phase 1: promote vb2_dma_resv RFC v2 + add ka-status + ampere as 2nd aarch64 host
Three changes that together flip kernel-agent from spec'd to operational
in the manual-orchestrated form. Real ka-* CLI verbs come in later phases;
this commit gets a first iteration through the pipeline and proves the
flow at the artifact level.
1. Promote vb2_dma_resv RFC v2 series into the scope-tagged tree
Markus iterated v2 locally on boltzmann (kernel-agent-bootstrap dir,
reaching linux-fresnel-fourier pkgrel=14). v2 attaches the producer
fence at device_run in slept-OK context per Dufresne's v1 review on
linux-media. The three patches land under
patches/subsystem/media/videobuf2/dma-resv-release-fence/:
- 0004 (helper) — opt-in vb2 dma_resv producer-fence helper
- 0005 (driver opt-in) — hantro device_run attach
- 0006 (driver opt-in) — rockchip-rga device_run attach
Numbered 4/5/6 because the fresnel build PKGBUILD applies them after
the three 0001/0002/0003 PBP DTS patches; this directory's numbering
follows that apply-order, not the upstream lore series numbering.
README at the scope dir documents fleet eligibility, decision history,
and the v1 → v2 design pivot.
2. Update fleet/fresnel.yaml to include the v2 series
Pre-v2 manifest had a comment block 'Explicitly NOT included … vb2
dma-resv-release-fence … defer until v2 lands'. v2 has landed. Move
those three lines from 'excluded' to 'includes', annotate the decision
inline.
3. README updates
- Build hosts table: add ampere (CoolPi GenBook, RK3588 32GB) as
secondary aarch64 host. Same uarch as boltzmann, on-demand wake via
His. Gives the fleet a second native build target for when boltzmann
is busy (e.g. carrying a firefox-fourier 4h build).
- 'Out of scope this round' bootstrap section: mark vb2_dma_resv as
resolved 2026-05-15, keep panfrost IOMMU_CACHE deferred.
4. First ka-* CLI verb implemented: bin/ka-status
bash, ~120 lines. Reads fleet/*.yaml manifests, queries Gitea for
open [ka:*] issues, probes each reachable host for the installed
kernel-package version. Read-only — no sudo, no host writes. Picks
GITEA_TOKEN from /opt/herding/etc/claude-identities/<host>.creds or
env override.
Proves the agent's Gitea-API + manifest-parsing skeleton works
end-to-end without committing to a full ka-promote/build/install
implementation. Smoke-tested locally:
$ bin/ka-status
kernel-agent status (repo: marfrit/kernel-agent)
open [ka:*] issues total: 1
══ fresnel ══
manifest: arch=arm64 soc=rockchip/rk3399 board=pinebook-pro
package: linux-fresnel-fourier
installed: host-down # (fresnel is currently powered off)
open ka-issues: (none for this host)
No PKGBUILD update in this PR — that lives in marfrit-packages and
ships as a sibling PR (the actual linux-fresnel-fourier-7.0-14 publish).
This commit is contained in:
@@ -115,7 +115,7 @@ ka-pause-prune / ka-resume-prune
|
|||||||
ka-restore-archive <job-id>
|
ka-restore-archive <job-id>
|
||||||
ka-snooze <issue-id> [--for <duration>]
|
ka-snooze <issue-id> [--for <duration>]
|
||||||
ka-debug <job-id> # shells into the same container that ran the build
|
ka-debug <job-id> # shells into the same container that ran the build
|
||||||
ka-status # per-host one-liner with drift/pending state
|
ka-status # per-host one-liner with drift/pending state [bin/ka-status — implemented Phase 1]
|
||||||
ka-migrate-tree --from <p> --to <p>
|
ka-migrate-tree --from <p> --to <p>
|
||||||
ka-wake-data # wraps wake-host data through His
|
ka-wake-data # wraps wake-host data through His
|
||||||
```
|
```
|
||||||
@@ -162,11 +162,13 @@ manifest rewrite); paths stable otherwise.
|
|||||||
## Build hosts
|
## Build hosts
|
||||||
|
|
||||||
```
|
```
|
||||||
Host Where Role Wake? Notes
|
Host Where Role Wake? Notes
|
||||||
──────────────────────────────────────────────────────────────────────────
|
──────────────────────────────────────────────────────────────────────────
|
||||||
boltzmann Rock 5 ITX+ aarch64 primary always container kbuild-aarch64
|
boltzmann Rock 5 ITX+ aarch64 primary always container kbuild-aarch64
|
||||||
fermi hertz LXD aarch64 fallback always matches kbuild-aarch64 profile
|
ampere CoolPi GenBook aarch64 secondary on-demand RK3588 32GB; same uarch as boltzmann,
|
||||||
kbuild-x86 data CT x86_64 on-demand wakes via His; idle 30 min → release
|
wakes via His; idle 30 min → release
|
||||||
|
fermi hertz LXD aarch64 fallback always matches kbuild-aarch64 profile
|
||||||
|
kbuild-x86 data CT x86_64 on-demand wakes via His; idle 30 min → release
|
||||||
```
|
```
|
||||||
|
|
||||||
Native make on the assigned build host. **No distcc** for kernel-agent
|
Native make on the assigned build host. **No distcc** for kernel-agent
|
||||||
@@ -288,10 +290,16 @@ build.
|
|||||||
|
|
||||||
### Out of scope this round (explicit defer)
|
### Out of scope this round (explicit defer)
|
||||||
|
|
||||||
- **vb2 dma_resv RFC v2** + panfrost IOMMU_CACHE for RK3399 — would have closed
|
- **vb2 dma_resv RFC v2** — *resolved 2026-05-15.* Markus iterated v2 locally
|
||||||
the fresnel-fourier campaign criterion-4 readback transitive-proof gap, but
|
on boltzmann reaching pkgrel=14; the v2 series attaches the fence at
|
||||||
v2 isn't implemented (RFC v1 rejected upstream). Deferred to a follow-up
|
`device_run` (slept-OK context per Dufresne's v1 review). Now carried in
|
||||||
build once v2 lands. See `marfrit/dmabuf-modifier-triage#3`.
|
`patches/subsystem/media/videobuf2/dma-resv-release-fence/` and included
|
||||||
|
in `fleet/fresnel.yaml`. Still in scope for upstream targeting; default
|
||||||
|
remains "build-tree only, no PR until explicitly asked"
|
||||||
|
(`feedback_no_upstream.md`).
|
||||||
|
- **panfrost IOMMU_CACHE for RK3399** — sibling kernel work that targets the
|
||||||
|
readback transitive-proof gap that vb2_dma_resv alone doesn't close.
|
||||||
|
Still deferred until that lands; ship together when ready.
|
||||||
- **Replace** `linux-eos-arm` rather than coexist alongside — preserves easy
|
- **Replace** `linux-eos-arm` rather than coexist alongside — preserves easy
|
||||||
rollback at u-boot. Can flip to `provides=(linux-eos-arm) conflicts=(...)`
|
rollback at u-boot. Can flip to `provides=(linux-eos-arm) conflicts=(...)`
|
||||||
later once burn-in proves the OC kernel reliable.
|
later once burn-in proves the OC kernel reliable.
