Skip to content

device-bluejay 0.1.3: publish the persistent accelerated Pixel 6a bring-up - #543

Closed
latentoperator wants to merge 4 commits into
derekross:mainfrom
latentoperator:bluejay-development-upstream
Closed

latentoperator wants to merge 4 commits into
derekross:mainfrom
latentoperator:bluejay-development-upstream

Conversation

@latentoperator

@latentoperator latentoperator commented Nov 8, 2026 •

Copy link
Copy Markdown
Contributor

Split revisions, 2026-11-09: this draft remains the immutable development integration reference. Proposed replacements are documentation #561, descriptor-driven initramfs #562, installer policy #563, and experimental bring-up files #565. A passive 0.1.1 candidate based on #346 is prepared locally pending coordination; it installs no experimental startup or slot-success writer. The audited devinfo write, missing exact executable hash, and still-pending fresh device backup are disclosed in #561. Further gs101 marking remains on hold. Kernel #225–#230 revisions are now published with verified human DCO tags and unchanged source trees. Signed-head policy/style/clean-merge gates pass; eight W=1 C objects, ten DTs and four bindings pass. All six published revised kernel heads now have green CI. They have not been flashed. All six remain drafts for the documented hardware and dependency gates. Prior submitted commits remain backed up locally, and the exact tested kernel remains unchanged. Trusty #231 has been marked ready at the maintainer's invitation.

What and why

The Pixel 6a now boots a persistent Arch Linux ARM/Omarchy desktop from slot B, with automatic polling ALIX touch, BCM4389 station reconnect and Mali-G78/Panfrost rendering. Upstream still describes Bluejay as unbooted. Publish the development source, guarded startup, installer/root protections and measured results so the port can be reviewed and continued. Tracking issue: #2; earlier first-light package proposal: #346.

Draft development milestone, not a release installer. The working phone uses a locally built kernel/module set, privately extracted firmware and a native patched Aquamarine library. These dependencies and the build-specific provenance pins are documented in device/bluejay/UPSTREAM.md. No firmware, credentials, images, captures or per-device addresses are included.

Changes:

  • Check the root label before repair/RW mounting, reject unsafe fsck results, and retain a pending marker until installed SSH is ready. Unknown development selectors stay in RAM rescue.
  • Honor BOOT_UPDATE_AUTO_WRITE=no in the installer and an optional fastboot serial pin; other descriptors retain the existing default.
  • Add omamobile-device-bluejay 0.1.3: passive storage/USB checks, watchdog policy, blacklists, bounded subsystem startup and GPU software fallback. Hardware/Wi-Fi/GPU require explicit opt-ins; automatic boot-image and boot-success writes stay disabled.
  • Add the development kernel fragment, qualified bring-up overlays, dual-silicon image repacking, a headless GLES pixel probe, current notes and an honest device-matrix row.
  • Keep the earlier private RAM-transition harness out of this submission. The published image tool can create its selector in a fresh project image; it does not require a previous private image archive.

Kernel companion source

The exact tested source at c934e316b starts at published omamobile commit 29a7a9326bf0. It is not a released kernel. All fourteen author DCO sign-offs have now been added locally by the author and verified. The signed topic PRs based on 1a9d1d4e9 are open as drafts below:

The common review integration passes W=1 builds of all nine changed C files, ten device trees with no new warnings, and all four binding checks. The seven signed topic publication gates now pass. Their code was verified unchanged by sign-off rebasing; the Wi-Fi DT topic is stacked on the signed PCIe/PHY dependency with identical own patch IDs. DT/MAINTAINERS reminders remain disclosed, and merge qualification is separate from these publication checks. The review correction adds PCI identity plus the generic fmac fallback and incorporates the already qualified INTx indices. That compatible-list correction and the newly rebased stack have not been flashed. The tested source is retained separately.

