daedalus-v4l2{,-dkms}: 3dd0eb0 -> 462aa4b — kernel ctrl-binding fix

Upstream PR #2 landed the one-line kernel fix that was the missing
half of issue libva-v4l2-request-fourier#8: device_run now calls
v4l2_ctrl_request_setup() before reading ctrl->p_cur, so the
daedalus_h264_meta the daemon receives reflects the in-flight
media_request's bound H.264 stateless control values instead of
stale/default ones.

Pairs with libva-v4l2-request-fourier 1.0.0+r382+gc1bb444 (max_num_
ref_frames fallback + Fix 4 instrumentation that exposed the
control-binding gap in the first place).

Effect on Pi 5 / CM5 hosts (higgs): ffmpeg -hwaccel vaapi against
H.264 sources now produces actual decoded content (per-frame
fnv1a hashes differ, zero MB-decode errors) instead of the
constant 0x6a6a05c5 "best-effort give-up" hash and cascading
decode warnings.

Both packages: pkgver 0.1.0.r22.462aa4b, pkgrel reset to 1.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

.gitea/workflows/build.yml

@@ -1144,39 +1144,25 @@ jobs:
           echo "$result" >> "$GITHUB_OUTPUT"
           echo "decision: $result"
 
-      - name: install build-deps (sans ffmpeg — see [marfrit] step)
+      - name: install build-deps
         if: steps.skip-check.outputs.skip != '1'
         run: |
           set -e
           retry() { for i in 1 2 3; do "$@" && return 0; rc=$?; echo "retry $i (exit=$rc)" >&2; sleep $((i*5)); done; return 1; }
-          # Do NOT pull stock 'ffmpeg' here: the arch-aarch64 runner has
-          # ffmpeg-v4l2-request-fourier pre-installed from the mpv-aarch64
-          # job (configured via [marfrit]), and pacman -S ffmpeg would
-          # conflict on the libav* drop-in.  Daedalus build only needs
-          # libavcodec/libavformat headers, which the fourier package
-          # already supplies.  Keep cmake/ninja/pkgconf/libdrm here; the
-          # ffmpeg-dev equivalent comes via the next step.
-          retry pacman -Syu --noconfirm --needed \
-            dpkg openssh rsync curl base-devel git cmake ninja pkgconf \
-            libdrm
-
-      - name: ensure ffmpeg-v4l2-request-fourier installed (link-time ABI source)
-        if: steps.skip-check.outputs.skip != '1'
-        run: |
-          set -e
-          # Idempotent: pre-install the marfrit fourier ffmpeg so cmake
-          # finds libavcodec / libavformat / libavutil headers + .so's.
-          # Mirrors mpv-fourier-debian's [marfrit] step.
-          curl -sLo /tmp/marfrit.gpg https://packages.reauktion.de/marfrit.gpg
-          pacman-key --add /tmp/marfrit.gpg
-          pacman-key --lsign-key 92D5E96D8F63C75E4116AA1FF5C8C4603D0D250C
-          rm -f /tmp/marfrit.gpg
-          if ! grep -q '^\[marfrit\]' /etc/pacman.conf; then
-            printf '\n[marfrit]\nServer = https://packages.reauktion.de/arch/$arch\nSigLevel = Required\n' >> /etc/pacman.conf
-          fi
-          pacman -Sy --noconfirm
-          rm -f /var/cache/pacman/pkg/ffmpeg-v4l2-request-fourier-*-aarch64.pkg.tar.*
-          printf 'y\ny\ny\n' | pacman -S --needed marfrit/ffmpeg-v4l2-request-fourier
+          export DEBIAN_FRONTEND=noninteractive
+          retry apt-get update -qq
+          # libav*-dev provide the headers daedalus daemon dlopens at
+          # runtime — Debian's stock packages match the trixie ABI the
+          # daemon will encounter on Pi 5 hosts (both ship libavcodec
+          # 61.x).  The fourier ffmpeg fork isn't needed here; the
+          # daemon never link-binds against libav (Option γ — dlopen
+          # at runtime), so any header set with the right struct
+          # definitions works.
+          retry apt-get install -y --no-install-recommends \
+              build-essential cmake ninja-build pkg-config git \
+              libavcodec-dev libavformat-dev libavutil-dev libdrm-dev \
+              linux-libc-dev \
+              curl ca-certificates openssh-client rsync dpkg-dev
 
       - name: install hertz deploy ssh key
         if: steps.skip-check.outputs.skip != '1'
@@ -1238,7 +1224,10 @@ jobs:
         run: |
           set -e
           retry() { for i in 1 2 3; do "$@" && return 0; rc=$?; echo "retry $i (exit=$rc)" >&2; sleep $((i*5)); done; return 1; }
-          retry pacman -Syu --noconfirm --needed dpkg openssh rsync curl tar gzip
+          export DEBIAN_FRONTEND=noninteractive
+          retry apt-get update -qq
+          retry apt-get install -y --no-install-recommends \
+              dpkg-dev openssh-client rsync curl ca-certificates tar gzip
 
