main.yml

---
name: CXL test runner

# yamllint disable-line rule:truthy
on:
  workflow_dispatch:
    inputs:
      kernel_repo:
        description: "Kernel repo: owner/name OR full git URL"
        required: true
        default: "AlisonSchofield/linux-kernel"
      kernel_branch:
        description: "Kernel branch (or tag/SHA)"
        required: true
        default: "cxl-testme"

      ndctl_repo:
        description: "ndctl repo: owner/name OR full git URL"
        required: true
        default: "pmem/ndctl"
      ndctl_branch:
        description: "ndctl branch (or tag/SHA)"
        required: true
        default: "pending"

      timeout_min:
        description: "Guest timeout (minutes)"
        required: true
        default: "35"

jobs:
  build:
    runs-on: ubuntu-24.04

    env:
      # Inputs
      KERNEL_REPO: ${{ inputs.kernel_repo }}
      KERNEL_BRANCH: ${{ inputs.kernel_branch }}
      NDCTL_REPO: ${{ inputs.ndctl_repo }}
      NDCTL_BRANCH: ${{ inputs.ndctl_branch }}
      RUN_TIMEOUT_MIN: ${{ inputs.timeout_min }}

      # Pinned runner behavior (not user-configurable ATM)
      RUN_OPTS: --cxl-test-run --ndctl-build

      # Pinned image selection
      MKOSI_DISTRO: ubuntu
      MKOSI_RELEASE: noble

      # Pinned runner identity strings (for output / cache keys)
      RUNNER_OS_LABEL: ubuntu-24.04
      ARCH_LABEL: x86_64

    steps:
      - name: checkout run_qemu (this repo)
        uses: actions/checkout@v4
        with:
          path: run_qemu

      # Freeing disk space takes time. Check if needed.
      - name: check free disk space
        id: disk
        shell: bash
        run: |
          set -euo pipefail
          avail=$(df --output=avail -BG / | tail -1 | tr -dc '0-9')
          echo "available_gb=$avail" >> "$GITHUB_OUTPUT"
          echo "Free space: ${avail}G"
          df -h

      # Free disk space on GitHub runner
      - name: apt-get remove for disk space
        if: ${{ steps.disk.outputs.available_gb < 20 }}
        run: |
          df -h
          sudo apt-get remove -y '^aspnetcore-.*' '^dotnet-.*'
          sudo apt-get remove -y 'php.*' '^mongodb-.*' '^mysql-.*'
          sudo apt-get remove -y azure-cli '^google-cloud*'
          sudo apt-get remove -y '^llvm-.*'
          sudo apt-get autoremove -y
          sudo apt-get install -y bc
          df -h

      - name: rm -rf for disk space
        if: ${{ steps.disk.outputs.available_gb < 20 }}
        run: |
          df -h
          sudo docker image prune --all --force || true
          df -h
          _dirs=( /usr/local/lib/android /usr/share/dotnet /usr/local/.ghcup )
          sudo du -shc "${_dirs[@]}" || true
          sudo rm -rf "${_dirs[@]}"
          df -h

      - name: apt install mkosi and run_qemu requirements
        run: |
          sudo apt-get update

          # Kernel build deps
          sudo apt install -y build-essential flex bison libelf-dev libssl-dev ccache

          # run_qemu deps
          sudo apt install -y mkosi
          sudo apt install -y dracut-core qemu-utils ovmf mtools qemu-system-x86
          pip3 install qemu.qmp

      - name: mkosi v20 fixes on Ubuntu 24.04
        run: |
          cd /usr/lib/python3/dist-packages
          sudo git apply "${{ github.workspace }}/run_qemu/Patches/ubuntu/24.04/mkosi/"*.patch

      - name: apt install Ubuntu 24.04 requirements for mkosi
        run: |
          sudo apt install -y systemd-ukify systemd-boot

      - name: apt install ndctl build dependencies
        run: |
          grep -vE '^[[:blank:]]*#' .github/workflows/ubuntu-ndctl-build-deps | \
          xargs sudo apt install -y

      - name: argbash
        run: |
          AB_VER=2.10.0
          wget https://github.com/matejak/argbash/archive/refs/tags/${AB_VER}.tar.gz
          tar xf ${AB_VER}.tar.gz
          sudo apt install -y autoconf
          sudo make -C argbash-${AB_VER}/resources install PREFIX=/usr/local/

      - name: set week number for caches
        id: weeks
        run: |
          printf 'now=%s\n' "$(date +%Y-w%U)" >> "$GITHUB_OUTPUT"
          printf 'previous=%s\n' "$(date +%Y-w%U -d '7 days ago')" >> "$GITHUB_OUTPUT"

      - name: compute cache key suffixes
        id: cachekeys
        shell: bash
        run: |
          set -euo pipefail
          k="$(printf '%s\n' "$KERNEL_REPO@$KERNEL_BRANCH" | sha256sum | awk '{print $1}')"
          n="$(printf '%s\n' "$NDCTL_REPO@$NDCTL_BRANCH" | sha256sum | awk '{print $1}')"
          echo "kernel_key=$k" >> "$GITHUB_OUTPUT"
          echo "ndctl_key=$n" >> "$GITHUB_OUTPUT"

      - name: Fetch ccache
        uses: actions/cache@v4
        with:
          path: ~/.cache/ccache/
          key: ccache_${{ env.RUNNER_OS_LABEL }}_${{ env.ARCH_LABEL }}_${{ steps.cachekeys.outputs.kernel_key }}_${{ steps.weeks.outputs.now }}
          restore-keys: |
            ccache_${{ env.RUNNER_OS_LABEL }}_${{ env.ARCH_LABEL }}_${{ steps.cachekeys.outputs.kernel_key }}_${{ steps.weeks.outputs.previous }}
            ccache_${{ env.RUNNER_OS_LABEL }}_${{ env.ARCH_LABEL }}_${{ steps.cachekeys.outputs.kernel_key }}
            ccache_${{ env.RUNNER_OS_LABEL }}_${{ env.ARCH_LABEL }}

      - name: ensure clean checkout dirs
        shell: bash
        run: |
          set -euo pipefail
          rm -rf ndctl kernel

      # Checkout that supports either "owner/name" OR full git URL.
      - name: checkout ndctl
        shell: bash
        run: |
          set -euo pipefail
          if [[ "$NDCTL_REPO" =~ ^https?://|^git://|^ssh://|^git@ ]]; then
            git clone --depth 1 --branch "$NDCTL_BRANCH" "$NDCTL_REPO" ndctl
          else
            git clone --depth 1 --branch "$NDCTL_BRANCH" "https://github.com/${NDCTL_REPO}.git" ndctl
          fi

      - name: checkout kernel
        shell: bash
        run: |
          set -euo pipefail
          export GIT_LFS_SKIP_SMUDGE=1
          if [[ "$KERNEL_REPO" =~ ^https?://|^git://|^ssh://|^git@ ]]; then
            git -c submodule.recurse=false clone \
              --depth 1 --single-branch \
              --branch "$KERNEL_BRANCH" \
              "$KERNEL_REPO" kernel
          else
            git -c submodule.recurse=false clone \
              --depth 1 --single-branch \
              --branch "$KERNEL_BRANCH" \
              "https://github.com/${KERNEL_REPO}.git" kernel
          fi

