kwin_overlay_subsurface — KWin per-frame compositor cost on parent + wp_subsurface

A scope-narrow campaign to identify and remove the per-frame cost in KWin's overlay-subsurface composite path on Mali-G52 / panfrost / KDE Plasma 6.6.4 Wayland (PineTab2, RK3568). Spun off from ohm_gl_fix on 2026-05-02 after Phase 3 re-measurement isolated KWin specifically as the residual drop source on the in-scope use case (Brave/Chromium 1080p30 H.264).

• Local dossier: ~/src/kwin_overlay_subsurface/ on boltzmann (canonical scope + reference list)
• Reference target: ohm — PineTab2, Rockchip RK3568 (4× Cortex-A55, Mali-G52 MP2, hantro G1/G2 VPU). Kernel 6.19.10-danctnix1-1-pinetab2. Mesa 26.0.5. KDE Plasma 6.6.4 Wayland.

• README — campaign overview — scope, hardware target, sources of truth, pre-MR checklist, non-upstreaming default.
• Worklist — phase plan with leading question — Phase 1 leading research question, Phase 2 source-archaeology reading list, Phases 3-8 outline.
• Phase 0 findings — handover from ohm_gl_fix — decisive cage vs KWin A/B, what's ruled out, A2 trajectory pattern, software stack at handover.
• Phase 1 lock — locked metric cells — one-line goal, four binding cells, reference floor, measurement protocol.
• Phase 2 source archaeology — Phase 1 leading question answer + Phase 2 stub — the new artefact. Short answer NO; Phase 4 design space narrows to the import-caching hypothesis only.
• Phase 3 measurement protocol — pre-registered hypothesis, falsification table, orchestration handshake — the new artefact. 9-run perf-record campaign locked 2026-05-02; H1: glEGLImageTargetTexture2DOES ≥ 20 % of kwin_wayland self-time in C1, ≥ 10 pp above cage.
• Phase 3 measurement findings — pass-1 LOCKED, H1 rejected, H1' emerges — the new artefact. Pass-1 verdict 2026-05-02: target symbol peaks at 0.15 % vs the 20 % threshold; per-frame Wayland-protocol dispatch dominates instead; Brave-vs-chromium-fourier inversion shows drops are not CPU-caused.
• metrics.csv — locked Phase 1 cells — schema + locked rows; raw CSV as attachment.
• ohm tooling revert log — every package / one-off installed on ohm during the campaign, for Phase 8 close-out.

• ohm_drm_info_2026-05-02.json (inlined) — full drm_info JSON dump from ohm (rockchip-drm plane format/modifier blob + connector/CRTC enumeration).
• ohm_modetest_planes_2026-05-02.txt (inlined) — modetest -p plane output on ohm, used to cross-check the JSON.
• metrics.csv (inlined) — raw CSV of locked binding cells.

• ohm_gl_fix — Mali-G52 EGL/dmabuf import efficiency campaign. Closed 2026-05-02 with the explicit re-scope into this campaign. Phase 0 of this campaign inherits its evidence, A2 trajectory, decisive A/B, and software stack.

Namespace landing created 2026-05-02 alongside the Phase 1 leading-question lock (commit 595a130 on master in the local dossier).