Skip to content

Device port: Pixel 6a (bluejay) #348

Open

Description

@PeterXMR

The Pixel 6a's own thread, split out of #2 (the Pixel 6's) as asked there. The gs101 context, the bring-up checklist and the safety rules in #2 apply here too.

Status

Wanted: a 6a owner who can unlock it and boot slot B

I can't wipe mine. Read CONTRIBUTING §3 first: only slot B and userdata are ours, slot A's stock is the way back.

  1. The read-only stock capture first (CONTRIBUTING §5.1), on stock, before anything is written: bin/omamobile capture bluejay-stock and fastboot getvar all. A locked phone can do the unrooted part. Keep the capture in ~/android/reference/ and post facts only: ro.boot.dtbo_idx (the descriptor assumes 0x20506/0x10000, [I]), hw-revision, build id and patch level, the A/B partition list and slot, and with root the soc_id (A0 0x9845000 or B0 0x9845001).
  2. Once the AVB settings are in, build from main (DEVICE=bluejay bin/omamobile kernel, initramfs, images; kernel clone on omamobile) and flash slot B only, with you at the phone: DEVICE=bluejay bin/omamobile flash. The screen stays on the boot logo (no display register is written on bluejay yet); the sign of life is the USB network at 172.16.42.1 and SSH. A fallback to slot A is the expected failure.
  3. Post uname -r and the tools/phone-smoke report, or how far it got (fastboot, boot logo, fallback).

No serials, IMEIs, MACs or captures. I'll record the result in the notes and take #346 out of draft on top of it.

Activity

  1. derekross commented on Nov 3, 2026

    @derekross
    Owner

    Thank you, Peter, this is a clear and safe call for testers. #350 settles the AVB question and is merged, so DEVICE=bluejay bin/omamobile images adds boot's footer, which needs avbtool (tools/fetch-toolchain fetches it).

    A few things worth adding to the steps, for whoever picks this up:

    • Both bootloader slots on the same version (CONTRIBUTING §3.1). If getvar all doesn't show both matching, sideload the full OTA twice from stock so each slot gets the same build, as the 8a's notes describe. Please don't flash bootloader partitions by hand.
    • Your stock build. The descriptor's AVB rollback index is from CP3A.261005.002.A1. If your phone runs an older build, tell us first and we'll use its value.
    • What a failed boot looks like. Slot B isn't marked successful, so each boot attempt spends one of the bootloader's three tries, and after about a minute without a USB host our initramfs restarts the phone, spending them quickly. Then stock Android boots from slot A. flash doesn't touch userdata, so stock should come up normally. If you see a recovery screen offering "Try again" or "Factory data reset", choose neither: hold Volume Down + Power for about 20 s to reach fastboot.
    • Fastboot is by the buttons. reboot-to bootloader won't reach it on a first boot.

    Once a 6a has booted, #346 comes out of draft on top of it.

  2. latentoperator commented on Nov 11, 2026

    @latentoperator
    Contributor

    Pixel 6a boot/smoke update, November 11, 2026: this handset now boots slot B into a persistent Arch Linux ARM root on userdata, with systemd, USB SSH, SDDM, Hyprland and Omarchy's shell. The earlier “never booted” status is superseded by this development result.

    uname -r: 7.3.0-rc6-omamobile-bluejay-input-gated0; tested kernel source: c934e316b. This uses historical, manually matched modules and overlays. It is not a qualification of today's release kernel or a ready-to-flash installer.

    The November 11 tools/phone-smoke report passed all six required checks:

    Check Result
    Boot Expected kernel, slot B
    Storage guard 49 partitions checked; only userdata and permitted slot-B boot images writable
    Root Persistent ext4 on userdata
    Kernel health No oops/BUG, no new warning/error lines
    USB gadget Bound, SSH working
    Session Hyprland and Omarchy shell running

    Optional checks found retained simpledrm 1080×2400 output, Panfrost rendering, ALIX polling touch, GPIO keys and connected BCM4389 station Wi-Fi; no failed systemd units. Owner checks passed unlock, app launch/scrolling and both physical browser Back edges, keyboard-first dismissal and native Qt dialog dismissal with the combined userspace artifact 83326ed6d09684ae80f38c9360e8c4e864d32614 (omamobile-shell 0.2.119-1). The newer kernel topics have not been booted by those userspace checks.

    The durable boot, AVB/image settings, dependency and limitation notes landed in #561: NOTES, UPSTREAM, PERSISTENCE-PLAN and GPU-PLAN. Current userspace results are also committed in #582, with browser fitting in #581. The split initramfs, installer and historical bring-up work remains in #562, #563 and #565; I am coordinating the passive package with #346 rather than opening a competing package PR.

    Limits remain: retained display rather than native panel modesetting; GPU qualification at 151 MHz; polling touch; Power-key release unresolved; Wi-Fi station only. Audio, modem, Bluetooth, battery/charging, sensors, cameras, full DVFS/Vulkan and suspend are unqualified. USB replug and laptop-sleep behavior still need reference-device qualification of the revised NCM path.

    AUTO_WRITE=no and masked automatic boot-success remain in force. The historical slot-success write and its audit are explicitly disclosed in NOTES; further GS101 slot marking is on hold under #252. No protected-data write occurred during the November 11 userspace checks. Private captures, device identifiers, calibration data and vendor blobs are not attached.

  3. latentoperator commented on Nov 13, 2026

    @latentoperator
    Contributor

    Pixel 6a development update, November 13, 2026:

    The handset is now on Omarchy 4.5.0-3, omamobile-omarchy 0.1.44-1 and omamobile-shell 0.2.127-1. The filtered mobile migration run has zero normal pending migrations: 71 processed (including conditional no-ops), 24 held by the packaged mobile filter and three optional app/toolchain installs skipped for this lab profile. Missing diffutils was installed and the affected mise-wrapper migration rerun. All 28 mobile patches verify and pacman -Dk passes.

    The signed revised kernel integration 720a49523373, release 7.3.0-rc6-omamobile-bluejay-review-20261111, has now passed staged headless qualification:

    • RAM-rescue and persistent Arch-root baselines with 49 partition guards.
    • Normal USI/SPI/ALIX probing, followed by an owner two-contact/full-release check, without the removed staging/probe-hold parameters (kernel #229).
    • A 64 MiB restricted pool; then revised PHY/controller binding and a one-try PCIe link to 14e4:4441 with the endpoint initially unbound (#227).
    • One matched BCM4389 bind, verified restricted DMA below 4 GiB, managed-station association/DHCP and interface-bound HTTPS 200; IRQ 725 → 744, observed tx_oversize_dropped=0 (#228, DT #230).
    • Active zstd zram at approximately half kernel-visible RAM after installing the missing generator. This was activation/configuration proof, not a memory-pressure test.

    The phone has since returned to its known working graphical historical kernel, 7.3.0-rc6-omamobile-bluejay-input-gated0. Touch, Panfrost, Wi-Fi, Hyprland/shell and all six required smoke checks pass. The revised kernel's desktop/GPU startup integration remains the next step; these headless results do not qualify native panel modesetting, latency, suspend/endurance, roaming/hotspot or a release installer.

    The locally active device package remains held with IgnorePkg=omamobile-device-bluejay; passive package #346 has not been installed or qualified. AUTO_WRITE=no, masked automatic boot-success and the GS101 marking hold remain. A separate Fastboot-only retry refill observation is recorded in #252. No identifying captures, firmware or calibration data are attached.

  4. latentoperator commented on Nov 15, 2026

    @latentoperator
    Contributor

    Pixel 6a attended-baseline update, November 14–15:

    The revised signed integration 720a49523373, release 7.3.0-rc6-omamobile-bluejay-review-20261111, now starts the persistent accelerated desktop automatically through a separate opt-in development path. Two corrected boots passed, first at B2 and then one normal reboot to B1. On both, the owner confirmed unlock, scrolling, opening apps and both battery percentage displays.

    Each corrected boot passed pinned USB SSH, initial host logging acknowledgment, full B/nine protected/module checks, 49 storage guards, all eight startup units, nine GLES pixel readbacks, Panfrost compositor access, ALIX touch and station-bound verified HTTPS. The saved final smoke report passed all six required checks: boot, storage guard, root, kernel health, USB gadget and session, with no failed units or pending startup markers.

    Three station-profile-only reconnects passed in 6.56, 6.15 and 3.06 seconds. The original bounded phone traffic unit completed 600.000115 seconds, sixty verified 1 MiB downloads and 62,914,560 bytes. Its host controller failed when blocking oneshot dispatch exceeded the SSH deadline; the same phone unit kept running. A read-only observer resumed that existing run and recovered the full retained boot journal. There was a live host-logging gap. The phone result passed; the original host harness did not have a clean end-to-end pass. The pre-hardware command-line refusal is also retained as a failed attempt.

    Exact tested mobile source is now preserved at bringup/bluejay-daily-baseline-tested-20261114, runtime cde98c4. Draft #821 publishes it on current main with unchanged runtime bytes, clearer prerequisite/fallback documentation and fourteen controls wired into the test suite. Local tests/all passed on that publication tree. Private handset profiles, firmware, snapshots and consumed qualification/recovery controllers are not published or reusable install instructions.

    This supersedes my earlier report that the revised desktop integration was still pending and that the phone had returned to the historical kernel. It remains an attended development baseline, not daily-driver or release-install readiness. Startup still requires the logging-host gate. Actual pack/model/scaling, battery health/time estimates and charge direction are unqualified: the reported 628-hour estimate is a defect, and kernel/UPower states have disagreed. No charger/model/Type-C trial was performed. Native display, reliable lock/wake/sleep, idle drain, host-free startup, update packaging and sustained carry remain open. Modem/calls/text are deferred.

    Current saved boundary is unsuccessful B1: no further ordinary reboot or push without the prepared attended recovery path. AUTO_WRITE=no, masked boot-success, historical pins, device-package hold and storage guards remain. The passive package #346 is still not qualified by these results.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions