daedalus-v4l2-dkms: 79256dc -> 6ffe92b — fix kernel panic from #7

Kernel-only bump. Fixes the hard-reboot regression introduced by the daedalus-v4l2#7 split-completion design and observed on higgs (Pi CM5) during the first mpv vaapi-copy playback of 720p H.264: device_run now removes src + dst from m2m_ctx's rdy_queue at the moment it picks them up, not at buf_done time. Without this, a parked dst_buf (waiting for libavcodec's display-order release) stayed in the rdy_queue and got re-picked by the next device_run after SRC_CONSUMED's job_finish released the scheduler — two inflight entries on the same vb2_buffer, later HAS_PIXELS calls list_del on an already-detached list_head, panic. DAEDALUS_PROTO_VERSION stays at 1 — daemon (userspace daedalus-v4l2) need NOT bump in lockstep with this DKMS update. The existing daedalus-v4l2 0.1.0+r28+g79256dc is wire-compatible with daedalus-v4l2-dkms 0.1.0+r30+g6ffe92b. Refs: * reauktion/daedalus-v4l2#8
Merge pull request 'libva-v4l2-request-fourier: bump pin to c454618 (#15 transparent resize)' (#68 ) from claude-noether/marfrit-packages:bump-libva-fourier-c454618-issue-15 into main
2026-05-21 13:56:42 +02:00 · 2026-05-21 11:25:39 +00:00 · 2026-05-21 13:24:21 +02:00
5 changed files with 70 additions and 42 deletions
@@ -16,14 +16,17 @@
 pkgname=daedalus-v4l2-dkms
 _module=daedalus_v4l2

-# Same pin as arch/daedalus-v4l2 — keep kernel module + daemon
-# bit-versioned together so the chardev wire protocol stays in sync.
-# PROTO_VERSION 0 → 1 at this pin (H.264 B-frame reorder fix); must
-# install both packages atomically.
-_commit=79256dc7ef41f83873ca9c23db20f5888858e65d
+# Kernel module-only bump to 6ffe92b — fixes the panic regression
+# introduced by 79256dc's split-completion design (daedalus-v4l2#8):
+# claim src/dst from m2m's rdy_queue at device_run time, not at
+# buf_done; otherwise a parked dst_buf gets re-picked by the next
+# device_run after SRC_CONSUMED's job_finish releases the scheduler.
+# PROTO_VERSION stays at 1; daedalus-v4l2 (daemon) need NOT bump in
+# lockstep — daemon ABI is unchanged.
+_commit=6ffe92bcacccbf228c398a10c21a6818f07c67d2

-pkgver=0.1.0.r28.79256dc
-pkgrel=1  # reset for new upstream pin (79256dc — H.264 B-frame reorder fix)
+pkgver=0.1.0.r30.6ffe92b
+pkgrel=1  # reset for new upstream pin (6ffe92b — claim src/dst at device_run, fixes #8 panic)
 pkgdesc="V4L2 stateless decoder shim kernel module (DKMS) — Pi 5 / CM5"
 arch=('any')
 url="https://git.reauktion.de/reauktion/daedalus-v4l2"
@@ -24,28 +24,29 @@ pkgname=libva-v4l2-request-fourier
 epoch=1
 _upstreampkg=libva-v4l2-request

-# Pin the fork tip. 2860d75 = PR #14 merge "picture: bounds-check
-# codec_store_buffer slice writes against source_size (#13)" — addresses
-# marfrit/libva-v4l2-request-fourier#13. Guards the three append sites
-# in codec_store_buffer's VASliceDataBufferType branch (H.264 Annex-B
-# start code, VP8 uncompressed-header pad, slice payload) against the
-# OUTPUT pool slot's fixed sizeimage; returns
-# VA_STATUS_ERROR_ALLOCATION_FAILED with a request_log line instead of
-# memcpy'ing past the mmap on a resolution upshift mid-stream (SIGSEGV
-# in mpv --hwdec=vaapi-copy, heap-corruption hazard in Firefox RDD).
-# The root-cause refactor (re-init OUTPUT pool / re-create surfaces on
-# resolution change, or grow source_data on demand) is tracked as the
-# follow-up backlog item; this PR is the memory-safety floor.
+# Pin the fork tip. c454618 = PR #16 merge "picture, request_pool:
+# transparent OUTPUT-pool resize on bitstream overrun (#15)" —
+# follow-up root-cause fix to #13/#14. On a mid-stream bitstream-
+# budget overrun (typical cause: SPS-driven resolution upshift in an
+# adaptive-bitrate stream), codec_store_buffer now snapshots the in-
+# flight surface's accumulated bytes, releases its OUTPUT pool slot,
+# calls request_pool_resize (STREAMOFF → REQBUFS(0) → S_FMT with
+# 2×sizeimage hint, capped at 1 GiB, page-aligned → CREATE_BUFS →
+# mmap → media_request_alloc → STREAMON), re-acquires a slot, re-
+# mirrors the surface's source_{data,size,request_fd}, restores the
+# bytes, and continues. The frame survives instead of being dropped
+# back to libavcodec for surface recreation. CAPTURE side untouched
+# (per-queue V4L2 streaming independence).
 #
-# Prior pin (77f9236) = PR #12 merge — av1_set_controls
-# (V4L2_CID_STATELESS_AV1_SEQUENCE for the daedalus daemon track).
-_commit=2860d75afe6a8e34df6afb508ff85c822bf9e908
+# Prior pin (2860d75) = PR #14 merge — codec_store_buffer bounds-
+# check floor (#13).
+_commit=c454618ae11addce2e17b560f4deeacbed067d98

