Files
bfe-core1106-sdk/.github/workflows/ci.yml
T
BFE EngineeringandClaude Fable 5 973a414f07 review: iteration-2 fixes (fragment assertion, evidence paths, hardening)
- build-kernel.sh fragment assertion: survives a missing trailing newline
  (the read footgun, reproduced) and treats an absent symbol on a disable
  line as FATAL, symmetric with the enable arm.
- fetch-kernel-tarball.sh checks the pin before downloading; both fetchers
  add --retry-connrefused.
- mkimage rejects '.'/'..' state keys.
- ui-shot: VM liveness checked before every QMP call, console.log preserved
  as evidence on every failure path, repaint deadline widened to 90s with
  the contended-runner rationale documented.
- rs485-bridge: overflow discards back off one gap and rate-limit their log
  line, mirroring the accept-loop fix; clippy nit fixed.
- .gitignore ignores *.elf/*.map so the untracked artifacts cannot silently
  return; CI shellcheck glob now covers build/ and the rootfs boot scripts
  (directives added for the deliberate in-guest source paths).
- Docs: NPU parity row matches its sibling verification docs; line-pinned
  audit cross-references unpinned; CROSS_COMPILE documented in the build
  header; payload README lists warden-ui; ci-cd tense settled.

Verified: guards negative-tested (bad state keys, no-newline fragment);
boot smoke, portal scenario, ui-shot all PASS; 53 tests green; shellcheck
clean across the widened glob; clippy zero.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018HUayid7W5w7jBdb9Rrj1K
2026-08-30 08:34:47 -06:00

286 lines
12 KiB
YAML

# warden-sdk CI.
#
# Policy (mirrors flare-edge): only GitHub-owned actions get the repo token; the one
# third-party helper (taiki-e/install-action) is pinned and never handed a token.
# Every job runs on GitHub-hosted runners — no self-hosted runner may be reachable
# from this repo's workflows (ADR-0007: public-repo fork PRs would otherwise be
# able to run code on private infrastructure). kernel-build is dispatch-only.
name: ci
on:
push:
paths-ignore: ['.github/badges/**']
pull_request:
workflow_dispatch:
concurrency:
group: ci-${{ github.ref }}
cancel-in-progress: true
# Least privilege by default: every job gets a read-only token; only `badges`
# (which commits rendered SVGs) overrides this with contents: write below.
permissions:
contents: read
jobs:
test:
runs-on: ubuntu-latest
timeout-minutes: 25
outputs:
passed: ${{ steps.result.outputs.passed }}
coverage: ${{ steps.result.outputs.coverage }}
steps:
- uses: actions/checkout@v4
- name: cargo test (all crates)
run: |
set -o pipefail
: > /tmp/test.log
for d in sim tools/config-lint qemu/rs485-bridge; do
echo "== cargo test in $d ==" | tee -a /tmp/test.log
( cd "$d" && cargo test --locked ) 2>&1 | tee -a /tmp/test.log
done
- name: coverage (cargo-llvm-cov on sim)
run: rustup component add llvm-tools-preview
- uses: taiki-e/install-action@b6ff580856c41316412a0b9b60540fbc6f8c82cc # v2.86.7
with:
tool: cargo-llvm-cov
- name: run coverage
working-directory: sim
run: cargo llvm-cov --locked --json --summary-only --output-path /tmp/cov.json
- name: parse results
id: result
run: |
passed=$(grep -oE '[0-9]+ passed' /tmp/test.log | awk '{s+=$1} END{print s+0}')
pct=$(python3 -c 'import json;print("%.0f"%json.load(open("/tmp/cov.json"))["data"][0]["totals"]["lines"]["percent"])')
echo "passed=$passed" >> "$GITHUB_OUTPUT"
echo "coverage=$pct" >> "$GITHUB_OUTPUT"
echo "tests passed: $passed | sim line coverage: ${pct}%"
mcdc:
# 100% MC/DC (condition coverage) enforced on every Tier-1 driver harness.
runs-on: ubuntu-latest
timeout-minutes: 15
steps:
- uses: actions/checkout@v4
- name: install gcc-14
run: sudo apt-get update -qq && sudo apt-get install -y -qq gcc-14
- name: enforce 100% MC/DC on drivers/*/test
run: |
fail=0; ran=0
for t in drivers/*/test; do
[ -f "$t/Makefile" ] || continue
ran=1
echo "== MC/DC: $t =="
make -C "$t" check CC=gcc-14 GCOV=gcov-14 || fail=1
done
[ "$ran" = 1 ] || { echo "no driver MC/DC harnesses found"; exit 1; }
exit $fail
bench:
# Smoke-run the sim micro-benchmarks and emit the ns/op trend JSON. Regression
# gating against stored history is future work (no flare-edge pattern to copy).
runs-on: ubuntu-latest
timeout-minutes: 20
steps:
- uses: actions/checkout@v4
- name: run sim benchmarks
working-directory: sim
run: |
cargo bench --locked --bench sim_bench 1> bench.txt 2> bench.json
echo "== timings =="; cat bench.txt
echo "== trend json =="; grep '"bench"' bench.json
- name: run rs485-bridge benchmarks
working-directory: qemu/rs485-bridge
run: |
cargo bench --locked --bench bridge_bench 1> bench.txt 2> bench.json
echo "== timings =="; cat bench.txt
echo "== trend json =="; grep '"bench"' bench.json
qemu-tools:
# The qemu/ device-sim build tooling must stay healthy on a plain hosted
# runner: shellcheck the scripts, build the initramfs (pinned busybox,
# fail-closed sha), and build the A/B disk image (unprivileged mkfs -d).
# Booting needs a zImage and therefore lives in kernel-build's smoke step.
runs-on: ubuntu-latest
timeout-minutes: 20
steps:
- uses: actions/checkout@v4
- name: shellcheck qemu scripts
run: |
sudo apt-get update -qq && sudo apt-get install -y -qq shellcheck
shellcheck -x qemu/*.sh qemu/tests/*.sh build/*.sh \
qemu/rootfs/etc/warden-lib.sh qemu/rootfs/etc/rc \
qemu/rootfs/sbin/init qemu/rootfs/init
- name: cache pinned busybox
uses: actions/cache@v4
with:
path: qemu/out/busybox-armv7l
key: busybox-armv7l-${{ hashFiles('qemu/busybox.sha256') }}
- name: build initramfs
run: bash qemu/mkinitramfs.sh
- name: build A/B disk image
run: bash qemu/mkimage.sh
patches-apply:
# The RV1106 series must apply cleanly onto pristine linux-6.18.46.
runs-on: ubuntu-latest
timeout-minutes: 20
steps:
- uses: actions/checkout@v4
- name: cache pristine kernel tarball
uses: actions/cache@v4
with:
path: ~/linux-6.18.46.tar.xz
key: linux-6.18.46-tarball
- name: fetch + verify pristine (shared fail-closed fetcher)
run: bash build/fetch-kernel-tarball.sh "$HOME/linux-6.18.46.tar.xz"
- name: apply the series in order
run: |
tar -C /tmp -xf ~/linux-6.18.46.tar.xz
for p in patches/*.patch; do
git -C /tmp/linux-6.18.46 apply --whitespace=nowarn "$GITHUB_WORKSPACE/$p" \
&& echo "applied $p" || { echo "FAILED to apply $p"; exit 1; }
done
echo "full series applied cleanly onto pristine 6.18.46"
# Automatic artifact GC (mirrors flare-edge). An exceeded account-wide Actions
# storage quota blocks ALL new runs (startup_failure), not just uploads, so
# before a new kernel artifact is uploaded, drop older kernel-rv1106 artifacts
# beyond the newest few — storage stays bounded across dispatches. gh + the
# built-in token (no third-party action); needs actions:write to delete.
# Best-effort: it never fails the run, so it can't block the build that needs it.
prune-artifacts:
if: github.event_name == 'workflow_dispatch'
runs-on: ubuntu-latest
timeout-minutes: 10
permissions:
actions: write
steps:
- name: Keep only the newest few kernel artifacts
env:
GH_TOKEN: ${{ github.token }}
REPO: ${{ github.repository }}
KEEP: "3"
run: |
ids=$(gh api --paginate "repos/$REPO/actions/artifacts?per_page=100" \
--jq "[.artifacts[] | select(.name==\"kernel-rv1106\")]
| sort_by(.created_at) | reverse | .[$KEEP:] | .[].id" \
2>/dev/null) || { echo "list failed; skipping"; exit 0; }
n=0
for id in $ids; do
if gh api -X DELETE "repos/$REPO/actions/artifacts/$id" 2>/dev/null; then
n=$((n + 1)); echo "pruned artifact $id"
fi
done
echo "pruned $n old kernel artifact(s); kept newest $KEEP"
exit 0
kernel-build:
# Full hermetic build on a GitHub-hosted runner (ADR-0007; supersedes the
# self-hosted half of ADR-0004 — a self-hosted runner must never be reachable
# from a public repo's workflows). Manual-dispatch by design — a full kernel
# build is heavy; trigger via `gh workflow run ci.yml` / the Actions UI.
if: github.event_name == 'workflow_dispatch'
needs: [prune-artifacts]
runs-on: ubuntu-latest
timeout-minutes: 60
steps:
- uses: actions/checkout@v4
- name: install cross toolchain + kernel build deps + qemu
run: |
sudo apt-get update -qq
sudo apt-get install -y -qq gcc-arm-linux-gnueabihf qemu-system-arm \
cpio bc bison flex libssl-dev ccache
- name: provision `python` (SDK quirk — build calls bare python)
run: |
mkdir -p "$RUNNER_TEMP/bin"
ln -sf "$(command -v python3)" "$RUNNER_TEMP/bin/python"
echo "$RUNNER_TEMP/bin" >> "$GITHUB_PATH"
- name: cache pristine kernel tarball
uses: actions/cache@v4
with:
path: ~/linux-6.18.46.tar.xz
key: linux-6.18.46-tarball
# Ephemeral runners rebuild the whole tree every dispatch (~9 min of
# compile); ccache recovers most of it for an unchanged/lightly-changed
# series. Keyed on the config + patches so a real change misses cleanly.
- name: cache ccache
uses: actions/cache@v4
with:
path: ~/.ccache
key: kbuild-ccache-${{ hashFiles('build/warden_defconfig', 'patches/*.patch') }}
restore-keys: kbuild-ccache-
- name: build zImage + rv1106-warden.dtb
env:
# WORK must be OUTSIDE the repo checkout: build-kernel.sh applies the patch
# series with `git apply`, which silently ignores out-of-subdir paths when
# run inside another git repo (issue #1). $RUNNER_TEMP is outside the checkout.
WORK: ${{ runner.temp }}/kbuild-out
# The kernel is freestanding; the generic arm cross toolchain links it.
CROSS_COMPILE: arm-linux-gnueabihf-
WARDEN_CCACHE: 1
CCACHE_DIR: /home/runner/.ccache
# KERNEL_TARBALL is exported from the SHELL so $HOME expands — a literal
# `~` in a YAML env: value is never tilde-expanded and broke every
# dispatch until caught in review.
run: |
export KERNEL_TARBALL="$HOME/linux-6.18.46.tar.xz"
bash build/build-kernel.sh
ccache -s | head -4
# Boot smoke under QEMU: the zImage this job just built must reach the
# initramfs sentinel on -M virt (verified 2026-08-29: the canonical
# config boots virt as-is). FAIL-CLOSED on a missing qemu-system-arm.
- name: boot smoke (qemu-system-arm -M virt)
run: |
bash qemu/mkinitramfs.sh
bash qemu/tests/boot-smoke.sh \
"$RUNNER_TEMP/kbuild-out/linux-6.18.46/arch/arm/boot/zImage"
# Best-effort: the build IS the gate. Uploading the zImage/dtb to GitHub
# artifact storage can fail on an account-wide storage-quota hit (recalculated
# every 6-12h) that has nothing to do with this build — don't red-X a good
# kernel build over it.
- uses: actions/upload-artifact@v4
continue-on-error: true
with:
name: kernel-rv1106
path: |
${{ runner.temp }}/kbuild-out/linux-6.18.46/arch/arm/boot/zImage
${{ runner.temp }}/kbuild-out/linux-6.18.46/arch/arm/boot/dts/rockchip/rv1106-warden.dtb
# Short so kernel images self-expire instead of piling into the
# account-wide storage quota; the prune-artifacts job above is the
# active bound, this is the backstop.
retention-days: 5
badges:
needs: [test]
if: github.event_name == 'push' && github.ref == 'refs/heads/main'
runs-on: ubuntu-latest
timeout-minutes: 15
permissions:
contents: write
steps:
- uses: actions/checkout@v4
- name: install cloc
run: sudo apt-get update -qq && sudo apt-get install -y -qq cloc
- name: render badges
env:
PASSED: ${{ needs.test.outputs.passed }}
COVERAGE: ${{ needs.test.outputs.coverage }}
run: |
mkdir -p .github/badges
loc=$(cloc --quiet --json --exclude-dir=target,build,build-target,patches,data,docs . \
| python3 -c 'import sys,json; print(json.load(sys.stdin)["SUM"]["code"])')
col=orange; [ "${COVERAGE:-0}" -ge 60 ] && col=yellow; [ "${COVERAGE:-0}" -ge 80 ] && col=brightgreen
curl -fsSL "https://img.shields.io/badge/lines%20of%20code-${loc}-blue" -o .github/badges/loc.svg
curl -fsSL "https://img.shields.io/badge/tests-${PASSED}%20passing-brightgreen" -o .github/badges/tests.svg
curl -fsSL "https://img.shields.io/badge/coverage-${COVERAGE}%25-${col}" -o .github/badges/coverage.svg
- name: commit badges
run: |
git config user.name "github-actions[bot]"
git config user.email "github-actions[bot]@users.noreply.github.com"
git add .github/badges/loc.svg .github/badges/tests.svg .github/badges/coverage.svg
if ! git diff --cached --quiet; then
git commit -m "ci: update loc/tests/coverage badges [skip ci]"
git push
fi