 | **9** | **+4 (all ours events match)** |
|
||||
| canary tid=7 → ours tid=2 | 26 | **29** | **+3 (all canary events match)** |
|
||||
| canary tid=12 → ours tid=7 | 2 | 2 | 0 (different bug — KeWaitForSingleObject=258 vs 0) |
|
||||
| canary tid=14 → ours tid=9 | 39 | 39 | 0 (different bug — XAudio vs RtlEnterCS) |
|
||||
| canary tid=15 → ours tid=10 | (no div) | (no div) | 0 |
|
||||
|
||||
Two sister chains advanced exactly as predicted by the C+6½ memory
|
||||
note. Main chain (XamTaskCloseHandle at 102158) is **explicitly
|
||||
unchanged** — the KeSetEvent fix doesn't touch the XamTaskHandle
|
||||
path; that divergence remains the next session's target.
|
||||
|
||||
## Gate 5 — emitter determinism
|
||||
|
||||
Phase A emitter run twice over the same boot, with all
|
||||
non-deterministic fields (`host_ns`, `guest_cycle`, `engine`,
|
||||
`deterministic`) stripped:
|
||||
|
||||
```
|
||||
90fb28202b70cb43a63def7a2f8b470d ours.jsonl
|
||||
90fb28202b70cb43a63def7a2f8b470d ours-determ.jsonl
|
||||
```
|
||||
|
||||
Byte-identical det fields. New emitter-det baseline replaces
|
||||
C+6½'s `11a07772…` — the return-value field is included in the
|
||||
deterministic schema, and KeSetEvent's contribution flipped from 0
|
||||
to 1 on every emit.
|
||||
|
||||
## Gate 6 — unit tests
|
||||
|
||||
`cargo test -p xenia-kernel --release` — **152/152 pass**. New 6
|
||||
tests:
|
||||
|
||||
* `ke_set_event_returns_constant_one_on_unsignaled_auto_reset`
|
||||
* `ke_set_event_returns_constant_one_on_already_signaled_manual_reset`
|
||||
* `nt_set_event_null_prev_ptr_returns_status_success_no_write`
|
||||
* `nt_set_event_valid_prev_ptr_writes_constant_one_and_returns_success`
|
||||
* `nt_set_event_on_signaled_event_writes_one`
|
||||
* `ke_set_event_post_fix_still_wakes_waiter`
|
||||
|
||||
All existing 146 tests still pass — no regressions from the
|
||||
return-value flip (other test bodies don't assert on the return
|
||||
value of `ke_set_event` / `nt_set_event`).
|
||||
Reference in New Issue
Block a user