Files
kernel-agent/fleet/ohm.yaml
T
test0r 3d15c5367d fleet/ohm: pkgrel=6 — per-series converged with tx-sdio-dma-oob + join-confirm-reset
Two additions to fleet/ohm.yaml's includes for the bes2600 driver scope:

1. driver/bes2600/tx-sdio-dma-oob-danctnix/ — already on disk from
   ka#17 but not previously included. The cumulative-c5x-danctnix
   shipped in pkgrel=3 did NOT have this fix; pkgrel=4 per-series
   regressed because the staging-prep series was excluded. KFENCE
   caught the OOB during pkgrel=4 soak; pkgrel=5 included it.

2. driver/bes2600/join-confirm-reset-danctnix/ — NEW scope.
   cw1200 ancestor port (sta.c:1339-1344) with bes2600-specific
   PASSIVE-gate compensation in bes2600_unjoin_work. Closes
   besser#25. Verified pkgrel=6 srcversion 0E16463F: cascade gone,
   periodic ~600ms latency jitter also gone (same root cause).

Status note: per-series reconstruction is now converged. The
cumulative-c5x-danctnix entry is left as historical fallback;
ka#29's blocker (per-series mirrors not applying cleanly) was
resolved by manually reconstructing the per-series in
marfrit/bes2600-dkms bes2600/join-confirm-failure-reset (top
commit 3d833f8).

Build still hand-managed via boltzmann:~/src/besser/marfrit-besser/
danctnix-besser-pkgbuild/kernel/PKGBUILD; ka-promote / ka-build
template rendering still pending per the original TODOs.