       - name: install hertz deploy ssh key
         if: steps.skip-check.outputs.skip != '1'

arch/daedalus-v4l2-dkms/PKGBUILD

@@ -18,10 +18,10 @@ _module=daedalus_v4l2
 
 # Same pin as arch/daedalus-v4l2 — keep kernel module + daemon
 # bit-versioned together so the chardev wire protocol stays in sync.
-_commit=481279c9bffd19e32c8f3299897e9b63fc5a24aa
+_commit=462aa4b4804972189d2f1b1ab5782f1ca1e8c423
 
-pkgver=0.1.0.r18.481279c
-pkgrel=1  # reset for new upstream pin (481279c — Phase 8.13 close)
+pkgver=0.1.0.r22.462aa4b
+pkgrel=1  # reset for new upstream pin (3dd0eb0 — DAEMON-PPS H.264 SPS/PPS NAL synth)
 pkgdesc="V4L2 stateless decoder shim kernel module (DKMS) — Pi 5 / CM5"
 arch=('any')
 url="https://git.reauktion.de/reauktion/daedalus-v4l2"

arch/daedalus-v4l2/PKGBUILD

@@ -21,12 +21,12 @@ _upstreampkg=daedalus-v4l2
 # ffmpeg -hwaccel vaapi → libva → /dev/video0 → daemon path lands a
 # pixel-correct decoded frame back in ffmpeg.  Promote to a later pin
 # only after a future phase closes cleanly.
-_commit=481279c9bffd19e32c8f3299897e9b63fc5a24aa
+_commit=462aa4b4804972189d2f1b1ab5782f1ca1e8c423
 
 # 0.1.0 (pre-1.0) + commit count + short sha.  Bump the .Y on each
 # Phase 8.x close.  pkgver() recomputes at build time.
-pkgver=0.1.0.r18.481279c
-pkgrel=1  # reset for new upstream pin (481279c — Phase 8.13 close)
+pkgver=0.1.0.r22.462aa4b
+pkgrel=1  # reset for new upstream pin (3dd0eb0 — DAEMON-PPS H.264 SPS/PPS NAL synth)
 pkgdesc="Userspace daemon for the daedalus-v4l2 V4L2 stateless decoder shim (VP9/AV1/H.264 on Pi 5 / CM5)"
 arch=('aarch64')
 url="https://git.reauktion.de/reauktion/daedalus-v4l2"

arch/libva-v4l2-request-fourier/PKGBUILD

@@ -24,31 +24,29 @@ pkgname=libva-v4l2-request-fourier
 epoch=1
 _upstreampkg=libva-v4l2-request
 
-# Pin the fork tip. de27e95 = "v4l2: log error_idx + failing ctrl id
-# on S_EXT_CTRLS failure" — Phase 8.13 diagnostic that surfaced the
-# real root cause of the libva→daedalus_v4l2 request-completion
-# timeout (turned out the EINVAL libva was logging was a harmless
-# H264/HEVC probe; actual VP9 stateless control SET worked all along).
+# Pin the fork tip. c1bb444 = PR #9 merge "h264: max_num_ref_frames
+# fallback + libva-boundary instrumentation (#8)" — addresses the
+# libva-side portion of marfrit/libva-v4l2-request-fourier#8: the
+# daedalus_v4l2-via-libavcodec strict consumer rejected frames whose
+# sps.max_num_ref_frames was 0 (older ffmpeg-vaapi paths leave the
+# field unset and HW decoders tolerated it). Adds a per-profile spec
+# minimum fallback (counts valid DPB entries first; 1 for baseline /
+# 4 for main+high if even that is 0) and a one-line request_log at
+# h264_set_controls entry dumping raw VAAPI bitfields (seq_fields,
+# pic_fields, num_ref_frames, bit_depth_*) for disambiguating where
+# the still-open PPS-flag-zero portion of issue #8 originates.
 #
-# Prior pin (7ac934e) was iter38b — fresnel-fourier multi-device probe
-# + MAX_PROFILES bounds-check fix. de27e95 added the daedalus_v4l2
-# probe slot (b5b3acf), the meson option gate (2146341), and the
-# S_EXT_CTRLS diagnostic (de27e95 itself).  c332d34 (LIBVA-1) added
-# the per-codec dispatch: rpi-hevc-dec + daedalus_v4l2 both probe each
-# other as alts, VP9/AV1/H.264 route to daedalus via new 'd' kind,
-# HEVC stays on 'p' (rpi-hevc-dec).  9898331 (LIBVA-2) completes that
-# by adding video_fd_daedalus to any_fd_supports_output_format's probe
-# array — without it, H.264/VP9/AV1 profiles never got advertised on
-# Pi 5 mixed deployments (rpi-hevc-dec primary, daedalus alt) and
-# ffmpeg bailed with "No support for codec h264 profile 578" before
-# the per-codec dispatch could fire.
-_commit=989833114a7708ad999dc68309cbc181d9913bdb
+# Prior pin (9898331) was the LIBVA-2 close — config-side video_fd_daedalus
+# added to the profile enumeration probe array.  c332d34 (LIBVA-1) added
+# the per-codec dispatch.  Earlier de27e95 (Phase 8.13 diagnostic),
+# 7ac934e (iter38b multi-device probe + MAX_PROFILES bounds-check fix).
+_commit=c1bb444d0731ca7ca831c7e01ec4b52ed736ca8e
 