      - name: show revisions under test
        run: |
          echo "Kernel repo:   $KERNEL_REPO"
          echo "Kernel branch: $KERNEL_BRANCH"
          git -C kernel rev-parse HEAD
          git -C kernel log -1 --oneline
          echo
          echo "ndctl repo:    $NDCTL_REPO"
          echo "ndctl branch:  $NDCTL_BRANCH"
          git -C ndctl rev-parse HEAD
          git -C ndctl log -1 --oneline
          echo
          echo "Runner:        $RUNNER_OS_LABEL"
          echo "Arch:          $ARCH_LABEL"
          echo "mkosi image:   $MKOSI_DISTRO $MKOSI_RELEASE"
          echo "run_qemu opts: $RUN_OPTS"
          echo "timeout (min): $RUN_TIMEOUT_MIN"

      - name: defconfig
        working-directory: ${{ github.workspace }}/kernel
        run: |
          make defconfig ARCH=x86_64
          bash ./scripts/kconfig/merge_config.sh .config ../run_qemu/.github/workflows/cxl-test.cfg
          make olddefconfig ARCH=x86_64

      - name: disable AppArmor
        run: |
          if test -e /proc/sys/kernel/apparmor_restrict_unprivileged_unconfined; then
            sudo sysctl -w kernel.apparmor_restrict_unprivileged_unconfined=0
            sudo sysctl -w kernel.apparmor_restrict_unprivileged_userns=0
          fi

      - name: run_qemu.sh
        run: |
          set -x

          mkosi --version
          qemu-system-x86_64 --version || true
          which qemu-system-x86_64 || true

          ccache --show-stats
          cp run_qemu/contrib/tmpfs_workspace.tmpl run_qemu/mkosi_tmpl_portable/

          rm -f /tmp/rq_0.log
          touch /tmp/rq_0.log
          chmod a+rw /tmp/rq_0.log

          cd kernel

          (
            PATH=/usr/lib/ccache:"$PATH" \
            distro=${MKOSI_DISTRO} rev=${MKOSI_RELEASE} \
            ndctl='${{ github.workspace }}/ndctl' \
            ../run_qemu/run_qemu.sh ${RUN_OPTS} \
              --rebuild=kmod \
              --no-kvm \
              --timeout=${RUN_TIMEOUT_MIN} \
              --debug
          ) 2>&1 | tee -a /tmp/rq_0.log

          exit ${PIPESTATUS[0]}

      - name: failure triage snapshot
        if: ${{ always() }}
        shell: bash
        run: |
          set +e
          echo "=== df -h ==="
          df -h
          echo
          echo "=== free -h ==="
          free -h
          echo
          echo "=== dmesg tail ==="
          dmesg | tail -n 200 || true
          echo
          echo "=== last 200 lines of /tmp/rq_0.log ==="
          tail -n 200 /tmp/rq_0.log 2>/dev/null || true

      - name: check if qemu died early
        if: ${{ always() }}
        run: |
          echo "=== qemu processes ==="
          ps -ef | grep -v grep | grep qemu-system-x86_64 || echo "no qemu process"
          echo "=== QMP socket ==="
          ls -la /tmp/run_qemu_qmp_0 || true

      - name: show bios and serial logs
        if: ${{ always() }}
        run: |
          echo "=== /tmp/rq_0.log tail ==="
          tail -n 200 /tmp/rq_0.log || true
          echo "=== bios_debug.log ==="
          tail -n 200 kernel/bios_debug.log 2>/dev/null || tail -n 200 bios_debug.log 2>/dev/null || true

      - name: CXL unit test run results
        if: always()
        run: |
          echo "==== CXL unit test run results ===="
          sed -e 's/^\[[^]]*\] //' -e 's/\r$//' /tmp/rq_0.log \
            | awk '/^======= meson-test.log ========/{p=1} p; /^======= Done \/usr\/local\/bin\/rq_cxl_tests.sh ========/{p=0}'

      - name: upload logs
        if: ${{ always() }}
        uses: actions/upload-artifact@v4
        with:
          name: logs
          if-no-files-found: warn
          path: |
            /tmp/rq_0.log
            kernel/bios_debug.log
            bios_debug.log

      - name: ccache stats
        run: |
          ccache --show-stats
          ccache --show-config | grep dir || true