|
||||||
|
|||||||
Executable
+136
@@ -0,0 +1,136 @@
|
|||||||
|
#!/usr/bin/env bash
|
||||||
|
#
|
||||||
|
# ka-status — per-host kernel-agent state summary.
|
||||||
|
#
|
||||||
|
# Reads fleet/*.yaml manifests + queries Gitea for open [ka:*] issues +
|
||||||
|
# probes each host (where reachable) for the installed kernel-package
|
||||||
|
# version. Designed to give a first-look "what's the state" before any
|
||||||
|
# ka-promote / ka-install action.
|
||||||
|
#
|
||||||
|
# Usage:
|
||||||
|
# ka-status # summary across all manifests
|
||||||
|
# ka-status <host> # detail for one host
|
||||||
|
#
|
||||||
|
# Read-only. Never mutates state. No sudo. No SSH-into-host writes.
|
||||||
|
#
|
||||||
|
# Phase 1 deliverable. Future ka-* CLI verbs (ka-promote / ka-close /
|
||||||
|
# ka-install) build on the same Gitea-API + manifest-parsing skeleton.
|
||||||
|
|
||||||
|
set -euo pipefail
|
||||||
|
|
||||||
|
GITEA_URL="${GITEA_URL:-https://git.reauktion.de}"
|
||||||
|
REPO="${KERNEL_AGENT_REPO:-marfrit/kernel-agent}"
|
||||||
|
TOKEN_FILE="${KERNEL_AGENT_TOKEN_FILE:-/opt/herding/etc/claude-identities/noether.creds}"
|
||||||
|
|
||||||
|
# Resolve token from per-host claude-identity creds, or env override.
|
||||||
|
token=""
|
||||||
|
if [ -n "${GITEA_TOKEN:-}" ]; then
|
||||||
|
token="$GITEA_TOKEN"
|
||||||
|
elif [ -r "$TOKEN_FILE" ]; then
|
||||||
|
token="$(grep -E '^GITEA_TOKEN=' "$TOKEN_FILE" | head -1 | cut -d= -f2)"
|
||||||
|
fi
|
||||||
|
|
||||||
|
# Locate fleet/ — script lives in bin/ next to fleet/.
|
||||||
|
script_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
|
||||||
|
fleet_dir="${script_dir}/../fleet"
|
||||||
|
[ -d "$fleet_dir" ] || { echo "fleet/ not found relative to $script_dir" >&2; exit 2; }
|
||||||
|
|
||||||
|
api_get() {
|
||||||
|
local path="$1"
|
||||||
|
local args=(--silent --show-error --max-time 15)
|
||||||
|
[ -n "$token" ] && args+=(-H "Authorization: token $token")
|
||||||
|
curl "${args[@]}" "${GITEA_URL}/api/v1/${path}"
|
||||||
|
}
|
||||||
|
|
||||||
|
# Open ka-prefixed issues, JSON array on stdout.
|
||||||
|
fetch_ka_issues() {
|
||||||
|
api_get "repos/${REPO}/issues?state=open&type=issues&limit=50" \
|
||||||
|
| python3 -c '
|
||||||
|
import json, sys
|
||||||
|
issues = json.load(sys.stdin)
|
||||||
|
def kind(t):
|
||||||
|
if not t.startswith("["): return None
|
||||||
|
p = t.find("]")
|
||||||
|
return t[1:p] if p > 0 else None
|
||||||
|
out = [{
|
||||||
|
"number": i["number"],
|
||||||
|
"title": i["title"],
|
||||||
|
"kind": kind(i["title"]),
|
||||||
|
} for i in issues if kind(i["title"]) and kind(i["title"]).startswith("ka:")]
|
||||||
|
print(json.dumps(out))
|
||||||
|
' 2>/dev/null || echo "[]"
|
||||||
|
}
|
||||||
|
|
||||||
|
# Parse a fleet/<host>.yaml manifest (very-narrow YAML subset).
|
||||||
|
manifest_field() {
|
||||||
|
local file="$1" key="$2"
|
||||||
|
grep -E "^[[:space:]]*${key}:" "$file" | head -1 | sed -E "s/^[^:]*:[[:space:]]*//; s/[[:space:]]+#.*//; s/^[\"']//; s/[\"']$//"
|
||||||
|
}
|
||||||
|
|
||||||
|
manifest_pkgname() {
|
||||||
|
local file="$1"
|
||||||
|
awk '/^package:/{p=1; next} p && /^[a-z]/{p=0} p && /^[[:space:]]+name:/{sub(/^[[:space:]]+name:[[:space:]]*/,""); print; exit}' "$file"
|
||||||
|
}
|
||||||
|
|
||||||
|
# Probe a host for installed kernel-package version (best-effort, non-blocking).
|
||||||
|
probe_installed() {
|
||||||
|
local host="$1" pkg="$2"
|
||||||
|
ssh -o ConnectTimeout=3 -o BatchMode=yes -o StrictHostKeyChecking=accept-new \
|
||||||
|
"${host}.fritz.box" "pacman -Q '$pkg' 2>/dev/null || dpkg-query -W -f='\${Package} \${Version}\n' '$pkg' 2>/dev/null || echo 'host-up:not-installed'" 2>/dev/null \
|
||||||
|
|| echo "host-down"
|
||||||
|
}
|
||||||
|
|
||||||
|
issues_json="$(fetch_ka_issues)"
|
||||||
|
|
||||||
|
# Per-host issue grouping — match on title containing the host name (cheap heuristic;
|
||||||
|
# proper kernel-agent will tag issues with a host label).
|
||||||
|
issues_for_host() {
|
||||||
|
local host="$1"
|
||||||
|
echo "$issues_json" | python3 -c "
|
||||||
|
import json, sys
|
||||||
|
host = '$host'
|
||||||
|
issues = json.load(sys.stdin)
|
||||||
|
hits = [i for i in issues if host in i['title'].lower()]
|
||||||
|
for h in hits:
|
||||||
|
print(f\" #{h['number']} [{h['kind']}] {h['title']}\")
|
||||||
|
" 2>/dev/null
|
||||||
|
}
|
||||||
|
|
||||||
|
print_host() {
|
||||||
|
local file="$1"
|
||||||
|
local host pkg
|
||||||
|
host="$(basename "$file" .yaml)"
|
||||||
|
pkg="$(manifest_pkgname "$file")"
|
||||||
|
local arch="$(manifest_field "$file" arch)"
|
||||||
|
local soc="$(manifest_field "$file" soc)"
|
||||||
|
local board="$(manifest_field "$file" board)"
|
||||||
|
|
||||||
|
printf '\n══ %s ══\n' "$host"
|
||||||
|
printf ' manifest: arch=%s soc=%s board=%s\n' "$arch" "$soc" "$board"
|
||||||
|
printf ' package: %s\n' "$pkg"
|
||||||
|
if [ -n "$pkg" ]; then
|
||||||
|
printf ' installed: %s\n' "$(probe_installed "$host" "$pkg")"
|
||||||
|
fi
|
||||||
|
|
||||||
|
local n=0
|
||||||
|
while IFS= read -r line; do
|
||||||
|
[ -z "$line" ] && continue
|
||||||
|
if [ $n -eq 0 ]; then printf ' open ka-issues:\n'; fi
|
||||||
|
printf '%s\n' "$line"
|
||||||
|
n=$((n+1))
|
||||||
|
done < <(issues_for_host "$host")
|
||||||
|
[ $n -eq 0 ] && printf ' open ka-issues: (none for this host)\n'
|
||||||
|
}
|
||||||
|
|
||||||
|
if [ $# -ge 1 ]; then
|
||||||
|
f="${fleet_dir}/${1}.yaml"
|
||||||
|
[ -r "$f" ] || { echo "no manifest for host '$1'" >&2; exit 2; }
|
||||||
|
print_host "$f"
|
||||||
|
else
|
||||||
|
printf 'kernel-agent status (repo: %s)\n' "$REPO"
|
||||||
|
printf 'open [ka:*] issues total: %s\n' "$(echo "$issues_json" | python3 -c 'import json,sys; print(len(json.load(sys.stdin)))')"
|
||||||
|
for f in "$fleet_dir"/*.yaml; do
|
||||||
|
[ -e "$f" ] || continue
|
||||||
|
print_host "$f"
|
||||||
|
done
|
||||||
|
fi
|
||||||
+8
-4
@@ -23,13 +23,17 @@ includes:
|
|||||||
- board/pinebook-pro/0001-arm64-dts-rk3399-pinebook-pro-add-OC-OPP-tables-1704-2184.patch
|
- board/pinebook-pro/0001-arm64-dts-rk3399-pinebook-pro-add-OC-OPP-tables-1704-2184.patch
|
||||||
- board/pinebook-pro/0002-arm64-dts-rk3399-pinebook-pro-enable-hdmi-sound.patch
|
- board/pinebook-pro/0002-arm64-dts-rk3399-pinebook-pro-enable-hdmi-sound.patch
|
||||||
- board/pinebook-pro/0003-arm64-dts-rk3399-pinebook-pro-spi1-max-freq-10MHz.patch
|
- board/pinebook-pro/0003-arm64-dts-rk3399-pinebook-pro-spi1-max-freq-10MHz.patch