 # 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.r380.9898331
+pkgver=1.0.0.r382.c1bb444
 pkgrel=1
 pkgdesc="VA-API backend for V4L2 stateless decoders (multiplanar fork — fourier umbrella)"
 arch=('aarch64')

arch/mesa-panvk-bifrost/0003-panvk-bifrost-vk-ext-transform-feedback.patch

@@ -0,0 +1,328 @@
+--- a/src/panfrost/vulkan/panvk_shader.h	2026-04-29 22:19:00.000000000 +0200
++++ b/src/panfrost/vulkan/panvk_shader.h	2026-05-20 18:52:53.312698258 +0200
+@@ -150,6 +150,10 @@
+    struct {
+ #if PAN_ARCH < 9
+       int32_t raw_vertex_offset;
++      uint32_t num_vertices;       /* iter13: XFB needs per-draw vertex count */
++      /* aligned_u64 attribute below inserts the 4-byte alignment gap
++       * after num_vertices automatically — no explicit pad needed. */
++      aligned_u64 xfb_address[4];  /* iter13: 4 transform feedback buffer base addresses */
+ #endif
+       int32_t first_vertex;
+       int32_t base_instance;
+--- a/src/panfrost/vulkan/panvk_vX_physical_device.c	2026-05-20 19:09:29.711145446 +0200
++++ b/src/panfrost/vulkan/panvk_vX_physical_device.c	2026-05-20 18:52:54.832720445 +0200
+@@ -169,6 +169,7 @@
+       .EXT_provoking_vertex = true,
+       .EXT_queue_family_foreign = true,
+       .EXT_robustness2 = true,
++      .EXT_transform_feedback = PAN_ARCH < 9,   /* iter13: JM-class only for now */
+       .EXT_sampler_filter_minmax = PAN_ARCH >= 10,
+       .EXT_scalar_block_layout = true,
+       .EXT_separate_stencil_usage = true,
+@@ -495,6 +496,10 @@
+       .robustImageAccess2 = false,
+       .nullDescriptor = true,
+ 
++      /* VK_EXT_transform_feedback (iter13) */
++      .transformFeedback = PAN_ARCH < 9,
++      .geometryStreams = false,
++
+       /* VK_KHR_shader_clock */
+       .shaderSubgroupClock = device->kmod.dev->props.gpu_can_query_timestamp,
+       .shaderDeviceClock = device->kmod.dev->props.timestamp_device_coherent,
+@@ -1020,6 +1025,18 @@
+       .robustStorageBufferAccessSizeAlignment = 1,
+       .robustUniformBufferAccessSizeAlignment = 1,
+ 
++      /* VK_EXT_transform_feedback (iter13) */
++      .maxTransformFeedbackStreams = 1,
++      .maxTransformFeedbackBuffers = 4,
++      .maxTransformFeedbackBufferSize = UINT32_MAX,
++      .maxTransformFeedbackStreamDataSize = 512,
++      .maxTransformFeedbackBufferDataSize = 512,
++      .maxTransformFeedbackBufferDataStride = 2048,
++      .transformFeedbackQueries = false,
++      .transformFeedbackStreamsLinesTriangles = false,
++      .transformFeedbackRasterizationStreamSelect = false,
++      .transformFeedbackDraw = false,
++
+       /* VK_EXT_shader_object */
+       /* We do not currently support VK_EXT_shader_object but this is used
+        * internally by vk_shader
+--- a/src/panfrost/vulkan/panvk_vX_shader.c	2026-04-29 22:19:00.000000000 +0200
++++ b/src/panfrost/vulkan/panvk_vX_shader.c	2026-05-20 18:52:56.556745611 +0200
+@@ -21,6 +21,7 @@
+ #include "panvk_physical_device.h"
+ #include "panvk_sampler.h"
+ #include "panvk_shader.h"
++#include "pan_nir.h"   /* iter13: pan_nir_lower_xfb */
+ 
+ #include "spirv/nir_spirv.h"
+ #include "util/memstream.h"
+@@ -100,6 +101,20 @@
+    case nir_intrinsic_load_raw_vertex_offset_pan:
+       val = load_sysval(b, graphics, bit_size, vs.raw_vertex_offset);
+       break;
++   case nir_intrinsic_load_num_vertices:    /* iter13: XFB index calc */
++      val = load_sysval(b, graphics, bit_size, vs.num_vertices);
++      break;
++   case nir_intrinsic_load_xfb_address: {   /* iter13: XFB buffer N base address */
++      unsigned idx = nir_intrinsic_base(intr);
++      switch (idx) {
++      case 0: val = load_sysval(b, graphics, bit_size, vs.xfb_address[0]); break;
++      case 1: val = load_sysval(b, graphics, bit_size, vs.xfb_address[1]); break;
++      case 2: val = load_sysval(b, graphics, bit_size, vs.xfb_address[2]); break;
++      case 3: val = load_sysval(b, graphics, bit_size, vs.xfb_address[3]); break;
++      default: return false;
++      }
++      break;
++   }
+    case nir_intrinsic_load_layer_id:
+       assert(b->shader->info.stage == MESA_SHADER_FRAGMENT);
+       val = load_sysval(b, graphics, bit_size, layer_id);
+@@ -457,6 +472,7 @@
+             core_max_id);
+ 
+    pan_preprocess_nir(nir, pdev->kmod.dev->props.gpu_id);
++
+ }
+ 
+ static void
+@@ -870,6 +886,18 @@
+             nir_var_shader_in | nir_var_shader_out, UINT32_MAX);
+    NIR_PASS(_, nir, nir_lower_io, nir_var_shader_in | nir_var_shader_out,
+             glsl_type_size, nir_lower_io_use_interpolated_input_intrinsics);
++
++#if PAN_ARCH < 9
++   /* iter13: VK_EXT_transform_feedback — runs AFTER nir_lower_io so that
++    * shader outputs are now store_output intrinsics that pan_nir_lower_xfb
++    * can rewrite to nir_store_global+nir_load_xfb_address. */
++   if (nir->info.stage == MESA_SHADER_VERTEX &&
++       nir->info.has_transform_feedback_varyings) {
++      NIR_PASS(_, nir, nir_opt_constant_folding);
++      NIR_PASS(_, nir, nir_io_add_intrinsic_xfb_info);
++      NIR_PASS(_, nir, pan_nir_lower_xfb);
++   }
++#endif
+ }
+ 
+ static VkResult
+@@ -1288,6 +1316,9 @@
+       .view_mask = (state && state->rp) ? state->rp->view_mask : 0,
+       .robust2_modes = robust2_modes,
+       .robust_descriptors = dev->vk.enabled_features.nullDescriptor,
++      /* iter13: XFB shaders must disable IDVS (matches Panfrost-Gallium). */
++      .no_idvs = (info->stage == MESA_SHADER_VERTEX) &&
++                 info->nir->info.has_transform_feedback_varyings,
+    };
+ 
+    switch (info->stage) {
+--- a/src/panfrost/vulkan/panvk_cmd_draw.h	2026-04-29 22:19:00.000000000 +0200
++++ b/src/panfrost/vulkan/panvk_cmd_draw.h	2026-05-20 18:52:57.748763011 +0200
+@@ -135,6 +135,19 @@
+    struct panvk_graphics_sysvals sysvals;
+ 
+ #if PAN_ARCH < 9
++   /* iter13: VK_EXT_transform_feedback state (JM-class only for now). */
++   struct {
++      bool active;
++      uint32_t buffer_count;
++      struct {
++         uint64_t addr;
++         uint64_t offset;
++         uint64_t size;
++      } buffers[4];
++   } xfb;
++#endif
++
++#if PAN_ARCH < 9
+    struct panvk_shader_link link;
+ #endif
+ 
+--- a/src/panfrost/vulkan/panvk_vX_cmd_draw.c	2026-04-29 22:19:00.000000000 +0200
++++ b/src/panfrost/vulkan/panvk_vX_cmd_draw.c	2026-05-20 19:10:23.031919662 +0200
+@@ -10,6 +10,7 @@
+ #include "panvk_entrypoints.h"
+ 
+ #include "pan_desc.h"
++#include "pan_compiler.h"   /* PAN_SHADER_OOB_ADDRESS */
+ #include "pan_util.h"
+ 
+ static void
+@@ -722,6 +723,35 @@
+    set_gfx_sysval(cmdbuf, dirty_sysvals, vs.raw_vertex_offset,
+                   info->vertex.raw_offset);
+    set_gfx_sysval(cmdbuf, dirty_sysvals, layer_id, info->layer_id);
++
++   /* iter13: VK_EXT_transform_feedback sysvals — always set (per draw),
++    * reflect bound XFB state. set_gfx_sysval is a no-op if value unchanged. */
++   set_gfx_sysval(cmdbuf, dirty_sysvals, vs.num_vertices, info->vertex.count);
++   {
++      const struct panvk_cmd_graphics_state *_gfx = &cmdbuf->state.gfx;
++      /* iter13: default each XFB buffer address to PAN_SHADER_OOB_ADDRESS
++       * (= 1<<63). This is the Panfrost-Gallium memory-sink idiom — the
++       * Bifrost MMU silently discards stores to this address, so a pipeline
++       * with XFB outputs used in a non-XFB draw (or in an XFB draw with
++       * fewer bound buffers than the shader declares) is safe instead of
++       * faulting. See gallium/drivers/panfrost/pan_cmdstream.c PAN_SYSVAL_XFB. */
++      uint64_t _xa0 = PAN_SHADER_OOB_ADDRESS, _xa1 = PAN_SHADER_OOB_ADDRESS,
++               _xa2 = PAN_SHADER_OOB_ADDRESS, _xa3 = PAN_SHADER_OOB_ADDRESS;
++      if (_gfx->xfb.active) {
++         if (_gfx->xfb.buffer_count > 0 && _gfx->xfb.buffers[0].addr)
++            _xa0 = _gfx->xfb.buffers[0].addr + _gfx->xfb.buffers[0].offset;
++         if (_gfx->xfb.buffer_count > 1 && _gfx->xfb.buffers[1].addr)
++            _xa1 = _gfx->xfb.buffers[1].addr + _gfx->xfb.buffers[1].offset;
++         if (_gfx->xfb.buffer_count > 2 && _gfx->xfb.buffers[2].addr)
++            _xa2 = _gfx->xfb.buffers[2].addr + _gfx->xfb.buffers[2].offset;
++         if (_gfx->xfb.buffer_count > 3 && _gfx->xfb.buffers[3].addr)
++            _xa3 = _gfx->xfb.buffers[3].addr + _gfx->xfb.buffers[3].offset;
++      }
++      set_gfx_sysval(cmdbuf, dirty_sysvals, vs.xfb_address[0], _xa0);
++      set_gfx_sysval(cmdbuf, dirty_sysvals, vs.xfb_address[1], _xa1);
++      set_gfx_sysval(cmdbuf, dirty_sysvals, vs.xfb_address[2], _xa2);
++      set_gfx_sysval(cmdbuf, dirty_sysvals, vs.xfb_address[3], _xa3);
++   }
+ #endif
+ 
+    if (dyn_gfx_state_dirty(cmdbuf, CB_BLEND_CONSTANTS)) {
+--- a/src/panfrost/vulkan/meson.build	2026-04-29 22:19:00.000000000 +0200
++++ b/src/panfrost/vulkan/meson.build	2026-05-20 18:53:04.484861338 +0200
+@@ -73,6 +73,7 @@
+ jm_inc_dir = ['jm']
+ jm_files = [
+   'jm/panvk_vX_bind_queue.c',
++  'jm/panvk_vX_cmd_xfb.c',   # iter13
+   'jm/panvk_vX_cmd_buffer.c',
+   'jm/panvk_vX_cmd_dispatch.c',
+   'jm/panvk_vX_cmd_draw.c',
+--- a/src/panfrost/vulkan/jm/panvk_vX_cmd_buffer.c	2026-04-29 22:19:00.000000000 +0200
++++ b/src/panfrost/vulkan/jm/panvk_vX_cmd_buffer.c	2026-05-20 19:10:26.163965149 +0200
+@@ -473,5 +473,12 @@
+ 
+    vk_command_buffer_begin(&cmdbuf->vk, pBeginInfo);
+ 
++#if PAN_ARCH < 9
++   /* iter13: clear XFB state on Begin so a reused command buffer does not
