fourier — multi-layer V4L2 hardware-decode unlock for mainline ARM Wayland

A patch campaign that started as a single Chromium V4L2 dispatch unlock and expanded — every time a layer was patched, the next layer's bug surfaced — into 4 sibling patch series + 1 kernel RFC + 3 upstream submission drafts that together make hardware video decode actually work on mainline-Linux Wayland on Rockchip + Mali-class ARM hardware.

Public face: github.com/marfrit/fourier — sibling subdirs chromium-fourier/, qt6-base-fourier/, kwin-fourier/, docs/, tools/
Local Arch packaging + investigation notes: gitea/marfrit-packages under arch/{chromium,qt6-base,kwin,firefox}-fourier/, kernel/vb2-dma-resv-rfc/, upstream-submissions/
Reference target: ohm (PineTab2 / RK3566 / Mali-G52 / panfrost / mainline 6.19.10 / KDE Plasma 6.6.4 Wayland)

Status — paused 2026-04-30

Project paused pending the GL/EGL pipeline campaign baseline. The browser-side ship-gate (1080p30 YouTube ≤ 30 % CPU, no drops) is gated on the Mali-side EGLImage import cost being addressed; running browser measurements before that lands produces known-failing results that contaminate the 8-phase loop.

Freeze-state record: Phase 5 review handover, 2026-04-30.

Live tasks that continue independently:

chromium-fourier build — finish the build + publish; package ships, measurement deferred.
vb2 dma_resv RFC v2 — kernel-side, separate cadence.

Re-entry condition: GL fix campaign surfaces something useful that changes the GL-consumer baseline.

Status snapshot — 2026-04-30

Series	Patches	State	Validated where	Notes
chromium-fourier	4	end-to-end smooth playback	ohm	bbb_1080p30_h264.mp4: ~81 % chrome CPU vs ~131 % pre-patch baseline (~3.8× reduction)
qt6-base-fourier	3	shipped + upstream-targetable	ohm + fresnel	pure spec-correctness GLES3 fix; zero `GL_INVALID_VALUE` in journal post-relogin
kwin-fourier	1 shipped (0001-skip-wait)	shipped, MR 9157 closed	ohm	reverted to 0001 after MR review found 0002 technically wrong; goes no-op once vb2-dma-resv RFC lands
vb2-dma-resv RFC	3 sent 2026-04-29	v1 reviewed (Nicolas + Christian); v2 sketch backlogged	—	not a NACK; attach point moves to `device_run`, opt-in flag added, lockdep annotated for v2
firefox-fourier 150.0.1	4	partial — gates 1+4 verified, 2+3 corrected today, rebuild pending	fresnel	root cause identified via `MOZ_LOG`: must probe by `h264_v4l2request` codec name first
upstream submission drafts	3	written, awaiting validation gates	—	bugreports.qt.io, invent.kde.org MR, linux-media RFC

Sub-pages

chromium-fourier — what we get + the rug-pull on Brave (2026-04-26 capture; predates the 4-patch end-to-end win)
Patch series — the four sibling series in detail
vb2-dma-resv kernel RFC — the architectural correction
Per-host status — ohm / fresnel / boltzmann / CT220 / ampere
Journey — chronological discovery (chromium → qt6 → kwin → vb2)
Upstream submission drafts — Qt, KDE, linux-media

Sources

Public

github.com/marfrit/fourier
- chromium-fourier/ — 4 patches
- qt6-base-fourier/ — 3 patches
- kwin-fourier/ — 2 patches
- docs/dmabuf-zero-copy.md — panfrost external_only finding + cascade
- tools/ — capture / measurement helpers

Local (gitea/marfrit-packages)

arch/chromium-fourier/ — PKGBUILD + patches + NEXT.md campaign chronology + KWIN_PIVOT.md discovery story
arch/qt6-base-fourier/
arch/kwin-fourier/
arch/firefox-fourier/
kernel/vb2-dma-resv-rfc/ — 3-patch kernel RFC
upstream-submissions/qt6-base-fourier/qt-bug-report.md
upstream-submissions/kwin-fourier/kde-mr-body.md
upstream-submissions/vb2-dma-resv/lkml-submission-notes.md

Page index created 2026-04-28 from a fourier-campaign sync. The chromium-fourier sub-page predates the end-to-end win and reflects the state on 2026-04-26; the patch_series page carries the current 4-patch shape.

Zettelkasten

Table of Contents