|
||||||
|
# vb2_dma_resv RFC v2 series — added 2026-05-15 (Markus iterated v2 locally
|
||||||
|
# on boltzmann reaching pkgrel=14; pre-v2 decision was "defer". v2 attaches
|
||||||
|
# the fence at device_run in slept-OK context per Dufresne's v1 review).
|
||||||
|
- subsystem/media/videobuf2/dma-resv-release-fence/0004-media-videobuf2-add-opt-in-dma_resv-producer-fence-h.patch
|
||||||
|
- subsystem/media/videobuf2/dma-resv-release-fence/0005-media-hantro-attach-dma_resv-release-fence-at-device.patch
|
||||||
|
- subsystem/media/videobuf2/dma-resv-release-fence/0006-media-rockchip-rga-attach-dma_resv-release-fence-at-.patch
|
||||||
|
|
||||||
# Explicitly NOT included (tracked elsewhere, decision logged):
|
# Explicitly NOT included (tracked elsewhere, decision logged):
|
||||||
# - subsystem/media/videobuf2/dma-resv-release-fence/ (RFC v1 rejected;
|
|
||||||
# v2 in design — see marfrit/dmabuf-modifier-triage#3. Skip until v2 lands
|
|
||||||
# or we explicitly accept v1-shape parity with ohm.)
|
|
||||||
# - driver/panfrost/iommu-cache-rk3399/ (sibling kernel work; ship together
|
# - driver/panfrost/iommu-cache-rk3399/ (sibling kernel work; ship together
|
||||||
# with vb2_dma_resv when it lands.)
|
# once it lands. Targets the readback transitive-proof gap that vb2_dma_resv
|
||||||
|
# alone doesn't close.)
|
||||||
|
|
||||||
config:
|
config:
|
||||||
source: /proc/config.gz on running fresnel kernel (linux-eos-arm 6.19.9-99)
|
source: /proc/config.gz on running fresnel kernel (linux-eos-arm 6.19.9-99)
|
||||||
|
|||||||
+356
@@ -0,0 +1,356 @@
|
|||||||
|
From a202de1646d4c8f8ee2ebc2e4c100b621975754a Mon Sep 17 00:00:00 2001
|
||||||
|
In-Reply-To: <20260429195306.239666-1-mfritsche@reauktion.de>
|
||||||
|
References: <20260429195306.239666-1-mfritsche@reauktion.de>
|
||||||
|
From: Markus Fritsche <mfritsche@reauktion.de>
|
||||||
|
Date: Sat, 9 May 2026 16:16:07 +0200
|
||||||
|
Subject: [PATCH RFC v2] media: videobuf2: add opt-in dma_resv producer fence
|
||||||
|
helper
|
||||||
|
MIME-Version: 1.0
|
||||||
|
Content-Type: text/plain; charset=UTF-8
|
||||||
|
Content-Transfer-Encoding: 8bit
|
||||||
|
|
||||||
|
V4L2 producers historically don't propagate buffer-state-done into
|
||||||
|
the dmabuf's dma_resv exclusive fence. Userspace consumers that
|
||||||
|
import V4L2-produced dmabufs and wait on the dmabuf's implicit-sync
|
||||||
|
fence (poll(POLLIN), DMA_BUF_IOCTL_EXPORT_SYNC_FILE,
|
||||||
|
EGL_LINUX_DMA_BUF_EXT) currently see either zero fences or a stub
|
||||||
|
fence from dma_fence_get_stub(). This is correct by accident for the
|
||||||
|
common DQBUF-then-import case but represents a contract gap that
|
||||||
|
breaks Wayland compositors importing CAPTURE buffers from a stateless
|
||||||
|
H.264 decoder under continuous playback on implicit-sync GPU stacks
|
||||||
|
(observed on RK3566 + hantro VPU + Mali-G52 panfrost; manifests as
|
||||||
|
green frames -- BT.709 limited-range YUV(0,0,0) -> RGB(0,77,0) -- when
|
||||||
|
the GPU samples the dmabuf before the producer's decode completes).
|
||||||
|
|
||||||
|
Add an opt-in API gated by both a per-driver runtime flag
|
||||||
|
(vb2_queue::supports_release_fences) and a Kconfig
|
||||||
|
(CONFIG_VIDEOBUF2_RELEASE_FENCES, default n) that lets producers
|
||||||
|
populate a real dma_resv exclusive write fence on the dmabufs they
|
||||||
|
export. Drivers call vb2_buffer_attach_release_fence(vb) at a
|
||||||
|
finite-time-fenced point in their pipeline (typically m2m
|
||||||
|
device_run, just before the HW kick); vb2_buffer_done() signals and
|
||||||
|
puts the fence as part of its state transition.
|
||||||
|
|
||||||
|
The publish and signal paths are wrapped in
|
||||||
|
dma_fence_begin_signalling() / dma_fence_end_signalling() so
|
||||||
|
PROVE_LOCKING can validate that nothing taken in those critical
|
||||||
|
sections deadlocks against the signal path. dma_resv_lock is
|
||||||
|
sleepable but not taken on the signal path, so taking it inside the
|
||||||
|
publish critical section is safe under lockdep.
|
||||||
|
|
||||||
|
Skips planes whose vb2_plane.dbuf is NULL -- buffers never exported
|
||||||
|
via VIDIOC_EXPBUF (or imported via V4L2_MEMORY_DMABUF) have no
|
||||||
|
dmabuf for userspace to wait on.
|
||||||
|
|
||||||
|
Drivers that don't opt in pay nothing: the helper is a no-op stub
|
||||||
|
when CONFIG_VIDEOBUF2_RELEASE_FENCES=n, and an early-return check
|
||||||
|
of supports_release_fences when =y but the flag is unset.
|
||||||
|
|
||||||
|
Validated on RK3566 PineTab2 with PROVE_LOCKING enabled: 30s of
|
||||||
|
bbb_1080p30 H.264 stateless decode + zero-copy panfrost EGL import
|
||||||
|
via dmabuf-wayland (mpv 0.41 + KWin 6.6.4 + Mesa panfrost 26.0.5)
|
||||||
|
produces 31,816 dma_fence init/signal pairs across 5,724 vb2 buffer
|
||||||
|
cycles with zero lockdep splats from videobuf2 / dma_resv code paths.
|
||||||
|
|
||||||
|
Subsequent patches in this series opt the hantro and rockchip-rga
|
||||||
|
drivers in.
|
||||||
|
|
||||||
|
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
|
||||||
|
Cc: Christian König <christian.koenig@amd.com>
|
||||||
|
Cc: Nicolas Dufresne <nicolas@ndufresne.ca>
|
||||||
|
Cc: Sumit Semwal <sumit.semwal@linaro.org>
|
||||||
|
Cc: Hans Verkuil <hverkuil@xs4all.nl>
|
||||||
|
Cc: Tomasz Figa <tfiga@chromium.org>
|
||||||
|
Cc: linux-media@vger.kernel.org
|
||||||
|
Cc: dri-devel@lists.freedesktop.org
|
||||||
|
Cc: linaro-mm-sig@lists.linaro.org
|
||||||
|
Signed-off-by: Markus Fritsche <mfritsche@reauktion.de>
|
||||||
|
---
|
||||||
|
drivers/media/common/videobuf2/Kconfig | 29 ++++
|
||||||
|
.../media/common/videobuf2/videobuf2-core.c | 135 ++++++++++++++++++
|
||||||
|
include/media/videobuf2-core.h | 51 +++++++
|
||||||
|
3 files changed, 215 insertions(+)
|
||||||
|
|
||||||
|
diff --git a/drivers/media/common/videobuf2/Kconfig b/drivers/media/common/videobuf2/Kconfig
|
||||||
|
index d2223a12c..bbfa26984 100644
|
||||||
|
--- a/drivers/media/common/videobuf2/Kconfig
|
||||||
|
+++ b/drivers/media/common/videobuf2/Kconfig
|
||||||
|
@@ -30,3 +30,32 @@ config VIDEOBUF2_DMA_SG
|
||||||
|
config VIDEOBUF2_DVB
|
||||||
|
tristate
|
||||||
|
select VIDEOBUF2_CORE
|
||||||
|
+
|
||||||
|
+config VIDEOBUF2_RELEASE_FENCES
|
||||||
|
+ bool "videobuf2: opt-in dma_resv producer fences for V4L2 dmabuf exports"
|
||||||
|
+ depends on VIDEOBUF2_CORE
|
||||||
|
+ depends on DMA_SHARED_BUFFER
|
||||||
|
+ default n
|
||||||
|
+ help
|
||||||
|
+ Enables an opt-in API that lets vb2 producers populate a dma_resv
|
||||||
|
+ exclusive write fence on the dmabufs they export to userspace.
|
||||||
|
+ The fence is signalled when the buffer transitions to
|
||||||
|
+ VB2_BUF_STATE_DONE.
|
||||||
|
+
|
||||||
|
+ This gives userspace consumers that import V4L2-produced dmabufs
|
||||||
|
+ and wait on the dmabuf's implicit-sync fence (poll(POLLIN),
|
||||||
|
+ DMA_BUF_IOCTL_EXPORT_SYNC_FILE, EGL_LINUX_DMA_BUF_EXT) a real
|
||||||
|
+ producer fence to wait on, instead of a stub fence from
|
||||||
|
+ dma_fence_get_stub() that the dma_buf core substitutes when
|
||||||
|
+ dma_resv is empty.
|
||||||
|
+
|
||||||
|
+ Drivers individually opt in by setting
|
||||||
|
+ vb2_queue::supports_release_fences = true and calling
|
||||||
|
+ vb2_buffer_attach_release_fence() at the right point in their
|
||||||
|
+ pipeline (typically m2m device_run, just before HW kick).
|
||||||
|
+
|
||||||
|
+ Distributors leave this off unless targeting Wayland/EGL
|
||||||
|
+ consumers of V4L2 stateless decoder output on
|
||||||
|
+ implicit-sync-only GPU stacks (e.g. mainline panfrost).
|
||||||
|
+
|
||||||
|
+ If unsure, say N.
|
||||||
|
diff --git a/drivers/media/common/videobuf2/videobuf2-core.c b/drivers/media/common/videobuf2/videobuf2-core.c
|
||||||
|
index adf668b21..85d7fddbd 100644
|
||||||
|
--- a/drivers/media/common/videobuf2/videobuf2-core.c
|
||||||
|
+++ b/drivers/media/common/videobuf2/videobuf2-core.c
|
||||||
|
@@ -26,6 +26,12 @@
|
||||||
|
#include <linux/freezer.h>
|
||||||
|
#include <linux/kthread.h>
|
||||||
|
|
||||||
|
+#ifdef CONFIG_VIDEOBUF2_RELEASE_FENCES
|
||||||
|
+#include <linux/dma-fence.h>
|
||||||
|
+#include <linux/dma-resv.h>
|
||||||
|
+#include <linux/dma-buf.h>
|
||||||
|
+#endif
|
||||||
|
+
|
||||||
|
#include <media/videobuf2-core.h>
|
||||||
|
#include <media/v4l2-mc.h>
|
||||||
|
|
||||||
|
@@ -1173,6 +1179,120 @@ void *vb2_plane_cookie(struct vb2_buffer *vb, unsigned int plane_no)
|
||||||
|
}
|
||||||
|
EXPORT_SYMBOL_GPL(vb2_plane_cookie);
|
||||||
|
|
||||||
|
+#ifdef CONFIG_VIDEOBUF2_RELEASE_FENCES
|
||||||
|
+/*
|
||||||
|
+ * dma_resv release-fence integration.
|
||||||
|
+ *
|
||||||
|
+ * Optional, opt-in path that lets producers publish a real
|
||||||
|
+ * dma_fence on their CAPTURE-side dmabufs so userspace consumers
|
||||||
|
+ * (compositors, EGL importers) get spec-clean implicit-sync
|
||||||
|
+ * semantics instead of the dma_buf core's stub fence. Drivers
|
||||||
|
+ * call vb2_buffer_attach_release_fence() at a finite-time-fenced
|
||||||
|
+ * point (typically m2m device_run) and the fence is signalled by
|
||||||
|
+ * vb2_buffer_done(). Gated at runtime by
|
||||||
|
+ * vb2_queue::supports_release_fences and at compile time by
|
||||||
|
+ * CONFIG_VIDEOBUF2_RELEASE_FENCES.
|
||||||
|
+ */
|
||||||
|
+
|
||||||
|
+static const char *vb2_dma_resv_get_driver_name(struct dma_fence *fence)
|
||||||
|
+{
|
||||||
|
+ return "videobuf2";
|
||||||
|
+}
|
||||||
|
+
|
||||||
|
+static const char *vb2_dma_resv_get_timeline_name(struct dma_fence *fence)
|
||||||
|
+{
|
||||||
|
+ return "vb2-release-fence";
|
||||||
|
+}
|
||||||
|
+
|
||||||
|
+static const struct dma_fence_ops vb2_dma_resv_fence_ops = {
|
||||||
|
+ .get_driver_name = vb2_dma_resv_get_driver_name,
|
||||||
|
+ .get_timeline_name = vb2_dma_resv_get_timeline_name,
|
||||||
|
+};
|
||||||
|
+
|
||||||
|
+int vb2_buffer_attach_release_fence(struct vb2_buffer *vb)
|
||||||
|
+{
|
||||||
|
+ struct vb2_queue *q = vb->vb2_queue;
|
||||||
|
+ struct dma_fence *fence;
|
||||||
|
+ unsigned int plane;
|
||||||
|
+ bool cookie;
|
||||||
|
+
|
||||||
|
+ if (!q->supports_release_fences)
|
||||||
|
+ return 0;
|
||||||
|
+
|
||||||
|
+ if (WARN_ON(vb->release_fence))
|
||||||
|
+ return -EINVAL;
|
||||||
|
+
|
||||||
|
+ fence = kzalloc(sizeof(*fence), GFP_KERNEL);
|
||||||
|
+ if (!fence)
|
||||||
|
+ return -ENOMEM;
|
||||||
|
+
|
||||||
|
+ dma_fence_init(fence, &vb2_dma_resv_fence_ops, &q->dma_resv_fence_lock,
|
||||||
|
+ q->dma_resv_fence_context,
|
||||||
|
+ atomic64_inc_return(&q->dma_resv_fence_seqno));
|
||||||
|
+
|
||||||
|
+ /*
|
||||||
|
+ * Annotate the publish-side critical section. Per
|
||||||
|
+ * Documentation/driver-api/dma-buf.rst, lockdep validates
|
||||||
|
+ * that nothing taken in this region can deadlock against
|
||||||
|
+ * the signal path in vb2_buffer_signal_release_fence().
|
||||||
|
+ * dma_resv_lock is sleepable but is not taken on the signal
|
||||||
|
+ * path, so taking it inside the critical section is safe.
|
||||||
|
+ */
|
||||||
|
+ cookie = dma_fence_begin_signalling();
|
||||||
|
+ for (plane = 0; plane < vb->num_planes; plane++) {
|
||||||
|
+ struct dma_buf *dbuf = vb->planes[plane].dbuf;
|
||||||
|
+
|
||||||
|
+ if (!dbuf)
|
||||||
|
+ continue;
|
||||||
|
+
|
||||||
|
+ dma_resv_lock(dbuf->resv, NULL);
|
||||||
|
+ dma_resv_add_fence(dbuf->resv, fence, DMA_RESV_USAGE_WRITE);
|
||||||
|
+ dma_resv_unlock(dbuf->resv);
|
||||||
|
+ }
|
||||||
|
+ dma_fence_end_signalling(cookie);
|
||||||
|
+
|
||||||
|
+ /* One reference for the eventual signal in vb2_buffer_done. */
|
||||||
|
+ vb->release_fence = dma_fence_get(fence);
|
||||||
|
+
|
||||||
|
+ /* The dma_resv held its own reference per plane. Drop ours. */
|
||||||
|
+ dma_fence_put(fence);
|
||||||
|
+
|
||||||
|
+ return 0;
|
||||||
|
+}
|
||||||
|
+EXPORT_SYMBOL_GPL(vb2_buffer_attach_release_fence);
|
||||||
|
+
|
||||||
|
+static void vb2_buffer_signal_release_fence(struct vb2_buffer *vb,
|
||||||
|
+ enum vb2_buffer_state state)
|
||||||
|
+{
|
||||||
|
+ struct dma_fence *fence = vb->release_fence;
|
||||||
|
+ bool cookie;
|
||||||
|
+
|
||||||
|
+ if (!fence)
|
||||||
|
+ return;
|
||||||
|
+
|
||||||
|
+ cookie = dma_fence_begin_signalling();
|
||||||
|
+ if (state == VB2_BUF_STATE_ERROR)
|
||||||
|
+ dma_fence_set_error(fence, -EIO);
|
||||||
|
+ dma_fence_signal(fence);
|
||||||
|
+ dma_fence_end_signalling(cookie);
|
||||||
|
+
|
||||||
|
+ dma_fence_put(fence);
|
||||||
|
+ vb->release_fence = NULL;
|
||||||
|
+}
|
||||||
|
+#else /* !CONFIG_VIDEOBUF2_RELEASE_FENCES */
|
||||||
|
+
|
||||||
|
+int vb2_buffer_attach_release_fence(struct vb2_buffer *vb)
|
||||||
|
+{
|
||||||
|
+ return 0;
|
||||||
|
+}
|
||||||
|
+EXPORT_SYMBOL_GPL(vb2_buffer_attach_release_fence);
|
||||||
|
+
|
||||||
|
+static inline void vb2_buffer_signal_release_fence(struct vb2_buffer *vb,
|
||||||
|
+ enum vb2_buffer_state state)
|
||||||
|
+{
|
||||||
|
+}
|
||||||
|
+#endif /* CONFIG_VIDEOBUF2_RELEASE_FENCES */
|
||||||
|
+
|
||||||
|
void vb2_buffer_done(struct vb2_buffer *vb, enum vb2_buffer_state state)
|
||||||
|
{
|
||||||
|
struct vb2_queue *q = vb->vb2_queue;
|
||||||
|
@@ -1199,6 +1319,9 @@ void vb2_buffer_done(struct vb2_buffer *vb, enum vb2_buffer_state state)
|
||||||
|
if (state != VB2_BUF_STATE_QUEUED)
|
||||||
|
__vb2_buf_mem_finish(vb);
|
||||||
|
|
||||||
|
+ if (state != VB2_BUF_STATE_QUEUED)
|
||||||
|
+ vb2_buffer_signal_release_fence(vb, state);
|
||||||
|
+
|
||||||
|
spin_lock_irqsave(&q->done_lock, flags);
|
||||||
|
if (state == VB2_BUF_STATE_QUEUED) {
|
||||||
|
vb->state = VB2_BUF_STATE_QUEUED;
|
||||||
|
@@ -2651,6 +2774,18 @@ int vb2_core_queue_init(struct vb2_queue *q)
|
||||||
|
mutex_init(&q->mmap_lock);
|
||||||
|
init_waitqueue_head(&q->done_wq);
|
||||||
|
|
||||||
|
+#ifdef CONFIG_VIDEOBUF2_RELEASE_FENCES
|
||||||
|
+ /*
|
||||||
|
+ * Per-queue dma_resv release-fence context. Drivers that
|
||||||
|
+ * opt in via supports_release_fences and call
|
||||||
|
+ * vb2_buffer_attach_release_fence() use these to allocate
|
||||||
|
+ * fences on a single per-queue timeline.
|
||||||
|
+ */
|
||||||
|
+ q->dma_resv_fence_context = dma_fence_context_alloc(1);
|
||||||
|
+ atomic64_set(&q->dma_resv_fence_seqno, 0);
|
||||||
|
+ spin_lock_init(&q->dma_resv_fence_lock);
|
||||||
|
+#endif
|
||||||
|
+
|
||||||
|
q->memory = VB2_MEMORY_UNKNOWN;
|
||||||
|
|
||||||
|
if (q->buf_struct_size == 0)
|
||||||
|
diff --git a/include/media/videobuf2-core.h b/include/media/videobuf2-core.h
|
||||||
|
index 4424d481d..766ff2194 100644
|
||||||
|
--- a/include/media/videobuf2-core.h
|
||||||
|
+++ b/include/media/videobuf2-core.h
|
||||||
|
@@ -288,6 +288,16 @@ struct vb2_buffer {
|
||||||
|
unsigned int skip_cache_sync_on_finish:1;
|
||||||
|
|
||||||
|
struct vb2_plane planes[VB2_MAX_PLANES];
|
||||||
|
+#ifdef CONFIG_VIDEOBUF2_RELEASE_FENCES
|
||||||
|
+ /*
|
||||||
|
+ * Producer release fence published on each plane's
|
||||||
|
+ * dmabuf->resv when the driver opts in via
|
||||||
|
+ * vb2_buffer_attach_release_fence(). Signalled and put by
|
||||||
|
+ * vb2_buffer_done() on transition to DONE/ERROR. NULL when
|
||||||
|
+ * the driver did not opt in for this buffer.
