Skip to content

device-bluejay: the factory image settles AVB, boot gets a footer - #350

Merged
derekross merged 1 commit into
derekross:mainfrom
PeterXMR:device-bluejay-avb-facts
Nov 3, 2026
Merged

derekross merged 1 commit into
derekross:mainfrom
PeterXMR:device-bluejay-avb-facts

Conversation

@PeterXMR

@PeterXMR PeterXMR commented Nov 2, 2026

Copy link
Copy Markdown
Contributor

The stage-0 facts the maintainer asked for in #2 (comment), before anyone flashes a 6a. No phone was involved; everything comes from the current factory image bluejay-cp3a.261005.002.a1, read with HTTP range requests (about 42 MB of a 2.6 GB zip). Facts only, no files.

The AVB answer: stock vbmeta chains boot (rollback index location 2), vbmeta_system and vbmeta_vendor. vendor_boot and dtbo are hash descriptors inside vbmeta, with unsigned footers. boot's footer is SHA256_RSA4096, rollback index 1791158400, security patch 2026-10-05, padded to 0x4000000. So slot B's boot needs a footer and the other two don't, and device/bluejay/device.conf gets

AVB_FOOTERS="boot:0x4000000"
AVB_ROLLBACK_INDEX=1791158400
AVB_SECURITY_PATCH=2026-10-05

Also confirmed from the same image (the provenance table in NOTES.md moves these from [I] to [V]): build CP3A.261005.002.A1; no init_boot or vendor_kernel_boot; vendor_boot v4 at page size 0x800 with the two gs101 soc_ids and oriole's bootargs; dtbo.img has exactly the ten entries Google's device trees predicted, with 0x20506/0x10000 at index 9. Which entry a retail 6a boots (ro.boot.dtbo_idx) still needs a phone.

Separate from #346 on purpose: that package waits on a boot, this is a facts change that should land first. #346's copy of the descriptor will need the three new lines once this merges. The tester call (#348) can now name the image settings.

tools/pr-check passes. tests/images-layout's descriptor and layout checks pass on the laptop; its image steps need Linux sed and run in CI.

The notes listed AVB_FOOTERS as open and said no flash until it is
settled, because nobody knew whether a 6a's stock vbmeta chains boot and
vendor_boot. The maintainer asked for the facts from the current factory
zip before anyone flashes (derekross#2).

Read on the laptop with HTTP range requests from the current factory
image, bluejay-cp3a.261005.002.a1 (facts only, no files kept): vbmeta
chains boot (rollback index location 2), vbmeta_system and vbmeta_vendor;
vendor_boot and dtbo are hash descriptors inside vbmeta. boot's footer is
SHA256_RSA4096, rollback index 1791158400, security patch 2026-10-05,
padded to 0x4000000; vendor_boot's and dtbo's footers are unsigned. So
slot B's boot needs a footer and the other two don't, and device.conf
gets AVB_FOOTERS="boot:0x4000000", AVB_ROLLBACK_INDEX=1791158400 and
AVB_SECURITY_PATCH=2026-10-05, stock's own values, never higher.

The same read confirms the rest of the descriptor: build
CP3A.261005.002.A1 (OS_VERSION, OS_PATCH_LEVEL), no init_boot or
vendor_kernel_boot (BOOT_IMAGES), vendor_boot v4 at page size 0x800 with
the two gs101 soc_ids and oriole's bootargs, and a dtbo.img with exactly
the ten entries Google's device trees predicted, 0x20506/0x10000 at
index 9. Which entry a retail 6a boots stays to be read from a phone.

load_device accepts the descriptor, and tests/images-layout's descriptor
and layout checks pass on the laptop; its image steps need Linux sed and
run in CI.

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

Copy link
Copy Markdown
Owner

Thank you, Peter, this settles the question we stopped on, with facts anyone can re-read from Google's public image. The descriptor change is exactly the shape the 7 Pro and the 8a use. The index and patch are stock's own: 1791158400 is 2026-10-05, the same value akita and husky carry for that build. And push and the installer refuse a boot_b of a different size, so a wrong size guess costs a message, not a boot.

One thing for the tester call rather than this PR: the rollback index is the newest build's. A tester whose slot A runs an older 6a build should tell us, and we'll use their build's index instead. Merging; #346 then takes the three lines into its copy of the descriptor.

@derekross
derekross merged commit 45ae255 into derekross:main Nov 3, 2026
2 checks passed
PeterXMR added a commit to PeterXMR/omarchy-mobile that referenced this pull request Nov 3, 2026
…bled

The package device/bluejay/device.conf names, so tools/build-rootfs can
build a 6a image with the [omamobile] repository configured: felix's
first-light shape without the fold's lid and cover-touch parts, plus
the suspend-to-idle drop-in. The storage udev rule, the usb0 address
with USB as the last gateway, NetworkManager off usb0, logind ignoring
the power key, [Sleep] on s2idle with no hibernation, the mirrorlist
and pacman hook, reboot-to (cheetah's source) and a copy of the
descriptor; provides and conflicts omamobile-device, depends on
omamobile-keyring.

The sleep drop-in is raven's, byte for byte: gs101's firmware offers
PSCI SYSTEM_SUSPEND, so the kernel lists deep, and Linux has no gs101
sleep sequence to wake from it (device/raven/SUSPEND-PLAN.md).
Bluejay's bootargs, like oriole's, don't set mem_sleep_default, so the
file is what keeps systemctl suspend on s2idle, as asked in the review
of linux-omarchy-mobile#179.

No service is enabled, no modprobe policy ships, and there is no
omamobile-bootctl: marking slot B on the gs101 phones is on hold
(derekross#252), and bin/omamobile initramfs builds its freestanding helpers
without the source, as for oriole. No Pixel 6a has run it; every file
is a copy of one that runs on raven or cheetah, and test/all pins each
copy. The descriptor copy carries the AVB lines the factory image
settled (derekross#350), the same bytes as device/bluejay/device.conf. The
first boot's phone-smoke report goes to derekross#348, the 6a's own thread, and
NOTES.md.

test/all and tests/packages pass on the laptop. The README carries the
SPDX header line raven's and tegu's do.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
PeterXMR added a commit to PeterXMR/omarchy-mobile that referenced this pull request Nov 11, 2026
…bled

The package device/bluejay/device.conf names, so tools/build-rootfs can
build a 6a image with the [omamobile] repository configured: felix's
first-light shape without the fold's lid and cover-touch parts, plus
the suspend-to-idle drop-in. The storage udev rule, the usb0 address
with USB as the last gateway, NetworkManager off usb0, logind ignoring
the power key, [Sleep] on s2idle with no hibernation, the mirrorlist
and pacman hook, reboot-to (cheetah's source) and a copy of the
descriptor; provides and conflicts omamobile-device, depends on
omamobile-keyring.

The sleep drop-in is raven's, byte for byte: gs101's firmware offers
PSCI SYSTEM_SUSPEND, so the kernel lists deep, and Linux has no gs101
sleep sequence to wake from it (device/raven/SUSPEND-PLAN.md).
Bluejay's bootargs, like oriole's, don't set mem_sleep_default, so the
file is what keeps systemctl suspend on s2idle, as asked in the review
of linux-omarchy-mobile#179.

No service is enabled, no modprobe policy ships, and there is no
omamobile-bootctl: marking slot B on the gs101 phones is on hold
(derekross#252), and bin/omamobile initramfs builds its freestanding helpers
without the source, as for oriole. No Pixel 6a has run it; every file
is a copy of one that runs on raven or cheetah, and test/all pins each
copy. The descriptor copy carries the AVB lines the factory image
settled (derekross#350), the same bytes as device/bluejay/device.conf. The
first boot's phone-smoke report goes to derekross#348, the 6a's own thread, and
NOTES.md.

test/all and tests/packages pass on the laptop. The README carries the
SPDX header line raven's and tegu's do.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
robertzas pushed a commit to robertzas/omarchy-mobile that referenced this pull request Nov 14, 2026
…rekross#350)

The notes listed AVB_FOOTERS as open and said no flash until it is
settled, because nobody knew whether a 6a's stock vbmeta chains boot and
vendor_boot. The maintainer asked for the facts from the current factory
zip before anyone flashes (derekross#2).

Read on the laptop with HTTP range requests from the current factory
image, bluejay-cp3a.261005.002.a1 (facts only, no files kept): vbmeta
chains boot (rollback index location 2), vbmeta_system and vbmeta_vendor;
vendor_boot and dtbo are hash descriptors inside vbmeta. boot's footer is
SHA256_RSA4096, rollback index 1791158400, security patch 2026-10-05,
padded to 0x4000000; vendor_boot's and dtbo's footers are unsigned. So
slot B's boot needs a footer and the other two don't, and device.conf
gets AVB_FOOTERS="boot:0x4000000", AVB_ROLLBACK_INDEX=1791158400 and
AVB_SECURITY_PATCH=2026-10-05, stock's own values, never higher.

The same read confirms the rest of the descriptor: build
CP3A.261005.002.A1 (OS_VERSION, OS_PATCH_LEVEL), no init_boot or
vendor_kernel_boot (BOOT_IMAGES), vendor_boot v4 at page size 0x800 with
the two gs101 soc_ids and oriole's bootargs, and a dtbo.img with exactly
the ten entries Google's device trees predicted, 0x20506/0x10000 at
index 9. Which entry a retail 6a boots stays to be read from a phone.

load_device accepts the descriptor, and tests/images-layout's descriptor
and layout checks pass on the laptop; its image steps need Linux sed and
run in CI.

Co-authored-by: Claude Fable 5.1 <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