A later kernel review also confined the extra BCM4389 ring limits and changed bus-master ordering to Bluejay (c41e0aeba, unchanged from the reviewed narrowing); the changed object rebuilt with W=1 and the reviewer verified that other boards retain their original path. The intentionally generic NCM retry still needs reference-device runtime validation before merge. The Wi-Fi DT topic must follow its separate PCIe/PHY bindings. These are kernel merge conditions, not claims that the new rebased stack has run on another phone.

Device and build

  • Phone: Pixel 6a (bluejay), owner-tested.
  • Running kernel: 7.3.0-rc6-omamobile-bluejay-input-gated0.
  • Package source: omamobile-device-bluejay 0.1.3. The installed database still says 0.1.1 because later qualified startup files were staged directly; a freshly packaged/final rebased release install is not claimed.
  • Graphics: retained simpledrm 1080x2400, approximately 60 Hz; Panfrost Mali-G78 MC20, GLES 3.1, initial 151 MHz shader point. No native panel or full DVFS claim.

Checks

  • tools/pr-check origin/main passes at 6ad4651 (identifiers, blobs, AI trailers and parsing).
  • tests/all passes at the submitted head; rootfs-image smoke skips because this review worktree has no root image.
  • packages/omamobile-device-bluejay/test/all passes; independently rerun in the repository-required review.
  • Actual GNU CPIO extraction and device-tree readers verify selector creation/update, both RAM/persistent modes, both silicon variants, overlay application, unchanged kernel/board images and rejection of a forged DEVICE sidecar.
  • Actual boot_arch tests cover foreign/unlabelled roots, fsck status handling, marker persistence and stale-attempt rescue; the existing growth/installer/image/storage suites pass.
  • New package version is in the commit subject; commits have explanatory bodies and Codex co-author attribution.
  • Pre-submission repository review found no remaining mobile-code blocker. Companion kernel DCO and release integration remain separate.

On the phone

  • Persistent installation through the project installer passed its actual dry run and write, followed by normal reboots, USB SSH, root growth and automatic touch/Wi-Fi/SDDM/Hyprland/Quickshell startup.
  • Touch passed single/two-finger motion, all edge zones and complete releases with zero SPI errors/timeouts. Both volume-key pairs passed; Power release remains unresolved.
  • Wi-Fi passed firmware control, actual IRQ delivery, association/DHCP/DNS, repeated automatic reconnect and interface-bound HTTPS with TLS verification.
  • GPU passed all-pixel clear, solid shader and alpha-blended shader checks at 64x64, 257x257 and 1024x1024. The owner unlocked, opened an app and scrolled normally in the accelerated session. A subsequent normal reboot independently passed automatic GPU/touch/Wi-Fi/desktop startup with no GPU faults.
  • The initial GPU probe deferred on a missing clock provider and recovered through the bounded service. The corrected path holds unrelated CPU DVFS before registering ACPM clocks. The submission smoke also exposed a missing FUSE module; adding the exact matching module repaired the document portal.
  • New hardware/Wi-Fi opt-in conditions and the strengthened image/module provenance checks were verified offline; the underlying startup functionality ran on the phone. No additional phone flash was performed to publish this PR.
  • The final recent-ring smoke reports one DWC3 remote wakeup not configured line. The earlier full-boot capture also contains an inherited trace-format WARN and the intentional USI hold; the later dmesg ring has wrapped. These observations are disclosed, not filtered away or presented as a clean full-boot log.
  • Retained DECON refresh uses the existing project devmem helper at the audited GS101 display addresses; the blanket no-userspace-register-access checkbox would be inaccurate. No PMU userspace access is added. No bootloader/calibration/NV/vbmeta write occurs in this source; the original qualified install used the existing project boot-control helper once to mark B successful, with automatic marking still masked.
  • Not tested: fresh install from the final public/rebased stack, native panel, full GPU/CPU DVFS, Vulkan, Wi-Fi P2P/hotspot/roaming/endurance/suspend, Power release, audio, modem, Bluetooth, battery/charging, sensors, cameras or suspend. No claim about other phones.
tools/phone-smoke report after the FUSE repair

phone-smoke: bluejay

Kernel 7.3.0-rc6-omamobile-bluejay-input-gated0, tools at development, 2026-11-08T23:40Z.

Check Result Detail
boot (required) ok kernel 7.3.0-rc6-omamobile-bluejay-input-gated0, slot b, up 1476s
storage guard (required) ok 49 partitions; only userdata and slot B's boot images writable
root (required) ok ext4 on userdata, 100G free
kernel health (required) ok no oops or BUG; 0 WARN, 1 error lines in dmesg (0 known, 1 new)
new dmesg errors not ok N N: remote wakeup not configured x1
systemd units ok none failed
USB gadget (required) ok bound to 11110000.usb
display ok card0-Unknown-1 connected, 1080x2400
GPU ok render node on panfrost
session (required) ok Hyprland and the Omarchy shell running
lock ok PAM file present, the shell can lock (passwordPam true, locked true)
touch ok ALIX Touchscreen
buttons ok gpio-keys
Wi-Fi ok wlp1s0 on brcmfmac, link up
Bluetooth skip no controller
battery skip no battery power supply
sound cards skip none
sensors skip no sensor proxy
cameras skip no libcamera tools

Packages: hyprland 0.57.0-3.1, libcamera 0.7.2-4.22, mesa 1:26.2.4-1.4, omacam 0.3.3-1, omamessages 0.2.2-1, omamobile-apps 0.2.2-1, omamobile-device-bluejay 0.1.1-1, omamobile-firstboot 0.1.3-1, omamobile-keyring 20261026-1, omamobile-mirror 0.1.1-1, omamobile-omarchy 0.1.32-1, omamobile-owl-widgets 0.1.2-1, omamobile-shell 0.2.100-1, omamobile-voice 0.1.0-1, omaphone 0.2.3-1, omarchy 4.0.6-1, omatel-agent 0.2.4-2, quickshell 0.3.2.r7.g2b4501a-1

latentoperator and others added 4 commits November 8, 2026 17:34
Check the ext4 label before repair or writable mounting, stop on unsafe
fsck results and retain a pending marker until installed SSH is ready.
Keep development roots at image size until their first persistent boot
works, and leave an unknown development selector in RAM rescue.

Honor an explicit descriptor AUTO_WRITE policy and optional fastboot
serial pin. Existing descriptors retain the automatic-write default.
Stage both persistent and RAM-rescue Bluejay selectors through the image
tool, optionally applying an overlay to both silicon variants.

Exercise actual boot_arch ordering and failure holds, installer policy,
and synthetic vendor images through independent CPIO/device-tree tools.

Co-Authored-By: Codex <noreply@openai.com>
Record persistent Arch/Omarchy, polling ALIX touch, BCM4389 station
reconnect and Mali-G78 rendering on the owner's Pixel 6a. Add the
pinned development package, guarded startup and renderer fallback,
bring-up overlays, kernel fragment and source/reproduction limitations.

Keep hardware, Wi-Fi and GPU behind explicit opt-ins, preserve the
storage guard and leave automatic boot writes/success marking off.
Include no firmware, credentials, images or per-device captures.

This shares a tested development milestone, not a self-contained release
installer. Native panel, full DVFS, suspend and other untested hardware
remain marked unfinished. The kernel changes are submitted separately.

Co-Authored-By: Codex <noreply@openai.com>
Use the existing FAKE prefix convention so the publication identifier gate recognizes the deliberately mismatched test serial.

Co-Authored-By: Codex <noreply@openai.com>
Check both embedded silicon variants before creating repacked output, including a forged sidecar regression case. Verify the board and the four qualified stripped Wi-Fi module hashes before any load; their bytes match canonical build outputs after strip-debug.

Co-Authored-By: Codex <noreply@openai.com>
@derekross

Copy link
Copy Markdown
Owner

Thank you, Chris — getting a Pixel 6a to a persistent, accelerated Omarchy desktop with touch and Wi-Fi is a real milestone, and the notes, the smoke report and the honest "not tested" list make it easy to follow. Since this is a draft, here's what I think it needs to get out of draft:

  1. Please split it. The initramfs/init and tools/install-phone changes affect every phone and need their own PRs; the plans and notes could land on their own quite soon; the package should come last.
  2. The root label and fsck checks in boot_arch() run on every device, including my Pixel 7 Pro. Phones set up by hand before the installer may not have the omamobile-root label, and they'd stop in the initramfs. Could you put these behind a descriptor setting, or I'll check the label on the development phones first? Please also replace the omamobile-bluejay-firstroot marker with a device-neutral, descriptor-driven key, and explain why the bluejay image turns off the USB self-recovery reboot.
  3. The description says slot B was marked successful once with the boot-control helper. Marking B on the gs101 phones is on hold (Marking slot B successful: the plan after #125's hold, and for the Pixel 8 Pro #252), so please add to the notes which tool ran, the devinfo bytes before and after, and take a fresh read-only backup; please don't mark B again until we've agreed how.
  4. omamobile-bluejay-hardware-start writes DECON's trigger register with devmem at boot. We don't ship register writes from userspace; the init already does this from the descriptor's SIMPLEFB_TRIG_CON. Could you use that?
  5. The start scripts pin your kernel release and module checksums and depend on the kernel drafts (kernel config: the Pixel 9a's USB at boot, battery and charger, camera regulator, Wi-Fi host and speaker amplifiers #227-omamobile-omarchy 0.1.22: the app icon index keeps the largest size per app #229). That's right for your phone, but it belongs in device/bluejay/bringup/ until the kernel parts are merged.
  6. Peter's device-bluejay 0.1.0: the Pixel 6a's first-light package, nothing enabled #346 already proposes omamobile-device-bluejay 0.1.0. Could the two of you agree how to combine them? I'd suggest building on device-bluejay 0.1.0: the Pixel 6a's first-light package, nothing enabled #346 and continuing at 0.1.1. The 6a's thread is Device port: Pixel 6a (bluejay) #348 rather than Device port: Pixel 6 (oriole, gs101 / Tensor G1) – tracking issue, contributors welcome! #2.
  7. In docs/DEVICES.md, please mark the touch, Wi-Fi and GPU cells as running on a development kernel, with links to the kernel PRs in review.

Thanks again — this is a great base for the 6a.

@derekross

Copy link
Copy Markdown
Owner

Thanks for splitting this so carefully, Chris. With #561, #562, #563 and #565 covering everything here, I'll close this one in their favour. When you and Peter have agreed on the package, please open it as 0.1.1 on top of #346.

@derekross derekross closed this Nov 9, 2026
latentoperator added a commit to latentoperator/omarchy-mobile that referenced this pull request Nov 10, 2026
Split experimental services, overlays and probe tools from derekross#543. Retain
the exact old development-kernel pins and explicit enable gates; these
helpers refuse the revised kernel topics. Remove userspace display-register
access and use the descriptor/init trigger contract. The proposed SSH-ready
helper clears the neutral root marker only when USB and sshd are ready.
No new phone run or automatic installation is claimed.

Co-Authored-By: OpenAI Codex <noreply@openai.com>
latentoperator added a commit to latentoperator/omarchy-mobile that referenced this pull request Nov 11, 2026
Split experimental services, overlays and probe tools from derekross#543. Retain
the exact old development-kernel pins and explicit enable gates; these
helpers refuse the revised kernel topics. Remove userspace display-register
access and use the descriptor/init trigger contract. The proposed SSH-ready
helper clears the neutral root marker only when USB and sshd are ready.
No new phone run or automatic installation is claimed.

Co-Authored-By: OpenAI Codex <noreply@openai.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants