Skip to content

device-bluejay: descriptor and notes from Google's GPL bluejay device tree - #265

Closed
PeterXMR wants to merge 1 commit into
derekross:mainfrom
PeterXMR:device-bluejay-descriptor-v2
Closed

PeterXMR wants to merge 1 commit into
derekross:mainfrom
PeterXMR:device-bluejay-descriptor-v2

Conversation

@PeterXMR

@PeterXMR PeterXMR commented Nov 1, 2026 •

Copy link
Copy Markdown
Contributor

Stage 0, part 1 of the Pixel 6a port (#2): device/bluejay/device.conf, an empty board dtbo source, NOTES.md, the Wanted row in docs/DEVICES.md, and bluejay in the two pack-time tests (tests/storage-guard, tests/usb-product). Nothing has been flashed or booted, no 6a has been queried, and no factory image has been read. No matrix row: no gate has passed, so bluejay stays under Wanted as mustang does, with its notes linked.

Where the values come from. Google's GPL device trees for bluejay and gs101 (kernel/devices/google/bluejay and kernel/devices/google/gs101 at android-16.0.0_r0.9), with the raviole tree as the oriole control, and oriole's recorded image facts for what a 6a shares with a Pixel 6. The descriptor carries no comments; the table "What device.conf rests on" in NOTES.md gives each key its source and mark.

  • [V] from the GPL trees: the same two gs101 base DTBs as oriole and their soc_id (0x9845000 A0, 0x9845001 B0, which also settles the labels oriole left [I]); the six bluejay board ids, with MP 1.0 at 0x20506/0x10000; the S6E3FC3 panel at 1080x2400; an ST FTS touch controller on SPI; and why the board DTS leaves out the two LDOs upstream's oriole DTS describes (shared gs101 PMIC rails).
  • [I] from oriole, to be confirmed by the bluejay factory image: that a retail 6a boots the MP 1.0 dtbo entry (raven boots its DVT entry, so this needs the stock dtbo.img or ro.boot.dtbo_idx), VENDOR_CMDLINE, VENDOR_PAGESIZE, OS_VERSION/OS_PATCH_LEVEL (informational on a GKI header; stock ships 0), and the partition layout (BOOT_IMAGES, UFS_RW: no init_boot, no vendor_kernel_boot).
  • Open, and a stop before any flash: whether the 6a's vbmeta_b chains boot and vendor_boot, so whether slot B needs AVB_FOOTERS (RELEASES §3.1). The factory vbmeta.img and the boot images' footers decide it; the notes list what to read.
  • Unset: SIMPLEFB_TRIG_CON and BRINGUP_PROBE, until a booted 6a has read gs101's DECON register (L-4, L-365).

The kernel side. DTB=exynos/google/gs101-bluejay.dtb is a first-light board DTS (binding entry plus gs101-bluejay.dts: oriole's file with bluejay's model, compatible and soc_id, without the LDO block), prepared on a local omamobile-gs101-bluejay-dts branch and to be opened as a pull request against linux-omarchy-mobile's omamobile under the title "arm64: dts: exynos: google: Add Pixel 6a (bluejay) first-light board". It is in no published kernel, and the Wanted row says so (L-149). DEVICE=bluejay bin/omamobile images can't run until that DTS is in a kernel build; this descriptor PR doesn't need it to merge.

Checks. tools/pr-check origin/main and tests/all in the project's build container on this commit (results below); dtc -@ compiles the overlay and fdtoverlay applies it to the host-built gs101-bluejay.dtb with no change to the DTB; the identifier grep over the diff finds nothing; Google's tree was re-read at the tag and every cited value matches; the Wanted row is the only bluejay mention in DEVICES.md.

Container run, 2026-11-01, omamobile-build:24.04 over main 053583e (the commit was then rebased onto 7ea8ab5 with the same diff; on the host, tools/pr-check origin/main on the rebased commit passes its six identifier, blob and sign-off checks, and bash -n of both touched tests passes):

## tools/pr-check origin/main
ok - no MAC or Bluetooth addresses
ok - no getprop dumps
ok - no serial numbers
ok - no IMEIs
ok - no vendor blobs or binaries
ok - no Signed-off-by from an AI agent
ok - touched scripts parse
## tests/all: rc 0, 262 ok, no failures
# tests/storage-guard
ok - cheetah and felix pack the six-name list, oriole and bluejay their three boot images and userdata
# tests/usb-product
ok - cheetah keeps 'Pixel 7 Pro (M1)'; felix, oriole and bluejay use their NAME
skipped: tests/rootfs-smoke (no root image in out/), the phone-provision end-to-end run (no archlinux image), tests/local-builds (no gpg)

Kernel pull request: derekross/linux-omarchy-mobile#179 (omamobile-gs101-bluejay-dts, head e85110d61).

… tree

Stage 0, part 1 of the Pixel 6a port (derekross#2), from Google's GPL device
trees for bluejay and gs101 (kernel/devices/google/bluejay and
kernel/devices/google/gs101 at android-16.0.0_r0.9) and the image facts
oriole's notes record, with no bluejay factory image read and no phone:
device.conf, an empty board dtbo source, NOTES.md, the Wanted row in
DEVICES.md and bluejay in the two pack-time tests. No matrix row yet:
no gate has passed, so the device stays under Wanted as mustang does.

Confirmed from the GPL trees: bluejay boots the same two gs101 base DTBs
as oriole, soc_id 0x9845000 (A0) and 0x9845001 (B0), which also settles
the labels oriole's notes left [I]; six board ids, the MP 1.0 entry being
0x20506/0x10000; the S6E3FC3 panel at 1080x2400 and an ST FTS touch
controller on SPI; and that the two LDOs upstream's oriole DTS describes
are shared gs101 PMIC rails, so the board DTS leaves them out.

Still inferred from oriole and marked [I] in NOTES.md: that a retail 6a
boots the MP 1.0 dtbo entry (raven boots its DVT entry, so only the
stock dtbo.img or ro.boot.dtbo_idx settles it), the vendor command line,
the vendor_boot page size, the OS version and patch level, and the
partition layout (no init_boot, no vendor_kernel_boot). The bluejay
factory image confirms or corrects them. Open, and a stop before any
flash: whether the 6a's vbmeta_b chains boot and vendor_boot, so whether
slot B needs AVB_FOOTERS; stock's vbmeta.img and the boot images'
footers decide it. No SIMPLEFB_TRIG_CON and no BRINGUP_PROBE until a
booted 6a has read gs101's DECON register.

The DTB, exynos/google/gs101-bluejay.dtb, comes from the first-light
board DTS prepared on a local omamobile-gs101-bluejay-dts kernel branch,
to be opened as a pull request against linux-omarchy-mobile's omamobile;
images can't be built for bluejay until that DTS is in a kernel build.
The descriptor carries no comments: the provenance of every key is a
table in NOTES.md.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
@derekross

Copy link
Copy Markdown
Owner

Thank you, Peter, and welcome. This is a careful stage-0 PR: every value in the descriptor has a source and a mark, the [I] values are honest about what only a retail 6a or its factory image can settle, and the AVB question is written down as a stop before any flash rather than guessed.

What I checked:

  • Against Google's GPL trees at android-16.0.0_r0.9: gs101-bluejay-mp1_0.dts has board_id = <0x20506> and board_rev = <0x010000>, model "GS101 BLUEJAY MP 1.0", and gs101-b0.dts/gs101-a0.dts have soc_id 0x09845001 and 0x09845000. The panel compatible (samsung,s6e3fc3_6a), the deleted vddi-supply and the 1080x2400 mode match gs101-bluejay-display.dtsi.
  • bin/omamobile loads the descriptor, and on today's main merged with your branch tests/storage-guard, tests/usb-product and tests/device-overlays pass. bluejay packs boot_b vendor_boot_b dtbo_b userdata and "Pixel 6a". tools/pr-check is clean.
  • The descriptor carries no display-register write and no probe, so nothing of another phone's ends up in its initramfs.
  • With no omamobile-device-bluejay package, bin/omamobile initramfs does fall back as you say (no bootctl, freestanding devmem and musl reboot-to).

A few small things. I'm happy to make them myself before merging, since you've allowed edits:

  1. Link the kernel PR. The notes (the board DTS paragraph and the DTB row) and the docs/DEVICES.md row still describe the board DTS as "prepared on a local branch, to be opened as a pull request". It's open now as linux-omarchy-mobile#179, so they can link it. An open kernel PR is enough for this descriptor to merge. Images can't be built until device/tegu MODEM-PLAN: G3-A, the automatic chip select boot #179 lands, and the notes already say so.
  2. The "review lesson" references. The notes cite "review lesson L-365" twice, plus "L-4". Those are our internal review notes, and L-365 doesn't exist. Please say the point in plain words instead: the dtbo entry's index isn't its board id, and gs101's DECON trigger register gets read on a booted 6a before anything writes it.
  3. The P/ source. No value in the notes rests on it. Every fact is [V] from Google's trees or [I] from oriole's notes. I'd drop the P/ line and the "(P/ builds one)" aside, so the facts stand on Google's sources alone.

A question, not a blocker: do you have a Pixel 6a to bring up, or is this groundwork for whoever does? Either is welcome. If you have one, the next step is the read-only stock capture in your "Still needs a phone" list (CONTRIBUTING §5.1, with captures kept off the repository), and then the factory image's vbmeta.img, to answer the AVB question before anything is flashed.

Thanks again. With those three edits in, this is ready to merge.

@derekross derekross closed this in 8254cf4 Nov 1, 2026
@derekross

Copy link
Copy Markdown
Owner

Landed on main as 8254cf4 (squashed, authored by you), with the three edits I listed. Welcome aboard, Peter. The next step is the first boot from kernel #179.

robertzas pushed a commit to robertzas/omarchy-mobile that referenced this pull request Nov 14, 2026
… tree (derekross#265)

Stage 0 for the Pixel 6a (bluejay, gs101): the device descriptor and
notes, every value sourced from Google's GPL bluejay device tree at
android-16.0.0_r0.9 (board id 0x20506, rev 0x10000, A0/B0 soc_id, the
S6E3FC3 panel at 1080x2400). The descriptor carries no register write
and no probe. Images wait for the board DTS in linux-omarchy-mobile#179.

Maintainer edits at merge: link kernel derekross#179, describe the board-id rule
in plain words, and drop a source line that backs no value.

Closes derekross#265

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.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