++    * inherit stale xfb.buffer_count / xfb.active / xfb.buffers[] from a
++    * prior recording. */
++   memset(&cmdbuf->state.gfx.xfb, 0, sizeof(cmdbuf->state.gfx.xfb));
++#endif
++
+    return VK_SUCCESS;
+ }
+--- a/src/panfrost/vulkan/jm/panvk_vX_cmd_xfb.c	2026-05-18 12:50:53.067999996 +0200
++++ b/src/panfrost/vulkan/jm/panvk_vX_cmd_xfb.c	2026-05-20 19:10:27.175979847 +0200
+@@ -0,0 +1,111 @@
++/*
++ * Copyright © 2026 mfritsche / claude-noether
++ * SPDX-License-Identifier: MIT
++ *
++ * iter13: VK_EXT_transform_feedback command handlers for the JM
++ * architecture path (Bifrost v6/v7 + Valhall-JM v9).
++ *
++ * The runtime contract:
++ *   - vkCmdBindTransformFeedbackBuffersEXT: stash (gpu_addr, offset, size)
++ *     for each slot into cmdbuf->state.gfx.xfb.buffers[].
++ *   - vkCmdBeginTransformFeedbackEXT: set cmdbuf->state.gfx.xfb.active = true.
++ *     Mark sysvals dirty so the next draw re-emits vs.xfb_address[].
++ *   - vkCmdEndTransformFeedbackEXT: set active = false.
++ *
++ * Counter buffers (firstCounterBuffer/counterBufferCount/pCounterBuffers/
++ * pCounterBufferOffsets) are accepted by API but ignored — v1 doesn't
++ * support pause/resume. transformFeedbackDraw is advertised as false.
++ *
++ * Per-draw integration: jm/panvk_vX_cmd_draw.c reads cmdbuf->state.gfx.xfb
++ * and populates vs.xfb_address[i] for shader use. The pan_nir_lower_xfb
++ * pass in panvk_vX_shader.c emits nir_load_xfb_address(i) which lowers
++ * (via panvk_vX_shader.c sysval handler) to a load from the per-draw
++ * sysval push area.
++ */
++
++#include "vk_log.h"
++#include "util/log.h"
++
++#include "panvk_cmd_buffer.h"
++#include "panvk_cmd_draw.h"
++#include "panvk_buffer.h"
++#include "panvk_entrypoints.h"
++
++VKAPI_ATTR void VKAPI_CALL
++panvk_per_arch(CmdBindTransformFeedbackBuffersEXT)(
++   VkCommandBuffer commandBuffer,
++   uint32_t firstBinding,
++   uint32_t bindingCount,
++   const VkBuffer *pBuffers,
++   const VkDeviceSize *pOffsets,
++   const VkDeviceSize *pSizes)
++{
++   VK_FROM_HANDLE(panvk_cmd_buffer, cmdbuf, commandBuffer);
++   struct panvk_cmd_graphics_state *gfx = &cmdbuf->state.gfx;
++
++   for (uint32_t i = 0; i < bindingCount; i++) {
++      uint32_t slot = firstBinding + i;
++      if (slot >= 4)
++         continue;
++
++      VK_FROM_HANDLE(panvk_buffer, buf, pBuffers[i]);
++      gfx->xfb.buffers[slot].addr = panvk_buffer_gpu_ptr(buf, 0);
++      gfx->xfb.buffers[slot].offset = pOffsets[i];
++      gfx->xfb.buffers[slot].size =
++         (pSizes != NULL && pSizes[i] != VK_WHOLE_SIZE)
++            ? pSizes[i]
++            : (buf->vk.size - pOffsets[i]);
++   }
++
++   if (firstBinding + bindingCount > gfx->xfb.buffer_count)
++      gfx->xfb.buffer_count = firstBinding + bindingCount;
++}
++
++VKAPI_ATTR void VKAPI_CALL
++panvk_per_arch(CmdBeginTransformFeedbackEXT)(
++   VkCommandBuffer commandBuffer,
++   uint32_t firstCounterBuffer,
++   uint32_t counterBufferCount,
++   const VkBuffer *pCounterBuffers,
++   const VkDeviceSize *pCounterBufferOffsets)
++{
++   VK_FROM_HANDLE(panvk_cmd_buffer, cmdbuf, commandBuffer);
++   struct panvk_cmd_graphics_state *gfx = &cmdbuf->state.gfx;
++
++   /* Counter buffers ignored in v1 — see VkPhysicalDeviceTransformFeedback
++    * PropertiesEXT.transformFeedbackDraw = false in panvk_vX_physical_device.c.
++    * App is spec-compliant if it does not pass counter buffers (which our
++    * features advertisement allows), but warn loudly if it does so we do not
++    * silently produce wrong capture state. */
++   (void)firstCounterBuffer;
++   (void)pCounterBufferOffsets;
++   if (counterBufferCount > 0 && pCounterBuffers != NULL) {
++      mesa_logw("panvk: CmdBeginTransformFeedbackEXT: counter buffers not "
++                "implemented (transformFeedbackDraw=false); XFB resume will "
++                "restart at buffer offset 0");
++   }
++
++   gfx->xfb.active = true;
++   /* Per-draw set_gfx_sysval picks up the change automatically — no
++    * explicit dirty marking required (set_gfx_sysval uses memcmp +
++    * BITSET to detect state diffs and re-emit sysvals). */
++}
++
++VKAPI_ATTR void VKAPI_CALL
++panvk_per_arch(CmdEndTransformFeedbackEXT)(
++   VkCommandBuffer commandBuffer,
++   uint32_t firstCounterBuffer,
++   uint32_t counterBufferCount,
++   const VkBuffer *pCounterBuffers,
++   const VkDeviceSize *pCounterBufferOffsets)
++{
++   VK_FROM_HANDLE(panvk_cmd_buffer, cmdbuf, commandBuffer);
++   struct panvk_cmd_graphics_state *gfx = &cmdbuf->state.gfx;
++
++   (void)firstCounterBuffer;
++   (void)counterBufferCount;
++   (void)pCounterBuffers;
++   (void)pCounterBufferOffsets;
++
++   gfx->xfb.active = false;
++}

arch/mesa-panvk-bifrost/PKGBUILD

@@ -30,7 +30,7 @@
 
 pkgname=mesa-panvk-bifrost
 _mesaver=26.0.6
-pkgver=26.0.6.r2
+pkgver=26.0.6.r3
 pkgrel=1
 pkgdesc="Patched Mesa libvulkan_panfrost.so exposing Bifrost-gen Mali to Vulkan apps (panvk-bifrost campaign)"
 arch=('aarch64')
@@ -79,6 +79,7 @@ source=(
     "https://archive.mesa3d.org/mesa-${_mesaver}.tar.xz"
     "0001-panvk-expose-robustness2-nullDescriptor-bifrost.patch"
     "0002-panvk-expose-vulkan-1.1-1.2-on-bifrost.patch"
+    "0003-panvk-bifrost-vk-ext-transform-feedback.patch"
     "brave-vulkan"
     "icd.json"
 )
@@ -88,6 +89,7 @@ sha256sums=(
     'SKIP'
     'SKIP'
     'SKIP'
+    'SKIP'
 )
 
 prepare() {
@@ -107,12 +109,23 @@ prepare() {
     sed -i 's|bool has_vk1_1 = PAN_ARCH >= 10;|bool has_vk1_1 = true;|' src/panfrost/vulkan/panvk_vX_physical_device.c
     sed -i 's|bool has_vk1_2 = PAN_ARCH >= 10;|bool has_vk1_2 = true;|' src/panfrost/vulkan/panvk_vX_physical_device.c
 