Signed-off-by: Claude (noether) <claude@reauktion.de>
2026-05-21 12:23:47 +02:00

141 lines
7.0 KiB
YAML
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# kernel-agent manifest for ohm (PineTab2 / Rockchip RK3566 + BES2600 SDIO WiFi/BT)
#
# Status: scaffolding from 2026-05-16; per-series patchset converged 2026-05-21 (pkgrel=6). Patches/scopes are mirrored;
# the build pipeline (cumulative-patch generation, makepkg invocation,
# sign+publish) still relies on the hand-managed flow in
# boltzmann:~/src/besser/marfrit-besser/danctnix-besser-pkgbuild/kernel/.
# Once ka-promote / ka-build land, switch this manifest to authoritative
# and retire the boltzmann-side checkouts (canonical + the orphan at
# ~/src/besser/danctnix-besser-pkgbuild/ that the fourier-campaign
# sibling agent accidentally created).
#
# See kernel-agent issue #5 for the full migration plan.
host: ohm
arch: arm64
soc: rockchip/rk3566
board: pinetab2
distro: archlinux-arm # DanctNIX PineTab2 variant on Arch Linux ARM aarch64
baseline:
tree: DanctNIX/linux-pinetab2
url: https://codeberg.org/DanctNIX/linux-pinetab2
ref: v7.0-danctnix1 # _srcname=linux-pinetab2 _srctag=v${pkgver%.*}-${pkgver##*.}
upstream_compat: linux-7.0 # DanctNIX rebases off mainline + per-tablet delta
# Scope-tagged patch includes. Resolves to patches/<scope>/<file>.patch.
#
# 2026-05-18 audit: the per-series -danctnix mirrors in
# patches/driver/bes2600/*-danctnix/ created by kernel-agent#17 use
# DKMS-style root paths (bes2600/foo.c) rather than in-tree staging
# paths (drivers/staging/bes2600/foo.c), and at least one has corrupted
# mixed-prefix headers (a/drivers/staging/bes2600/... b/bes2600/...).
# They do NOT apply cleanly against the linux-pinetab2 baseline.
#
# 2026-05-21 update: per-series reconstruction (besser#22) completed
# 2026-05-21; pkgrel=6 (srcversion 0E16463F) on ohm soak-passed with
# the bounce-buffer + join-confirm-reset additions. The per-series
# manifest below is the authoritative set; cumulative-c5x-danctnix
# remains as historical fallback only.
#
# Until the per-series mirrors are reconstructed (kernel-agent followup
# issue), the bes2600 driver scope is satisfied by a single-file
# cumulative captured from the working hand-managed
# danctnix-besser-pkgbuild flow on boltzmann (see
# patches/driver/bes2600/cumulative-c5x-danctnix/README.md). This is
# the c5x stack as it shipped in pkgrel=3 on 2026-05-18.
includes:
# bes2600 driver (c5x stack as shipped in pkgrel=3) — single-file
# interim cumulative; per-series reconstruction tracked separately.
- driver/bes2600/cumulative-c5x-danctnix/
# close besser#1 — refuse multi-channel 5 GHz scans at driver boundary.
- driver/bes2600/scan-filter-5ghz-danctnix/
# GCC 15.2.1 build-fix for arm_neon.h + SHADOW_CALL_STACK interaction.
# Runtime no-op as long as the config has CONFIG_SHADOW_CALL_STACK=n
# (current ohm setting). Kept in the manifest for the day SCS gets
# re-enabled. See reference_arm64_scs_arm_neon_gcc15 memory.
- arch/arm64/scs-arm-neon-build-fix/
# close besser#18 — pending_record_lock SOFTIRQ-safe -> -unsafe inversion.
# Mirror of marfrit/bes2600-dkms#11 (d95453c). 5-site spin_lock -> _bh.
- driver/bes2600/queue-pending-record-lock-bh-danctnix/
# bounce-buffer fix for SDIO TX DMA OOB (KFENCE-detected on pkgrel=4 soak);
# the per-series mirror of marfrit/bes2600-dkms bes2600/tx-sdio-dma-oob.
# cumulative-c5x-danctnix did NOT include this — it was the regression
# surfaced during the per-series reconstruction.
- driver/bes2600/tx-sdio-dma-oob-danctnix/
# close besser#25 — wsm_reset + serialised unjoin on JOIN reject.
# cw1200 ancestor port (drivers/net/wireless/st/cw1200/sta.c:1339-1344)
# with bes2600-specific PASSIVE-gate compensation. pkgrel=6 verified.
- driver/bes2600/join-confirm-reset-danctnix/
# Explicitly NOT included (decision logged):
# - debian-copyright-fsf-address: Debian packaging metadata, not kernel
# - bare (non-danctnix) variants of the per-series mirrors: same
# root-path bug as the -danctnix variants per the 2026-05-18 audit
config:
source: hand-managed config file in boltzmann:~/src/besser/marfrit-besser/danctnix-besser-pkgbuild/kernel/config
strategy: snapshot, fold to baseline, accept-new with rationale on diff
TODO: migrate config into kernel-agent flow once kconfig-by-manifest lands
package:
name: linux-pinetab2-danctnix-besser
versioning: "${baseline_ref}.kafr${pkgrel}" # e.g. v7.0-danctnix1.kafr1
install_mode: alongside # coexists with linux-pinetab2 if user wants A/B
conflicts: []
provides: [linux-pinetab2] # drop-in replacement; pacman can satisfy linux-pinetab2 deps
kernel_suffix: -danctnix-besser # vmlinuz-linux-pinetab2-danctnix-besser
bootloader: extlinux # PineTab2 standard
# PKGBUILD currently at:
# boltzmann:~/src/besser/marfrit-besser/danctnix-besser-pkgbuild/kernel/PKGBUILD
# TARGET location (per issue #5):
# marfrit/marfrit-packages/arch/linux-pinetab2-danctnix-besser/PKGBUILD
# PENDING the kernel-agent template-rendering pipeline (b2sums regen,
# pkgrel bump, cumulative-patch generation from manifest scope).
template_at: TBD-marfrit-packages-arch-linux-pinetab2-danctnix-besser
replaces_dkms: bes2600-dkms # once cumulative series in this manifest is enough
verify:
bar1_ssh_heartbeat: required
bar2_pkg_version: required
bar3_module_loaded:
- bes2600 module loads
- wlan0 + bt0 (BT/UART) present after boot
- sdio_force_uhs=0 not needed (DMA-OOB-read fix in tx-sdio-dma-oob series)
bar4_per_patch_probe: opt-in
bar5_burn_in:
mode: opt-in
tests:
- "WiFi: 24h iperf3 to LAN host without rxhang"
- "PM: lid-close → wake cycles × 100 without bes2600 confirm-loss"
build_host:
primary: boltzmann # native aarch64 with ohm's identical .config
fallback: fermi # hertz LXD, ALARM aarch64
# No distcc per feedback_kernel_agent_no_distcc.md.
backup:
pre_install: hertz:/sparfuxdata/kernel-agent-backups/ohm/${replaced_version}/
# ─────────────────────────────────────────────────────────────────────
# OPEN — kernel-agent CLI work blocking full migration:
#
# 1. ka-promote: read includes[] above + apply_order field (to be added),
# concatenate referenced series into a single .patch in the build dir.
# Validate it applies cleanly on baseline.ref. Compute b2sum for
# template substitution.
#
# 2. PKGBUILD template renderer: stamp pkgrel, ${_cumulative_b2sum},
# ${_srctag} from manifest into a PKGBUILD draft at the
# template_at location. Sign + publish to packages.reauktion.de.
#
# 3. Orphan retirement: surface ~/src/besser/danctnix-besser-pkgbuild/
# (NO remote, ~/src/besser/marfrit-besser/... is canonical) to
# Markus / fourier campaign for working-state migration BEFORE
# deleting. See besser issue #17 for the regression that the
# orphan caused. After the kernel-agent flow is authoritative,
# delete both checkouts.
# ─────────────────────────────────────────────────────────────────────