Skip to content
Closed
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
119 changes: 118 additions & 1 deletion device/oriole/NOTES.md
Original file line number Diff line number Diff line change
@@ -1,14 +1,131 @@
<!-- SPDX-License-Identifier: MIT -->
# Pixel 6 (oriole) bring-up notes

Findings for the Pixel 6 port, newest first. The port is a community effort tracked in [issue #2](https://github.com/derekross/omarchy-mobile/issues/2); the 6 Pro (raven) and 6a (bluejay) share it. Facts are marked [V] (checked, with the source) or [I] (an inference to check on a phone). Nothing has been flashed or booted yet: the phone facts come from read-only adb and fastboot queries of a stock phone.
Findings for the Pixel 6 port, newest first. The port is a community effort tracked in [issue #2](https://github.com/derekross/omarchy-mobile/issues/2); the 6 Pro (raven) and 6a (bluejay) share it. Facts are marked [V] (checked, with the source) or [I] (an inference or a recovered report still to check). The October 14–16 phone facts below came from read-only queries of a stock phone; later tests booted Linux on a Pixel 6.

Sources:

- `F/`: Google's public factory image `oriole-cp3a.261005.002.a1` ([developers.google.com/android/images](https://developers.google.com/android/images#oriole)), read with HTTP range requests; facts only, no files kept in the repository.
- `U/`: upstream Linux v7.0, `arch/arm64/boot/dts/exynos/google/gs101*.dts(i)`.
- `P/`: postmarketOS pmaports [`device/testing/device-google-oriole`](https://gitlab.postmarketos.org/postmarketOS/pmaports/-/tree/main/device/testing/device-google-oriole), [`soc-google-gs101`](https://gitlab.postmarketos.org/postmarketOS/pmaports/-/tree/main/device/testing/soc-google-gs101) and [`linux-postmarketos-gs101`](https://gitlab.postmarketos.org/postmarketOS/pmaports/-/tree/main/device/testing/linux-postmarketos-gs101) (7.0.6), which builds [gs101-mainline/linux @ d28cfa42](https://github.com/gs101-mainline/linux/tree/d28cfa4275b95e93f4b7a1e743de31b723bd21b8).

## 2026-10-31: build provenance and recovery limits

This describes the unpublished stack in the October 30 smoke report. It is
not a reproducible build of the public Oriole descriptor or an installer
qualification. The records and recovered artifacts were audited offline;
this audit did not reflash or inspect the phone's current partitions.

- **Kernel** [V, recovered source metadata]: the base is
[`omamobile` commit `4bd35d34abb9`](https://github.com/derekross/linux-omarchy-mobile/commit/4bd35d34abb92a745479a2de62816b28304b4449).
The native build used a detached worktree at that commit with unpublished
local changes, producing
`7.3.0-rc6-oriole-hsi1-late-dev0-g4bd35d34abb9-dirty`.
There is no published branch containing that complete integration tree.
- **Packages** [V, recovered sources and build receipts]: the mobile source
was local branch `pixel6-bringup`, based on
[`1155a7c352ed`](https://github.com/derekross/omarchy-mobile/commit/1155a7c352edf6313fb4e2ff02f054c9cb1902b6),
with unpublished changes and separate package-source snapshots. The
`omamobile-device-oriole 0.1.9-2` source matches all 24 files in its build
receipt; its archived package and the GPU, Wi-Fi, Bluetooth, torch,
battery, charger, NFC and backlight packages are local builds, not
packages from the project's published repository. The device package
was built in an offline ARM64 package container. The smoke report
establishes the installed versions; source receipts alone do not.
- **Shell** [V, recovered package]: local `omamobile-shell 0.2.12-1` was
rebuilt from the installed local 0.2.10 source on
`omamobile-oriole-canvas-budget`, preserving its phone-specific files and
backporting the Canvas budget fix from
[`be98c2c22e2b`](https://github.com/derekross/omarchy-mobile/commit/be98c2c22e2b260300c59787861fee200ef2ad58).
The recovered package's `runtime.js` matches that commit byte for byte.
This is not a claim that upstream shell 0.2.12 was installed unchanged.
- **Keyring correction** [V, recovered root image]: the local rebuild
`omamobile-keyring 20261013.oriole1-1` retained the upstream public key
and added a local public signing key to `omamobile.gpg` and its trusted
list. The keyring payload matches the installed package's mtree in the
recovered image; the additional key verifies the archived local device
package. The local key therefore entered the packaged trust list, not
only the build environment. The rebuilt source recipe has not yet been
recovered; this establishes the package contents, not its exact build
command. This does not follow the contributor-build
policy, and that rebuilt keyring must not be distributed as the project's
keyring. Neither the modified keyring nor signing material is included
in this PR. The correction for the next image is the existing
[contributor build workflow](../../docs/RELEASES.md#10-signing):
unchanged upstream keyring, preserved published-package signatures,
separate `keys/oriole/` and `out/oriole/`, and local-key trust confined to
the build image. Any later on-device local trust is a separate owner
action. No installed trust database was changed by this audit.
- **Slot B installation** [I, historical run records]: the initial root
installation used the local `tools/install-phone` after its dry run,
writing `boot_b`, `vendor_boot_b`, `dtbo_b` and `userdata`. Later kernel
deployments used the local `bin/omamobile push` with B-image readback.
The flags-3 stock-copy write to `vbmeta_b` and the B-success flag
update at offset `0x35` in `devinfo` were done without project approval.
The run records report the other `devinfo` bytes preserved and read-only
protection restored. Early verification tests restored original
`vbmeta_b`; the later native-root run records retain the B verification
change. No further `devinfo` or `vbmeta` writes will be made without
the maintainer's explicit approval. These are historical facts, not
instructions or an extension of the public installer's permissions.
- **Local deployment tools** [V, recovered October 28 source snapshot]:
`tools/install-phone` is byte-identical to the version at the mobile
base `1155a7c352ed` above. `bin/omamobile` has local changes; its `push`
path adds a `PUSH_MODULES_POLICY=require-match` option, follows image
symlinks for readback length, and uses systemd's reboot on a native root.
The complete two-file comparison against that base is in
[PR #200](https://github.com/derekross/omarchy-mobile/pull/200).
This is a dated source-snapshot comparison, not proof of the exact
script bytes used for every earlier deployment, and not a comparison
against today's main. The recovered Oriole descriptor still allows
only `boot_b vendor_boot_b dtbo_b userdata`; the metadata writes were
separate bring-up operations.
- **Slot A and bootloaders** [I, historical run records]: before Linux
installation, the matching factory build and full OTA booted stock on
B and then A, both reporting `slider-17.0-15681480`. The separate
per-slot bootloader-version variables were unsupported, so independent
partition-version equality was not established. The subsequent Linux
records preserve A's images and slot record. A fresh partition audit
has not been performed for this PR. `userdata` now contains Linux;
returning to Android requires deliberate data restoration, not simply
switching to slot A.

### Historical B-success write: mechanism and evidence limits

[V, recovered October 18 source and execution transcript] The separate
`mark-b-success-37.py` wrapper made `devinfo` writable with `blockdev --setrw`
on its containing boot LUN and then the partition. It restored the partition
and then the disk to read-only with an EXIT trap, backed by a 20-second
systemd undo timer and a 120-second watchdog. The wrapper did not stop the
storage watcher or change its allowlist. The recovered init's watcher skipped
nodes already marked in `/run/ufs-ro/`; it did not continuously reassert their
read-only flags. This is how the explicit opening remained possible.

The helper wrote a **128-byte header**, with only bit 1 at offset `0x35`
changed, followed by comparison of the full 8192-byte record. The gate's
2026-10-18 09:04:51 UTC execution result reports the one-bit change, unchanged
slot-A/opaque bytes, both nodes restored read-only and refreshed backups
verified. This was not a one-byte write.

The next Fastboot controller completed at 09:06:19 UTC after requiring
`current-slot=b`, `slot-successful:b=yes` and `slot-unbootable:b=no`, without
issuing set-active. The native audit at 09:11:29 UTC reported B success
persisting and A's record unchanged. These are recovered executed assertions
and results; the original Fastboot output and contemporaneous retry count
have not been recovered.

The original gate's before/after images are listed in the private backup
inventory but are not available locally, and the backup host was unreachable
during this audit. The exact bytes at `0x30`–`0x37` therefore remain unverified
for the original pair. This gap is not filled by deriving bytes from the
helper's logic or from a later backup. No new metadata write was attempted.

## 2026-10-30: native boot and RTC alarm status

- [V] A Pixel 6 booted native Linux from slot B with its root on `userdata`, using the unpublished `7.3.0-rc6-oriole-hsi1-late-dev0-g4bd35d34abb9-dirty` kernel and a separate integration package set. A read-only `phone-smoke --here --json` run passed all required checks: storage guard, root, USB and desktop session. It reported no failed units, oops, BUG or WARN, and 48 existing dmesg error lines; sound, sensors and cameras were skipped. This is a test-phone result, not an installation claim for the public images.
- [V, source] On the published kernel's `omamobile` branch, the S2MPG10 alarm callback reads `S2MPS14_REG_ST2` at offset `0x08`. On S2MPG10 that is `INT3M`; [Google's register definitions](https://android.googlesource.com/kernel/google-modules/soc/gs/+/204078758737776959ecc85c527a00fd2cabb62f/include/linux/mfd/samsung/s2mpg10-register.h) place `STATUS2` at `0x0d`, and its [RTC driver](https://android.googlesource.com/kernel/google-modules/soc/gs/+/204078758737776959ecc85c527a00fd2cabb62f/drivers/rtc/rtc-s2mpg10.c) reads the alarm-0 status bit there. The proposed kernel fix is S2MPG10-specific; the S2MPG12 path remains unchanged. The test phone's unpublished kernel already contains the correction.
- [I] The smoke test does not exercise a live pending alarm, alarm delivery, suspend or wake. Those paths remain to be independently verified.

## 2026-10-16: stage 0, part 2: a retail Pixel 6 (stock, bootloader locked)

Read-only queries (`adb shell getprop`, `/dev/block/by-name`, `fastboot getvar all`) of a stock phone; facts only, no serial numbers.
Expand Down
Loading