test0r
713a856cdc
Workaround for the dmabuf-wayland green-frames bug (marfrit/dmabuf-modifier-triage#1). iter1 phase 2 source-read of KWin 6.6.4 + Mesa 26.0.6 ruled out the original H1/H2 hypotheses (panfrost offset handling and KWin wl_dmabuf import are clean) and matched the green tone to BT.601 limited-range YUV(0,0,0) -> RGB(0, 135, 0). Conclusion: panfrost reads zero-fill memory despite hantro having written real data — a cache-coherency / synchronization gap. V4L2 doesn't attach implicit fences (dma_resv) to CAPTURE buffers on VIDIOC_DQBUF; this gap is the same one our vb2_dma_resv RFC v2 addresses upstream. The userspace workaround is to issue DMA_BUF_IOCTL_SYNC(SYNC_START|SYNC_RW) + SYNC_END(SYNC_RW) on each EXPBUF fd before submitting to the compositor — invokes the producer driver's begin_cpu_access / end_cpu_access path, which on most ARM SoCs flushes write buffers and synchronizes coherent memory. Patch covers BOTH vaapi_dmabuf_importer (the path our test exercises via `mpv --hwdec=vaapi`) and drmprime_dmabuf_importer (for symmetry when used via `--hwdec=drmprime`). If this works, ship it; if it doesn't, hypothesis space narrows further to GPU-side cache invalidation in panfrost's kernel-mode dma_buf import path (H7). pkgrel 8 -> 9. Patch sha256 6c929bea7636b8d81b63a1275ba1d8a471fe2f249fc23509043ace6cf9b076a7.
mpv-fourier
mpv with the fourier-umbrella patches.
Why
The dmabuf-modifier-triage campaign isolated the green-frames bug on ohm
to mpv's vo_dmabuf_wayland.c plane-semantics handling for V4L2 stateless
decoder dmabufs. mpv currently emits a zwp_linux_buffer_params_v1
message that mixes per-plane fds (V4L2 MPLANE export) with a single-
allocation offset for plane 1, causing KWin to read the UV chroma plane
past-EOF on the UV-plane fd and render solid dark green.
This package is the delivery vehicle for the fix once it's written. PKGBUILD is in place with an empty patch slot; the patch itself is iter1 of the triage campaign.
Tracker
- Bug: marfrit/dmabuf-modifier-triage#1
- Symptom-tracker: marfrit/libva-multiplanar#1
- Acceptance criterion:
~/src/dmabuf-modifier-triage/screenshots/
Status
- 2026-05-08: scaffold landed. Builds vanilla mpv 0.41.0 with no
fourier patches applied (patch slot empty in
prepare()). pkgrel=1 pinned to mpv release v0.41.0.
When iter1 of the triage campaign produces the patch, bump pkgrel,
add the patch to source=() and sha256sums=(), uncomment the
patch -p1 line in prepare().
Hosts that benefit
Only hosts that exercise the V4L2 stateless dmabuf-wayland path — ohm (RK3566 + hantro G1) today, fresnel (RK3399 + hantro + rkvdec) once that campaign reaches its mpv-test phase. Other hosts on [marfrit] (boltzmann, hertz desktop) won't see any difference as the fix is no-op for non-V4L2-stateless dmabuf paths.