+    # iter13: VK_EXT_transform_feedback implementation for Bifrost (PAN_ARCH<9).
+    # Applied as a real unified-diff patch — the change is too large for sed.
+    # Phase-doc context: ~/src/panvk-bifrost/phase{4,5,6}_iter13_close.md.
+    # Unlocks ANGLE-Vulkan → GLES3 → WebGL2 / WebGPU on Brave (chrome://gpu
+    # reports "Hardware accelerated" across the board for the affected paths).
+    patch -p1 < "${srcdir}/0003-panvk-bifrost-vk-ext-transform-feedback.patch"
+
     # Sanity-check the patches landed.
     grep -q "KHR_robustness2 = true," src/panfrost/vulkan/panvk_vX_physical_device.c
     grep -q "EXT_robustness2 = true," src/panfrost/vulkan/panvk_vX_physical_device.c
     grep -q "nullDescriptor = true," src/panfrost/vulkan/panvk_vX_physical_device.c
     grep -q "has_vk1_1 = true;" src/panfrost/vulkan/panvk_vX_physical_device.c
     grep -q "has_vk1_2 = true;" src/panfrost/vulkan/panvk_vX_physical_device.c
+    # iter13 sanity:
+    grep -q "EXT_transform_feedback = PAN_ARCH < 9," src/panfrost/vulkan/panvk_vX_physical_device.c
+    grep -q "pan_nir_lower_xfb" src/panfrost/vulkan/panvk_vX_shader.c
+    test -f src/panfrost/vulkan/jm/panvk_vX_cmd_xfb.c
 }
 
 build() {

debian/daedalus-v4l2-dkms/build-deb.sh

@@ -14,9 +14,9 @@
 # Sibling userspace package: ../daedalus-v4l2/build-deb.sh
 set -euo pipefail
 
-UPSTREAM_COMMIT=481279c9bffd19e32c8f3299897e9b63fc5a24aa
-PKGVER=0.1.0+r18+g481279c
-PKGREL=1  # reset for new upstream pin (481279c — Phase 8.13 close)
+UPSTREAM_COMMIT=462aa4b4804972189d2f1b1ab5782f1ca1e8c423
+PKGVER=0.1.0+r22+g462aa4b
+PKGREL=1  # reset for new upstream pin (3dd0eb0 — DAEMON-PPS H.264 SPS/PPS NAL synth)
 MODULE_NAME=daedalus_v4l2
 
 HERE=$(dirname "$(readlink -f "$0")")

debian/daedalus-v4l2-dkms/debian/changelog

@@ -1,3 +1,30 @@
+daedalus-v4l2-dkms (0.1.0+r22+g462aa4b-1) bookworm trixie; urgency=medium
+
+  * Bump to 462aa4b — kernel device_run() now calls
+    v4l2_ctrl_request_setup() before reading the H.264 stateless
+    control values from the bound media_request, so the values
+    daedalus ships to the userspace daemon match what the V4L2
+    client (libva-v4l2-request-fourier) actually set.  Closes the
+    libva→kernel control-binding gap that was causing decoded
+    frames to come back as best-effort zero garbage from libavcodec.
+  * Wire-ABI lockstep with daedalus-v4l2 0.1.0+r22+g462aa4b.
+
+ -- Markus Fritsche <mfritsche@reauktion.de>  Wed, 20 May 2026 22:00:00 +0000
+
+daedalus-v4l2-dkms (0.1.0+r20+g3dd0eb0-1) bookworm trixie; urgency=medium
+
+  * Bump to 3dd0eb0 — DAEMON-PPS kernel-side changes.  device_run()
+    now reads the V4L2 H.264 stateless control values from the bound
+    media_request and ships them to the daemon inside REQ_DECODE
+    via the new struct daedalus_h264_meta block (gated on
+    DAEDALUS_REQ_FLAG_H264_META).  Required for H.264 decode to
+    work via the libva-v4l2-request -> daedalus daemon path; daemon
+    synthesises AnnexB SPS+PPS NAL units from the structs.
+  * Wire-ABI lockstep with daedalus-v4l2 0.1.0+r20+g3dd0eb0 — install
+    both packages together.
+
+ -- Markus Fritsche <mfritsche@reauktion.de>  Wed, 20 May 2026 21:00:00 +0000
+
 daedalus-v4l2-dkms (0.1.0+r18+g481279c-1) bookworm trixie; urgency=medium
 
   * Bump to 481279c in lockstep with the userspace daedalus-v4l2

debian/daedalus-v4l2/build-deb.sh

@@ -15,9 +15,9 @@ set -euo pipefail
 # end-to-end via libva (consumer target hit)" — first commit where the
 # full ffmpeg -hwaccel vaapi → libva → /dev/video0 → daemon path lands
 # a pixel-correct decoded frame back in ffmpeg.
-UPSTREAM_COMMIT=481279c9bffd19e32c8f3299897e9b63fc5a24aa
-PKGVER=0.1.0+r18+g481279c
-PKGREL=1  # reset for new upstream pin (481279c — Phase 8.13 close)
+UPSTREAM_COMMIT=462aa4b4804972189d2f1b1ab5782f1ca1e8c423
+PKGVER=0.1.0+r22+g462aa4b
+PKGREL=1  # reset for new upstream pin (3dd0eb0 — DAEMON-PPS H.264 SPS/PPS NAL synth)
 