|
||||||
|
+ */
|
||||||
|
+ struct dma_fence *release_fence;
|
||||||
|
+#endif
|
||||||
|
struct list_head queued_entry;
|
||||||
|
struct list_head done_entry;
|
||||||
|
#ifdef CONFIG_VIDEO_ADV_DEBUG
|
||||||
|
@@ -648,6 +658,19 @@ struct vb2_queue {
|
||||||
|
spinlock_t done_lock;
|
||||||
|
wait_queue_head_t done_wq;
|
||||||
|
|
||||||
|
+#ifdef CONFIG_VIDEOBUF2_RELEASE_FENCES
|
||||||
|
+ /*
|
||||||
|
+ * dma_resv release-fence context. Drivers that set
|
||||||
|
+ * supports_release_fences and call
|
||||||
|
+ * vb2_buffer_attach_release_fence() use these to allocate
|
||||||
|
+ * fences on a per-queue timeline.
|
||||||
|
+ */
|
||||||
|
+ u64 dma_resv_fence_context;
|
||||||
|
+ atomic64_t dma_resv_fence_seqno;
|
||||||
|
+ spinlock_t dma_resv_fence_lock;
|
||||||
|
+#endif
|
||||||
|
+
|
||||||
|
+ unsigned int supports_release_fences:1;
|
||||||
|
unsigned int streaming:1;
|
||||||
|
unsigned int start_streaming_called:1;
|
||||||
|
unsigned int error:1;
|
||||||
|
@@ -735,6 +758,34 @@ void *vb2_plane_cookie(struct vb2_buffer *vb, unsigned int plane_no);
|
||||||
|
*/
|
||||||
|
void vb2_buffer_done(struct vb2_buffer *vb, enum vb2_buffer_state state);
|
||||||
|
|
||||||
|
+/**
|
||||||
|
+ * vb2_buffer_attach_release_fence() - opt-in dma_resv release fence.
|
||||||
|
+ * @vb: the buffer being committed to the producer.
|
||||||
|
+ *
|
||||||
|
+ * Drivers that have set vb2_queue::supports_release_fences may call
|
||||||
|
+ * this from any sleepable context where they have committed to
|
||||||
|
+ * running the operation in finite time -- typically m2m
|
||||||
|
+ * device_run(), just before the HW kick. The helper allocates a
|
||||||
|
+ * dma_fence on the queue's per-queue timeline, attaches it as
|
||||||
|
+ * DMA_RESV_USAGE_WRITE on each plane's dmabuf->resv, and stashes
|
||||||
|
+ * it in vb->release_fence. vb2_buffer_done() signals and puts the
|
||||||
|
+ * fence as part of the buffer's state transition.
|
||||||
|
+ *
|
||||||
|
+ * Skips planes whose vb2_plane.dbuf is NULL -- buffers never
|
||||||
|
+ * exported via VIDIOC_EXPBUF (or imported via V4L2_MEMORY_DMABUF)
|
||||||
|
+ * have no dmabuf for userspace to wait on.
|
||||||
|
+ *
|
||||||
|
+ * No-op when vb2_queue::supports_release_fences is not set
|
||||||
|
+ * (regardless of CONFIG_VIDEOBUF2_RELEASE_FENCES). When
|
||||||
|
+ * CONFIG_VIDEOBUF2_RELEASE_FENCES=n, this is a stub that returns 0.
|
||||||
|
+ *
|
||||||
|
+ * Returns 0 on success or when the no-op stub is in effect,
|
||||||
|
+ * negative errno on allocation failure when fence publishing was
|
||||||
|
+ * attempted. Best-effort: drivers should ignore the return value
|
||||||
|
+ * unless they want diagnostics.
|
||||||
|
+ */
|
||||||
|
+int vb2_buffer_attach_release_fence(struct vb2_buffer *vb);
|
||||||
|
+
|
||||||
|
/**
|
||||||
|
* vb2_discard_done() - discard all buffers marked as DONE.
|
||||||
|
* @q: pointer to &struct vb2_queue with videobuf2 queue.
|
||||||
|
--
|
||||||
|
2.53.0
|
||||||
|
|
||||||
+95
@@ -0,0 +1,95 @@
|
|||||||
|
From 1844c263bde8dd244d7db46f8c508e7c70da459c Mon Sep 17 00:00:00 2001
|
||||||
|
In-Reply-To: <20260429195306.239666-1-mfritsche@reauktion.de>
|
||||||
|
References: <20260429195306.239666-1-mfritsche@reauktion.de>
|
||||||
|
From: Markus Fritsche <mfritsche@reauktion.de>
|
||||||
|
Date: Sat, 9 May 2026 16:24:01 +0200
|
||||||
|
Subject: [PATCH RFC v2] media: hantro: attach dma_resv release fence at
|
||||||
|
device_run
|
||||||
|
|
||||||
|
Opt the hantro driver into the new vb2 release-fence helper so its
|
||||||
|
CAPTURE-side dmabufs carry a real producer fence that wayland
|
||||||
|
compositors and other implicit-sync consumers can wait on, instead
|
||||||
|
of the dma_buf core's stub fence.
|
||||||
|
|
||||||
|
Attach point is m2m device_run, immediately after
|
||||||
|
v4l2_m2m_buf_copy_metadata() and before ctx->codec_ops->run().
|
||||||
|
Per Nicolas Dufresne's v1 review (lore.kernel.org/linux-media/
|
||||||
|
3d8deeb15581b754e4c061d4c4a13657aa08bc3c.camel@ndufresne.ca/),
|
||||||
|
this satisfies the dma_fence finite-time contract: the m2m core
|
||||||
|
has committed to running the job by this point, codec_ops->run
|
||||||
|
either kicks the HW (decode-complete signals the fence via
|
||||||
|
vb2_buffer_done) or fails immediately (job_finish with
|
||||||
|
VB2_BUF_STATE_ERROR signals with -EIO). PM and clocks are already
|
||||||
|
up by this point, so no allocation context restrictions.
|
||||||
|
|
||||||
|
The CAPTURE queue is opted in with supports_release_fences=true at
|
||||||
|
queue_init.
|
||||||
|
|
||||||
|
Userspace consumers that import hantro CAPTURE dmabufs and wait on
|
||||||
|
their implicit-sync fence (Wayland zwp_linux_dmabuf_v1 +
|
||||||
|
panfrost EGL_LINUX_DMA_BUF_EXT) now wait on a real fence
|
||||||
|
representing the producer's actual completion, fixing green-frame
|
||||||
|
corruption observed on RK3566 PineTab2 + Mali-G52 panfrost (the
|
||||||
|
GPU was sampling zero pages because the dmabuf's implicit fence
|
||||||
|
was the dma_buf core's pre-signalled stub).
|
||||||
|
|
||||||
|
Validated end-to-end on PineTab2 (RK3566 / hantro G1 / Mali-G52
|
||||||
|
mainline panfrost): 30s of bbb_1080p30 H.264 stateless decode +
|
||||||
|
zero-copy panfrost EGL import via dmabuf-wayland (mpv 0.41 +
|
||||||
|
KWin 6.6.4 + Mesa panfrost 26.0.5) renders correctly with no
|
||||||
|
green-frame corruption and no PROVE_LOCKING splats.
|
||||||
|
|
||||||
|
Cc: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>
|
||||||
|
Cc: Philipp Zabel <p.zabel@pengutronix.de>
|
||||||
|
Cc: Nicolas Dufresne <nicolas@ndufresne.ca>
|
||||||
|
Cc: linux-media@vger.kernel.org
|
||||||
|
Cc: linux-rockchip@lists.infradead.org
|
||||||
|
Signed-off-by: Markus Fritsche <mfritsche@reauktion.de>
|
||||||
|
---
|
||||||
|
.../media/platform/verisilicon/hantro_drv.c | 23 +++++++++++++++++++
|
||||||
|
1 file changed, 23 insertions(+)
|
||||||
|
|
||||||
|
diff --git a/drivers/media/platform/verisilicon/hantro_drv.c b/drivers/media/platform/verisilicon/hantro_drv.c
|
||||||
|
index 2e81877f6..6a66c47ed 100644
|
||||||
|
--- a/drivers/media/platform/verisilicon/hantro_drv.c
|
||||||
|
+++ b/drivers/media/platform/verisilicon/hantro_drv.c
|
||||||
|
@@ -186,6 +186,22 @@ static void device_run(void *priv)
|
||||||
|
|
||||||
|
v4l2_m2m_buf_copy_metadata(src, dst);
|
||||||
|
|
||||||
|
+ /*
|
||||||
|
+ * Attach a producer fence on the CAPTURE-side dmabuf so userspace
|
||||||
|
+ * importers (e.g. Wayland compositors) get spec-clean implicit-sync
|
||||||
|
+ * semantics. Called from device_run rather than buf_queue: the
|
||||||
|
+ * dma_fence finite-time contract requires that once a fence is
|
||||||
|
+ * published, the producer must signal it in finite time. By the
|
||||||
|
+ * time we reach device_run, the m2m core has committed to running
|
||||||
|
+ * this job, and the next hop (codec_ops->run) either kicks the HW
|
||||||
|
+ * (decode-complete signals the fence via vb2_buffer_done) or
|
||||||
|
+ * fails immediately (job_finish with VB2_BUF_STATE_ERROR signals
|
||||||
|
+ * the fence with -EIO). Either path resolves the fence in finite
|
||||||
|
+ * time. Best-effort: a NOMEM here means we lose implicit-sync
|
||||||
|
+ * precision for this frame, no functional regression.
|
||||||
|
+ */
|
||||||
|
+ (void)vb2_buffer_attach_release_fence(&dst->vb2_buf);
|
||||||
|
+
|
||||||
|
if (ctx->codec_ops->run(ctx))
|
||||||
|
goto err_cancel_job;
|
||||||
|
|
||||||
|
@@ -249,6 +265,13 @@ queue_init(void *priv, struct vb2_queue *src_vq, struct vb2_queue *dst_vq)
|
||||||
|
dst_vq->lock = &ctx->dev->vpu_mutex;
|
||||||
|
dst_vq->dev = ctx->dev->v4l2_dev.dev;
|
||||||
|
|
||||||
|
+ /*
|
||||||
|
+ * Opt the CAPTURE queue into vb2 release-fence publishing.
|
||||||
|
+ * No-op unless CONFIG_VIDEOBUF2_RELEASE_FENCES=y; runtime cost
|
||||||
|
+ * is one extra fence allocation + dma_resv update per device_run.
|
||||||
|
+ */
|
||||||
|
+ dst_vq->supports_release_fences = true;
|
||||||
|
+
|
||||||
|
return vb2_queue_init(dst_vq);
|
||||||
|
}
|
||||||
|
|
||||||
|
--
|
||||||
|
2.53.0
|
||||||
|
|
||||||
+117
@@ -0,0 +1,117 @@
|
|||||||
|
From 2c63a63bf65739763051dc4ce7ce2ffaf2d514c4 Mon Sep 17 00:00:00 2001
|
||||||
|
In-Reply-To: <20260429195306.239666-1-mfritsche@reauktion.de>
|
||||||
|
References: <20260429195306.239666-1-mfritsche@reauktion.de>
|
||||||
|
From: Markus Fritsche <mfritsche@reauktion.de>
|
||||||
|
Date: Sat, 9 May 2026 16:50:51 +0200
|
||||||
|
Subject: [PATCH RFC v2] media: rockchip-rga: attach dma_resv release fence at
|
||||||
|
device_run
|
||||||
|
MIME-Version: 1.0
|
||||||
|
Content-Type: text/plain; charset=UTF-8
|
||||||
|
Content-Transfer-Encoding: 8bit
|
||||||
|
|
||||||
|
Opt the rockchip-rga driver into the new vb2 release-fence helper.
|
||||||
|
|
||||||
|
Same shape as the hantro patch: attach a producer fence on the
|
||||||
|
CAPTURE-side dmabuf at m2m device_run, signalled by
|
||||||
|
vb2_buffer_done() when RGA completes the m2m operation.
|
||||||
|
|
||||||
|
Differs from hantro in one mechanical detail: rga's device_run
|
||||||
|
wraps the entire body in spin_lock_irqsave(&rga->ctrl_lock). Our
|
||||||
|
helper calls dma_resv_lock(), which is sleepable, so the
|
||||||
|
buffer-fetch + fence-attach sequence has to run above the spinlock.
|
||||||
|
Restructure device_run so:
|
||||||
|
|
||||||
|
- v4l2_m2m_next_src_buf / next_dst_buf,
|
||||||
|
- src->sequence increment,
|
||||||
|
- vb2_buffer_attach_release_fence(&dst->vb2_buf)
|
||||||
|
|
||||||
|
run before spin_lock_irqsave; only the rga->curr assignment and
|
||||||
|
rga_hw_start() (the actual HW kick) remain inside the spinlock.
|
||||||
|
|
||||||
|
This is safe under the m2m-job ownership model: by the time
|
||||||
|
device_run is called, the m2m core has selected this context and
|
||||||
|
serializes one device_run per context, so v4l2_m2m_next_*_buf
|
||||||
|
returns stable pointers until the corresponding *_buf_remove in
|
||||||
|
rga_isr. ctrl_lock was previously protecting per-device state
|
||||||
|
(rga->curr) and the HW register access, neither of which depends on
|
||||||
|
the buffer-fetch happening inside the lock.
|
||||||
|
|
||||||
|
The CAPTURE queue is opted in with supports_release_fences=true at
|
||||||
|
queue_init.
|
||||||
|
|
||||||
|
Userspace consumers of RGA-produced dmabufs (image-processing
|
||||||
|
pipelines, screen-rotation servers, gstreamer flows on Rockchip
|
||||||
|
boards) get spec-clean implicit-sync semantics, matching what
|
||||||
|
hantro does in the previous patch in this series.
|
||||||
|
|
||||||
|
Sven Püschel's ongoing "media: platform: rga: Add RGA3 support"
|
||||||
|
v5 series (linux-rockchip 2026-04-28) restructures rga.c
|
||||||
|
substantially. If that lands first, the device_run restructure
|
||||||
|
here will need a rebase against the new shape; the locking story
|
||||||
|
itself is invariant.
|
||||||
|
|
||||||
|
Cc: Jacob Chen <jacob-chen@iotwrt.com>
|
||||||
|
Cc: Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>
|
||||||
|
Cc: Sven Püschel <s.pueschel@pengutronix.de>
|
||||||
|
Cc: Heiko Stuebner <heiko@sntech.de>
|
||||||
|
Cc: Hans Verkuil <hverkuil@xs4all.nl>
|
||||||
|
Cc: linux-media@vger.kernel.org
|
||||||
|
Cc: linux-rockchip@lists.infradead.org
|
||||||
|
Signed-off-by: Markus Fritsche <mfritsche@reauktion.de>
|
||||||
|
---
|
||||||
|
drivers/media/platform/rockchip/rga/rga.c | 27 +++++++++++++++++++----
|
||||||
|
1 file changed, 23 insertions(+), 4 deletions(-)
|
||||||
|
|
||||||
|
diff --git a/drivers/media/platform/rockchip/rga/rga.c b/drivers/media/platform/rockchip/rga/rga.c
|
||||||
|
index fea63b94c..03030c7ea 100644
|
||||||
|
--- a/drivers/media/platform/rockchip/rga/rga.c
|
||||||
|
+++ b/drivers/media/platform/rockchip/rga/rga.c
|
||||||
|
@@ -38,15 +38,28 @@ static void device_run(void *prv)
|
||||||
|
struct vb2_v4l2_buffer *src, *dst;
|
||||||
|
unsigned long flags;
|
||||||
|
|
||||||
|
- spin_lock_irqsave(&rga->ctrl_lock, flags);
|
||||||
|
-
|
||||||
|
- rga->curr = ctx;
|
||||||
|
-
|
||||||
|
+ /*
|
||||||
|
+ * Fetch the next-job buffers and (best-effort) attach a producer
|
||||||
|
+ * fence on CAPTURE before taking ctrl_lock below.
|
||||||
|
+ * vb2_buffer_attach_release_fence() takes dma_resv_lock, which is
|
||||||
|
+ * sleepable; ctrl_lock is taken with spin_lock_irqsave so any
|
||||||
|
+ * sleepable call must happen above it. Buffer ownership is
|
||||||
|
+ * already committed at this point: the m2m core has selected
|
||||||
|
+ * this context for device_run and serializes one device_run per
|
||||||
|
+ * context, so v4l2_m2m_next_*_buf returns stable pointers until
|
||||||
|
+ * the corresponding *_buf_remove in rga_isr.
|
||||||
|
+ */
|
||||||
|
src = v4l2_m2m_next_src_buf(ctx->fh.m2m_ctx);
|
||||||
|
src->sequence = ctx->osequence++;
|
||||||
|
|
||||||
|
dst = v4l2_m2m_next_dst_buf(ctx->fh.m2m_ctx);
|
||||||
|
|
||||||
|
+ (void)vb2_buffer_attach_release_fence(&dst->vb2_buf);
|
||||||
|
+
|
||||||
|
+ spin_lock_irqsave(&rga->ctrl_lock, flags);
|
||||||
|
+
|
||||||
|
+ rga->curr = ctx;
|
||||||
|
+
|
||||||
|
rga_hw_start(rga, vb_to_rga(src), vb_to_rga(dst));
|
||||||
|
|
||||||
|
spin_unlock_irqrestore(&rga->ctrl_lock, flags);
|
||||||
|
@@ -123,6 +136,12 @@ queue_init(void *priv, struct vb2_queue *src_vq, struct vb2_queue *dst_vq)
|
||||||
|
dst_vq->lock = &ctx->rga->mutex;
|
||||||
|
dst_vq->dev = ctx->rga->v4l2_dev.dev;
|
||||||
|
|
||||||
|
+ /*
|
||||||
|
+ * Opt the CAPTURE queue into vb2 release-fence publishing.
|
||||||
|
+ * Compile-time gated by CONFIG_VIDEOBUF2_RELEASE_FENCES.
|
||||||
|
+ */
|
||||||
|
+ dst_vq->supports_release_fences = true;
|
||||||
|
+
|
||||||
|
return vb2_queue_init(dst_vq);
|
||||||
|
}
|
||||||
|
|
||||||
|
--
|
||||||
|
2.53.0
|
||||||
|
|
||||||
@@ -0,0 +1,61 @@
|
|||||||
|
# vb2_dma_resv release-fence — RFC v2 series
|
||||||
|
|
||||||
|
Three-patch series that opts V4L2 m2m drivers into attaching real
|
||||||
|
producer fences to CAPTURE-side dmabufs, so implicit-sync GPU
|
||||||
|
consumers (Wayland / panfrost / panthor) wait correctly on the
|
||||||
|
producer's decode-completion rather than seeing the dma_buf core's
|
||||||
|
stub fence.
|
||||||
|
|
||||||
|
| Patch | Subject |
|
||||||
|
|---|---|
|
||||||
|
| `0004` | media: videobuf2: add opt-in dma_resv producer fence helper |
|
||||||
|
| `0005` | media: hantro: attach dma_resv release fence at device_run |
|
||||||
|
| `0006` | media: rockchip-rga: attach dma_resv release fence at device_run |
|
||||||
|
|
||||||
|
Numbered with the leading 4/5/6 because the fresnel build series carries
|
||||||
|
these alongside the 3 board-scoped `0001/0002/0003` Pinebook Pro DTS
|
||||||
|
patches; the numbers reflect apply-order in the PKGBUILD, not the
|
||||||
|
upstream lore series ordering (which starts at 1/4..4/4 with a cover).
|
||||||
|
|
||||||
|
## Status
|
||||||
|
|
||||||
|
**RFC v2.** Iterated on `lore.kernel.org/linux-media` after v1 was
|
||||||
|
rejected over the dma_fence finite-time contract gap and bus-locked
|
||||||
|
allocation issues. v2 attaches the fence at `device_run` instead of
|
||||||
|
QBUF, which puts allocation in slept-OK context (PM and clocks up,
|
||||||
|
job committed) — per Nicolas Dufresne's v1 review feedback.
|
||||||
|
|
||||||
|
Cover-letter reference: `marfrit/dmabuf-modifier-triage#3` (campaign
|
||||||
|
session that owns the upstream-targeting work; this directory ships
|
||||||
|
the build-tree-ready form for kernel-agent fleet consumption).
|
||||||
|
|
||||||
|
## Scope
|
||||||
|
|
||||||
|
`subsystem/media/videobuf2/` for the helper (0004), with two
|
||||||
|
driver opt-ins (0005/0006) shipped together because hantro and
|
||||||
|
rockchip-rga both need the helper to be useful on the RK35xx fleet.
|
||||||
|
Splitting 0005/0006 into `driver/hantro/` and `driver/rockchip-rga/`
|
||||||
|
was considered but rejected: they're a single contract series, and
|
||||||
|
their apply-order matters (0004 must precede). Series-as-unit beats
|
||||||
|
per-driver promote eligibility here.
|
||||||
|
|
||||||
|
## Fleet eligibility
|
||||||
|
|
||||||
|
- **fresnel** (RK3399 + hantro + Mali-G52): eligible. Carried in
|
||||||
|
`fleet/fresnel.yaml` since 2026-05-15 (decision flipped from "defer
|
||||||
|
to v2" to "include — v2 in this tree").
|
||||||
|
- **ohm** (RK3566 + hantro + Mali-G52): eligible. Was the original
|
||||||
|
reproducer for the green-frames symptom. Will be carried in
|
||||||
|
`fleet/ohm.yaml` once that manifest lands.
|
||||||
|
- **ampere** (RK3588 + hantro for some codecs): eligibility deferred —
|
||||||
|
RK3588 uses rkvdec2 for primary decode, hantro role is narrower.
|
||||||
|
Re-assess when `fleet/ampere.yaml` lands per issue #6.
|
||||||
|
- **boltzmann** (RK3588): same as ampere — defer.
|
||||||
|
|
||||||
|
## Upstream targeting
|
||||||
|
|
||||||
|
Not yet posted to linux-media in v2 form. Per `feedback_no_upstream.md`
|
||||||
|
the default is "build-tree only, wait for explicit ask". When/if the
|
||||||
|
upstream submission happens, this directory's `0004/0005/0006` are the
|
||||||
|
canonical source — they include the v2 commit headers (`PATCH RFC v2`,
|
||||||
|
`In-Reply-To` chain to the v1 cover-letter Message-Id).
|
||||||
Reference in New Issue
Block a user