 Markus Fritsche
|
96e2d439c9
|
iter6 v3 plan: apply Sonnet round-3 amendments
6 targeted fixes, no rewrite:
1. Add DEBUG_ATOMIC_SLEEP + DEBUG_LIST explicitly to config delta
2. Drop PAGE_POISONING_NO_SANITY (keep the sanity check)
3. Fix mpv hwdec flag (vaapi-copy is SW! use v4l2m2m or rkmpp).
Extend monitoring window 2min → 20min (KASAN slows runtime 3-8x).
Add glmark2 + ffmpeg-null concurrent option.
4. Add R11: /boot/firmware headroom pre-flight (KASAN_OUTLINE +15-25% Image size).
5. Replace serial fallback with netconsole (no clip work needed).
6. Add H4 isolation exit path: headless-only DRM ffmpeg loop if
compositor test silent — isolates panthor from vb2.
All round-3 ACCEPT/AMENDs applied. v3 ready for execution.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
|
2026-05-16 19:28:19 +00:00 |
|
|
Markus Fritsche
|
2de6870cb4
|
iter6 v3 plan: KASAN+KCSAN delta on lockdep substrate
Pending architect delta-review before any execution.
Adds: KASAN_GENERIC + KCSAN + DEBUG_PREEMPT + DEBUG_OBJECTS +
SLUB_DEBUG_ON + PAGE_POISONING. CONFIG_LOCALVERSION=-lockdep-kasan
so separate from existing lockdep modules.
Smoke test extended: after headless ffmpeg passes, manually log in
to plasma + run mpv hwdec=vaapi-copy to reproduce v2's compositor
wedge condition. Monitor pstore + journal for KASAN/KCSAN reports.
Risk additions: KASAN/KCSAN boot slowness, KASAN shadow region,
KASAN false positives, GPU repro requires physical login.
Exit conditions: success (sanitizer report fires) / partial (still
silent — switch to serial) / failure (KASAN kernel won't boot).
~75-90 min build + ~45 min test cycle for 0004 verdict.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
|
2026-05-16 19:26:55 +00:00 |
|