 HERE=$(dirname "$(readlink -f "$0")")
 

debian/daedalus-v4l2/debian/changelog

@@ -1,3 +1,36 @@
+daedalus-v4l2 (0.1.0+r22+g462aa4b-1) bookworm trixie; urgency=medium
+
+  * Bump to 462aa4b — kernel-side fix for control-binding gap that
+    closes the libva→daemon SPS/PPS pipeline.  Kernel device_run now
+    calls v4l2_ctrl_request_setup() before reading ctrl->p_cur, so
+    the daemon's daedalus_h264_meta block actually carries THIS
+    request's V4L2 stateless H.264 control values instead of stale
+    /default ones.  Pairs with libva-v4l2-request-fourier r382+gc1bb444
+    (Fix 3 + Fix 4 from issue libva-v4l2-request-fourier#8).
+  * After-fix on higgs (Pi CM5): ffmpeg -hwaccel vaapi -i h264.mp4
+    produces unique decoded P-frames (per-frame fnv1a hashes differ)
+    and zero "error while decoding MB" / "reference frames exceeds
+    max" warnings.
+
+ -- Markus Fritsche <mfritsche@reauktion.de>  Wed, 20 May 2026 22:00:00 +0000
+
+daedalus-v4l2 (0.1.0+r20+g3dd0eb0-1) bookworm trixie; urgency=medium
+
+  * Bump to 3dd0eb0 — DAEMON-PPS H.264 SPS/PPS NAL synthesiser.
+    Daemon now reconstructs AnnexB SPS+PPS NAL units from the V4L2
+    stateless H.264 control structs (forwarded by the kernel via
+    a new struct daedalus_h264_meta block in REQ_DECODE) and
+    prepends them to the slice bitstream before feeding libavcodec.
+    Without this, ffmpeg -hwaccel vaapi on H.264 sources failed
+    with "non-existing PPS 0 referenced" even after LIBVA-1/-2
+    routing correctly delivered the request.
+  * Wire protocol: new DAEDALUS_REQ_FLAG_H264_META bit + struct
+    daedalus_h264_meta; daemon and kernel must be installed in
+    lockstep (this package + daedalus-v4l2-dkms 0.1.0+r20+g3dd0eb0).
+  * VP9 / AV1 paths unchanged.
+
+ -- Markus Fritsche <mfritsche@reauktion.de>  Wed, 20 May 2026 21:00:00 +0000
+
 daedalus-v4l2 (0.1.0+r18+g481279c-1) bookworm trixie; urgency=medium
 
   * Bump to 481279c.  Upstream landed the systemd unit + modules-load.d

debian/libva-v4l2-request-fourier/build-deb.sh

@@ -10,15 +10,21 @@
 # Upstream fork: https://git.reauktion.de/marfrit/libva-v4l2-request-fourier
 set -euo pipefail
 
-# Same pin as the Arch PKGBUILD.  9898331 = LIBVA-2 close — completes
-# the per-codec dispatch from c332d34 (LIBVA-1) by adding video_fd_
-# daedalus to any_fd_supports_output_format's probe array.  Without
-# it, H.264/VP9/AV1 profiles never got advertised on Pi 5 mixed
-# deployments (rpi-hevc-dec primary, daedalus_v4l2 alt) — ffmpeg
-# bailed with "No support for codec h264 profile 578" before the
-# per-codec dispatch could even fire.
-UPSTREAM_COMMIT=989833114a7708ad999dc68309cbc181d9913bdb
-PKGVER=1.0.0+r380+g9898331
+# Same pin as the Arch PKGBUILD.  c1bb444 = PR #9 merge "h264:
+# max_num_ref_frames fallback + libva-boundary instrumentation (#8)" —
+# addresses the libva-side portion of marfrit/libva-v4l2-request-fourier#8.
+# The daedalus_v4l2-via-libavcodec strict consumer rejected frames whose
+# sps.max_num_ref_frames was 0 (older ffmpeg-vaapi paths leave the field
+# unset and HW decoders tolerated it).  PR adds a per-profile spec
+# minimum fallback (counts valid DPB entries first; 1 for baseline / 4
+# for main+high if even that is 0) and a one-line request_log at
+# h264_set_controls entry dumping raw VAAPI bitfields for diagnosing
+# the still-open PPS-flag-zero portion of issue #8.
+#
+# Prior pin (9898331) was the LIBVA-2 close — config-side
+# video_fd_daedalus added to the profile enumeration probe array.
+UPSTREAM_COMMIT=c1bb444d0731ca7ca831c7e01ec4b52ed736ca8e
+PKGVER=1.0.0+r382+gc1bb444
 PKGREL=1
 
 HERE=$(dirname "$(readlink -f "$0")")