Three prefs that gate the patched VAAPI / V4L2-request code path are
'false' upstream and need to be 'true' for the libva-v4l2-request
backend to actually engage on RK3399 / panfrost EGL:
widget.dmabuf.force-enabled
media.hardware-video-decoding.force-enabled
media.ffvpx-hw.enabled
Without them, Gecko's DMA-BUF probe silently fails (panfrost EGL
doesn't trip the right detection), the (Intel-tuned) HW-vs-SW cost
heuristic picks SW, and the FFvpx PDM never selects its HW-capable
variant for VP8/VP9. Net result: fresh profiles SW-decode despite the
0001..0004 unlock patches being applied.
Shipped as a vendor-default file at:
/usr/lib/firefox-fourier/browser/defaults/preferences/rockchip-fourier-defaults.js
Lower precedence than user.js / about:config — power users debugging
HW decode can still flip them off without touching the package.
pkgrel 1 -> 2.
Closes: #8
Extends Mozilla's RDD sandbox to permit /dev/media* (driver-matched),
the MEDIA_IOC_* ioctl family ('|'), and the sysfs paths libudev would
need to enumerate the media controller (read-only AddTree on
/sys/class, /sys/bus, /sys/dev/char, /sys/devices/platform plus
/run/udev, /etc/udev/udev.conf, /proc/self, /dev/dma_heap).
Necessary but not sufficient on its own: Mozilla's OpenAtTrap
rejects fd-relative openat used by systemd's chase() inside libudev.
The companion ffmpeg-v4l2-request-git patch adds a brute-force
fallback that opens /dev/media[0..15] directly with absolute paths,
which composes with this broker policy.
Validated on RK3399 / Pinebook Pro / mainline rkvdec: with both
patches in place, default RDD sandbox runs HW decode at ~5% CPU on
1080p30 H.264 (vs ~64% software fallback before). Closes the
parity gap with MOZ_DISABLE_RDD_SANDBOX=1 baseline.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Earlier todays edit on patch 3 (commit e0af915) replaced the
hw_configs-only sanity check with the dual-mechanism (named codec
first, hw_configs fallback) probe. The patch text body was correct
but the unified-diff hunk header at the .cpp portion still claimed
the original +105-line content while the new content was 130 lines —
so patch -p1 failed with malformed-patch errors.
This commit regenerates the .cpp portion canonically via diff -u
from the boltzmann working tree (which has the corrected logic
applied). The .h portion is unchanged. Patch description body
updated to match the new dual-mechanism behavior.
Empirical evidence the new logic is correct lives in the boltzmann
build; this commit just makes the patch file replay correctly from
scratch.
Diagnostic on fresnel (RK3399 / Mali-T860 / mainline) showed
InitV4L2RequestDecoder running at runtime but failing the
hw_configs sanity check:
FFMPEG: Initialising V4L2 stateless (request API) FFmpeg decoder
FFMPEG: codec h264 has no DRM hwaccel —
libavcodec built without --enable-v4l2-request?
FFMPEG: Initialising V4L2-DRM FFmpeg decoder ← falls through to stateful
FFMPEG: V4L2 codec h264_v4l2m2m : V4L2 mem2mem H.264 decoder wrapper
FFMPEG: Couldn't initialise V4L2 decoder ← stateful also fails
…despite the system libavcodec.so being built with
--enable-v4l2-request and exposing h264_v4l2request, vp8_v4l2request,
etc. as named AVCodec entries.
Root cause: libavcodec exposes v4l2_request through one of two
mechanisms depending on the build:
(a) Named AVCodec entry (legacy, distro-portable): looked up via
avcodec_find_decoder_by_name("h264_v4l2request"). ALARM,
Debian, most distros use this.
(b) hw_configs entry on the generic codec (modern, upstream): the
generic codec's AVCodecHWConfig array advertises
AV_HWDEVICE_TYPE_DRM. Setting hw_device_ctx binds the hwaccel.
Patch 3 originally only probed (b). On builds that ship (a) — the
common case — the check failed even though v4l2_request was fully
functional.
Fix: probe (a) first via the codec-name lookup table, fall back to
(b) walking hw_configs. Both shapes work; the decoder is opened with
whichever AVCodec the probe selected. The DRM hwdevice ctx
attachment is unchanged (both mechanisms need it for surface
allocation).
Patch description and the in-code comment block updated to document
the dual mechanism. Behaviour on stateful-only boards (Pi4 / vendor
MPP) preserved: neither named codec nor hw_configs DRM is registered,
the function bails out, the existing InitV4L2Decoder runs as before.
claude-noether
test0r
marfrit