 # Project version from meson.build (1.0.0) + commit count + short sha,
 # matching the ffmpeg-v4l2-request-fourier convention. Recomputed at
 # build time by pkgver() below; the static value here is a placeholder
 # so AUR-style consumers see something coherent before src/ exists.
-pkgver=1.0.0.r388.2860d75
+pkgver=1.0.0.r390.c454618
 pkgrel=1
 pkgdesc="VA-API backend for V4L2 stateless decoders (multiplanar fork — fourier umbrella)"
 arch=('aarch64')
@@ -14,9 +14,9 @@
 # Sibling userspace package: ../daedalus-v4l2/build-deb.sh
 set -euo pipefail

-UPSTREAM_COMMIT=79256dc7ef41f83873ca9c23db20f5888858e65d
-PKGVER=0.1.0+r28+g79256dc
-PKGREL=1  # reset for new upstream pin (79256dc — H.264 B-frame reorder fix); still carries the #64 multi-kernel postinst fix
+UPSTREAM_COMMIT=6ffe92bcacccbf228c398a10c21a6818f07c67d2
+PKGVER=0.1.0+r30+g6ffe92b
+PKGREL=1  # reset for new upstream pin (6ffe92b — claim src/dst at device_run, fixes #8 panic); still carries the #64 multi-kernel postinst fix
 MODULE_NAME=daedalus_v4l2

 HERE=$(dirname "$(readlink -f "$0")")
@@ -1,3 +1,25 @@
+daedalus-v4l2-dkms (0.1.0+r30+g6ffe92b-1) bookworm trixie; urgency=medium
+
+  * Bump to 6ffe92b — fixes the kernel panic regression introduced
+    by 79256dc's split-completion design (closes daedalus-v4l2#8).
+    `device_run` now removes both src + dst from `m2m_ctx`'s
+    rdy_queue at pickup time, not at `buf_done` time.  Without
+    this, after `SRC_CONSUMED`'s `job_finish` released the m2m
+    scheduler, the NEXT `device_run` saw the still-queued parked
+    dst_buf and paired it with a fresh src — two inflight entries
+    referencing the same vb2_buffer, the later `HAS_PIXELS`
+    triggered list_del on an already-detached list_head, smashing
+    the rdy_queue → hard reboot on Pi CM5 during `mpv vaapi-copy`
+    playback of 720p H.264 (2026-05-21).
+  * Wire protocol unchanged — DAEDALUS_PROTO_VERSION stays at 1.
+    Daemon (userspace daedalus-v4l2 package) need NOT bump in
+    lockstep with this DKMS update; the existing
+    daedalus-v4l2 0.1.0+r28+g79256dc is wire-compatible with
+    daedalus-v4l2-dkms 0.1.0+r30+g6ffe92b.
+  * Carries forward the #64 multi-kernel postinst fix.
+
+ -- Markus Fritsche <mfritsche@reauktion.de>  Thu, 21 May 2026 14:00:00 +0000
+
 daedalus-v4l2-dkms (0.1.0+r28+g79256dc-1) bookworm trixie; urgency=medium

  * Bump to 79256dc — H.264 B-frame display reorder fix (closes
@@ -10,23 +10,25 @@
 # Upstream fork: https://git.reauktion.de/marfrit/libva-v4l2-request-fourier
 set -euo pipefail

-# Same pin as the Arch PKGBUILD.  2860d75 = PR #14 merge "picture:
-# bounds-check codec_store_buffer slice writes against source_size
-# (#13)" — guards the three append sites in codec_store_buffer's
-# VASliceDataBufferType branch (H.264 Annex-B start code, VP8
-# uncompressed-header pad, slice payload) against the OUTPUT pool
-# slot's fixed sizeimage; returns VA_STATUS_ERROR_ALLOCATION_FAILED
-# with a request_log line instead of memcpy'ing past the mmap on a
-# resolution upshift mid-stream.  Fixes a SIGSEGV in mpv
-# --hwdec=vaapi-copy and a heap-corruption hazard in Firefox RDD.
-# The root-cause refactor (re-init OUTPUT pool / re-create surfaces
-# on resolution change, or grow source_data on demand) is tracked
-# as the follow-up backlog item; this is the memory-safety floor.
+# Same pin as the Arch PKGBUILD.  c454618 = PR #16 merge "picture,
+# request_pool: transparent OUTPUT-pool resize on bitstream overrun
+# (#15)" — follow-up root-cause fix to #13/#14.  On a mid-stream
+# bitstream-budget overrun (typical cause: SPS-driven resolution
+# upshift in an adaptive-bitrate stream), codec_store_buffer now
+# snapshots the in-flight surface's accumulated bytes, releases its
+# OUTPUT pool slot, calls request_pool_resize (STREAMOFF →
+# REQBUFS(0) → S_FMT with 2×sizeimage hint, capped at 1 GiB, page-
+# aligned → CREATE_BUFS → mmap → media_request_alloc → STREAMON),
+# re-acquires a slot, re-mirrors the surface's source_{data,size,
+# request_fd}, restores the bytes, and continues.  The frame
+# survives instead of being dropped back to libavcodec for surface
+# recreation.  CAPTURE side untouched (per-queue V4L2 streaming
+# independence).
 #
-# Prior pin (77f9236) = PR #12 merge — av1_set_controls
-# (V4L2_CID_STATELESS_AV1_SEQUENCE for the daedalus daemon track).
-UPSTREAM_COMMIT=2860d75afe6a8e34df6afb508ff85c822bf9e908
-PKGVER=1.0.0+r388+g2860d75
+# Prior pin (2860d75) = PR #14 merge — codec_store_buffer bounds-
+# check floor (#13).
+UPSTREAM_COMMIT=c454618ae11addce2e17b560f4deeacbed067d98
+PKGVER=1.0.0+r390+gc454618
 PKGREL=1

 HERE=$(dirname "$(readlink -f "$0")")