sunfish-adminandCursor
2d145c2878
Restore disable_configstore, EuiccDisabler and thermal CDEV guards, branding and rebuild/sync helpers; document device roles, laptop migration, and Gitea. Co-authored-by: Cursor <cursoragent@cursor.com>
148 lines
6.9 KiB
Markdown
148 lines
6.9 KiB
Markdown
# 009 — Device polish after LAB daily-driver / GOLDEN rebacon (2026-10-16)
|
||
|
||
Owner test of the current Magisk ship zip. LAB is the daily driver (do not
|
||
wipe/flash without an explicit owner ask). GOLDEN was factory-reset and now
|
||
runs **our** build — it is no longer a live AXP.OS oracle.
|
||
|
||
**ADB / debug:** GOLDEN `11071JEC203839` (connected). Prefer GOLDEN for
|
||
`logcat` / vendor probes. LAB `11231JEC207477` stays main phone — read-only
|
||
unless the owner says otherwise.
|
||
|
||
**AXP parity oracle:** code and dumps on `a16.ssystems.de`
|
||
(`/srv/rom/references/axp/`, EXTENDROM, archived golden captures) — **not**
|
||
the live GOLDEN system anymore.
|
||
|
||
Measured on GOLDEN (our zip): `23.2-20261008-PRO-sunfish`,
|
||
`ro.build.date=Sun Oct 11 08:24:09 UTC 2026`, kernel
|
||
`4.14.357-openela-g28f9290ae067-dirty #81`.
|
||
|
||
## Priority high
|
||
|
||
### 1. configstore-hal / VINTF — **root cause + fix in WT**
|
||
|
||
- Every ~5s: `android.hardware.configstore@1.1::ISurfaceFlingerConfigs/default`
|
||
must be in VINTF manifest → service aborts and init restarts it.
|
||
- Present on current zip: `/vendor/etc/init/android.hardware.configstore@1.1-service.rc`
|
||
(service `vendor.configstore-hal`), binary under `/vendor/bin/hw/…`.
|
||
- **Root cause:** `build/make/target/product/base_vendor.mk` pulls
|
||
`android.hardware.configstore@1.1-service` via
|
||
`PRODUCT_PACKAGES_SHIPPING_API_LEVEL_29`. Lineage
|
||
`device/google/sunfish/device-lineage.mk` normally packages
|
||
`disable_configstore` (Soong `overrides` stub) so the HIDL HAL is **not**
|
||
installed. Wave 4 treated that as a “lab-only command” and removed it →
|
||
empty Configstore stanza → real HAL ships → A16 VINTF crash-loop.
|
||
- **Fix (local WT):** `apply_009_drop_lab_only_commands.py` no longer drops
|
||
`disable_configstore`; it **restores** it if missing. Bacon postflight
|
||
requires HAL+rc absent and `/vendor/bin/disable_configstore` present.
|
||
- Lands on next bacon/OTA; no live flash of LAB.
|
||
|
||
### 2. Thermal — keep both; fix false HAL ERROR (WT)
|
||
|
||
**Why two:** complementary, not duplicates.
|
||
|
||
| Process | Role |
|
||
|---------|------|
|
||
| `android.hardware.thermal-service.pixel` | Android HAL: temps + cooling device *reporting* (`thermal_info_config.json`) |
|
||
| `thermal-engine` | QC closed-loop: CPU/GPU/`fcc`/`modem` actions in `thermal-engine-sunfish.conf` |
|
||
|
||
Stock packaging on a16: `device.mk` packs the pixel HAL; vendor +
|
||
`init.hardware.rc` start `vendor.thermal-engine`.
|
||
|
||
**pixel-thermal ERROR root cause (confirmed):**
|
||
`thermal_info_config_sunfish.json` lists CoolingDevices
|
||
(`thermal-cpufreq-0/7`, `thermal-devfreq-0`, `fcc`) but has **zero**
|
||
`BindedCdev`/`CdevRequest`. Only sunfish among `device/google/*/thermal_info*`
|
||
is unbound like that — by design (engine mitigates). Newer pixel HAL still
|
||
runs `logCoolingDeviceStatus()` ~every few minutes and
|
||
`getCdevMaxRequest()` ERROR-spams for every unbound cdev. Sensors still read
|
||
OK (`skin-therm-monitor`, etc.). Parallel stack works; log was lying.
|
||
|
||
**Fix:** `apply_009_thermal_cdev_log.py` — skip unbound cdevs in the CDEV log
|
||
(VERBOSE note); real binded-path ERROR unchanged.
|
||
|
||
**thermal-engine (Magisk su 2026-10-16 after Magisk install):** both services
|
||
`running`, `vendor.thermal.link_ready=1`. Sysfs has full cooling set
|
||
(`fcc`, `thermal-cpufreq-*`, `thermal-devfreq-0`, `modem_*`, `adsp_vdd`, …).
|
||
Boot `dmesg`: early `Couldn't get temp USB port thermal zone rc=-22` (before
|
||
zones ready), charger WLC cooling missing (wireless), SELinux deny
|
||
`hal_power` → `hal_thermal` find (separate). No live ion/msm_thermal spam in
|
||
current buffers after reboot — treat earlier reports as boot noise unless
|
||
reproduced under load. Keep engine.
|
||
|
||
## Priority medium
|
||
|
||
### 3. OpenEUICC / EuiccDisabler — **fix in WT**
|
||
|
||
- `im.angry.openeuicc` installed (`/system_ext/priv-app/OpenEUICC`).
|
||
- Crash path is **GoogleParts**
|
||
`device/google/sunfish/parts/.../EuiccDisabler.java`
|
||
(`org.lineageos.settings.EuiccDisabler`), not EuiccPolicy Kotlin (that
|
||
already guards). Stack: `Unknown package: com.google.android.euicc`.
|
||
- **Fix:** `apply_009_euicc_disabler_guard.py` — skip missing packages before
|
||
`setApplicationEnabledSetting`. Does not retarget to OpenEUICC / does not
|
||
disable `im.angry.openeuicc`. Wired into ship bacon after wave1 euicc drop.
|
||
|
||
### 4. MediaCodec / media_quality — **closed (expected on phone)**
|
||
|
||
Reproduced 2026-10-16 on GOLDEN (Jelly browser → HW VP9 `c2.qti.vp9.decoder`):
|
||
|
||
- `media_quality service unavailable, skipping updatePictureProfile` is
|
||
**ALOGI**. `MediaQualityService` only starts when
|
||
`mediaQualityFw() && isTv` (`SystemServer.java`). Phones never publish
|
||
`media_quality`; MediaCodec skips picture profiles. Not a VINTF/codec bug.
|
||
- `Failed to query component interface …: 6` = `BAD_INDEX` while probing C2
|
||
allocator params; same session then creates pools **OK** and plays video.
|
||
- **No ROM fix.** Do not ship TV MediaQuality just to silence an INFO line.
|
||
|
||
### 5. Audio SLIM_0_RX Format — **root cause known; no simple ship fix**
|
||
|
||
Reproduced on every speaker open (browser video). Audio still works.
|
||
|
||
- Card id: `sm6150bolerosnd` (Bolero / CDC-DMA, not classic Slimbus).
|
||
- Proprietary `audio.primary.sm6150.so` (vendor blob) defaults
|
||
`is_slimbus_interface=true` and only clears it for an allowlist of
|
||
full card names (`sm6150-idp-snd-card`, `sm6150-qrd-snd-card`, …).
|
||
**`sm6150bolerosnd` is missing** → HAL looks up `SLIM_0_RX Format`, which
|
||
does not exist; Bolero expects `WSA_CDC_DMA_RX_0 Format` (same logic exists
|
||
in CAF `platform.c` behind `!is_slimbus_interface`).
|
||
- Soft-fail: ALOGE then continues (channel map / playback OK).
|
||
- **Fix would mean** rebuilding/replacing `audio.primary.sm6150` from CAF with
|
||
one card-name allowlist line — not a mixer_paths XML tweak. Deferred unless
|
||
we deliberately take on audio HAL from source (risk vs log noise).
|
||
|
||
## Build hygiene
|
||
|
||
- Kernel uname shows `-dirty` on this bacon — rebuild from a clean tree or
|
||
document intentional WT deltas in the artifact notes.
|
||
- Non–Direct Boot apps (e.g. Metrolist) can crash before unlock — app issue,
|
||
not ROM.
|
||
|
||
## Does not need changes (this session)
|
||
|
||
F2FS / storage, AVB/verity, encryption, SELinux enforcing, Wi‑Fi / BT / calls /
|
||
camera / video / audio functional path. Nora display-during-video is device-
|
||
specific, not ROM.
|
||
|
||
## Next bacon / OTA note
|
||
|
||
This list is **device polish**, not the full upstream OTA path. Still follow
|
||
`docs/009-ota-rebuild-checklist.md` + `docs/009-upstream-rebase-ota.md` for
|
||
sync / Magisk / apps gates. Land polish as vendor apply scripts + postflight
|
||
checks where possible.
|
||
|
||
### Host commands (2026-10-16 polish bacon)
|
||
|
||
```bash
|
||
# From laptop (after WT edits):
|
||
./scripts/009-sync-ota-stage-to-host.sh
|
||
./scripts/009-relaunch-polish-bacon.sh
|
||
|
||
# On rom-dev01:
|
||
cat /srv/rom/artifacts/detach-bacon009polish20261016.status
|
||
tmux attach -t bacon009polish20261016 # optional
|
||
```
|
||
|
||
Apply wiring in ship bacon (order): euicc guard → thermal CDEV log → … → wave4
|
||
`drop_lab_only_commands` (configstore restore). See
|
||
`scripts/009-bacon-axp-stage1-combo4-cve-hi-releasekeys.sh`.
|