Claude review: drm/ci: uprev mesa

[Claude Code Review Bot <claude-review@example.com>]

Patch Review

**Commit:** `daab4b8a1a5dc`

This is the largest patch in the series. It updates the DRM CI to track recent Mesa CI infrastructure changes. Key changes:

1. **Mesa CI commit bump:** `DRM_CI_COMMIT_SHA` updated from `02337aec...` to `25881c70...`, and `MESA_TEMPLATES_COMMIT` updated similarly.

2. **Rule refactoring:** `.container+build-rules` is split into separate `.container-rules` and `.build-rules`. Variable renames: `is-pre-merge` → `is-merge-request`, `is-fork-push` → `is-push-to-fork`.

3. **Build script fixes:**
   ```bash
   -git config --global pull.rebase true
   ...
   -rm -rf .git/rebase-merge
   +rm -rf .git/rebase-merge .git/rebase-apply
   ...
   -    git pull ${UPSTREAM_REPO} ${TARGET_BRANCH}-external-fixes
   +    git pull --no-rebase ${UPSTREAM_REPO} ${TARGET_BRANCH}-external-fixes
   ```
   The `--no-rebase` flag is added to all `git pull` commands, and the duplicate `rm -rf .git/rebase-apply` at the top is consolidated. This fixes issues with rebase operations during CI builds.

4. **LAVA submission overhaul** (`lava-submit.sh`): The old `get_path_to_artifact()` / `_check_artifact_path()` functions are replaced with `fdo_find_s3_path`. Section logging switches from `section_start`/`section_switch` to `fdo_log_section_start_collapsed`/`fdo_log_section_end`. The overlay handling now uses `LAVA_EXTRA_OVERLAYS` array for firmware overlays. The job submission switches from `PYTHONPATH=artifacts/ artifacts/lava/lava_job_submitter.py` to the containerized `lava-job-submitter`.

5. **Container/image changes:** Package names updated (e.g., `libasound2` → `libasound2t64`, `libdw1` → `libdw1t64`) reflecting Debian transitions. New `FIRMWARE_TAG`/`FIRMWARE_REPO` variables added. `CROSVM_TAG` added. `ALPINE_X86_64_LAVA_SSH_TAG` removed. `KERNEL_TAG` bumped from `v6.14` to `v6.16`.

6. **Job removals:** `python-artifacts` job removed. `.baremetal-igt-arm64` template removed. `msm:apq8016` and `msm:apq8096` jobs are dot-prefixed (`.msm:apq8016`, `.msm:apq8096`) to disable them pending re-enablement in patch 4.

7. **Shell script change:** `igt_runner.sh` shebang changed from `#!/bin/sh` to `#!/usr/bin/env bash` and sources `setup-test-env.sh`.

8. **Various new `when: never` rules** for Mesa CI jobs not used by DRM CI (e.g., `debian-x86_64-msan`, `debian-riscv64`, `.ci-tron-*`).

9. **Expectation file updates:** Extensive updates to xfails/flakes files for multiple platforms, reflecting IGT version 2.1 and kernel 6.16.0-rc2 results.

**Analysis of the lava-submit.sh changes:** The firmware overlay logic is worth examining:

```bash
if [ -n "${LAVA_FIRMWARE:-}" ]; then
    for fw in $LAVA_FIRMWARE; do
        LAVA_EXTRA_OVERLAYS+=(
            - append-overlay
              --name=linux-firmware
              --url="https://${S3_BASE_PATH}/${FIRMWARE_REPO}/${fw}-${FIRMWARE_TAG}.tar"
              --path="/"
              --format=tar
        )
    done
fi
```

The `--name=linux-firmware` is used for every firmware overlay iteration. If multiple firmware entries are specified in `LAVA_FIRMWARE`, they'd all get `--name=linux-firmware`. This appears to be how the lava-job-submitter expects it (appending overlays with the same logical name), so this is likely fine.

The kernel overlay uses a backslash continuation which is inside an array assignment — this is valid bash:
```bash
LAVA_EXTRA_OVERLAYS+=(
    - append-overlay \
      --name=kernel-build \
      ...
)
```

**Observations:**
- The patch is large but coherent — it adapts to upstream Mesa CI infrastructure changes.
- The `check-patch` job gets an explicit `stage: static-checks` added, which is needed because `.build` no longer inherits the correct stage from the removed `.container+build-rules`.
- The `mkdir -p /kernel` addition in the `.software-driver` section fixes what was presumably a missing directory for the vkms/software-driver jobs.

**Verdict:** This is a large but well-structured infrastructure update. The commit message could be slightly more specific about the individual sub-changes, but the v2 changelog addresses this. No functional regressions possible as this is pure CI infrastructure. Well-reviewed by Daniel Stone and Dmitry Baryshkov.

---

---
Generated by Claude Code Patch Reviewer