Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
18 commits
Select commit Hold shift + click to select a range
4646d1c
device-lynx: record UFS bring-up and the storage guard
latentoperator Oct 24, 2026
5c89629
device-lynx 0.1.0: device package and Arch root install
latentoperator Oct 24, 2026
c1851ef
device-lynx 0.2.0: mark the booted slot successful
latentoperator Oct 24, 2026
289ae06
device-lynx: record Wi-Fi bring-up so far
latentoperator Oct 24, 2026
cbb0e2a
device-lynx: packaged descriptor follows device/lynx
latentoperator Oct 24, 2026
99714b4
device-lynx: document recovery after installing Arch
latentoperator Oct 26, 2026
8be6c99
device-lynx: park Wi-Fi at the boot ROM's authentication stage
latentoperator Oct 27, 2026
b17370c
docs: DEVICES, the Pixel 7a row after its device package
latentoperator Oct 27, 2026
bd48fe0
device-lynx: WIFI-PLAN, the post-stall column and the cold-read caveat
latentoperator Oct 27, 2026
0e6b3b4
device-lynx: WIFI-PLAN, stock loads on the second WLAN_EN cycle
latentoperator Oct 27, 2026
d50af30
device-lynx: WIFI-PLAN, the post-stall column and a fourth difference
latentoperator Oct 27, 2026
04688d1
Merge origin/main into lynx-device
latentoperator Oct 27, 2026
aa06c3d
device-lynx: the Pixel 7 Pro's sleep drop-in
latentoperator Oct 27, 2026
877c298
tests: storage-guard checks lynx's allowlist
latentoperator Oct 27, 2026
3eb9536
docs: lynx after kernel #91
latentoperator Oct 27, 2026
a74e329
device-lynx: WIFI-PLAN after step 6
latentoperator Oct 27, 2026
f7c70bf
device-lynx: devinfo against fastboot
latentoperator Oct 27, 2026
8e9e967
device-lynx: WIFI-PLAN, the two chip revisions
latentoperator Oct 27, 2026
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
117 changes: 82 additions & 35 deletions device/lynx/BRINGUP-M1.md
Original file line number Diff line number Diff line change
@@ -1,22 +1,28 @@
<!-- SPDX-License-Identifier: MIT -->
# Pixel 7a (lynx) first light

Status (2026-10-24, this first-light/display branch): slot B boots and SSH
comes up at `root@172.16.42.1` over the USB NCM gadget. Both cluster
watchdogs are stopped at probe. With kernel #44's framebuffer reservation
and this branch's DECON0 trigger unmask, simpledrm/fbcon shows the boot log
and a shell prompt at 67x75. UFS remains disabled on this branch; the
checks below describe the initramfs boot, before installing Arch on
`userdata`. Kernel `7.3.0-rc6-omamobile-push1-g3081ac63fcc4`.
Last verified state (2026-10-27, after Arch was reinstalled from the stock
afternoon): slot B is active and marked successful by
`omamobile-device-lynx` 0.2.0. Arch runs from `userdata`, USB SSH works at
`root@172.16.42.1`, and the on-screen console works through simpledrm/fbcon.
Kernel: `7.3.0-rc6-omamobile-push1-g38b5a04d67e0`. `/sys/power/mem_sleep`
reads `s2idle [deep]`; the package's sleep drop-in keeps suspend on s2idle.
Wi-Fi is parked ([WIFI-PLAN.md](WIFI-PLAN.md)). Recheck the live state
before further hardware work.

**Stock Android on slot A is no longer a working recovery boot because
Arch owns `userdata`. Use the physical bootloader buttons for recovery.**
The first-light sections below describe earlier initramfs boots; their
original automatic fallback assumptions no longer apply to this phone.

## Facts

The stock capture and what each boot showed are in [NOTES.md](NOTES.md).
Tracking issue: #36.

## What booted
## What booted at first light (historical)

The board file (linux-omarchy-mobile, `gs201-lynx.dts`) enables the USB
The original board file (linux-omarchy-mobile, `gs201-lynx.dts`) enabled the USB
controller and relies on the SoC tree for the cluster watchdogs. UFS, the
panel, touch, the GPU and the radios stay disabled. It carries
`/soc_compatible/B0_IPOP` instead of `/soc_id`, because lynx's ABL matches
Expand All @@ -32,10 +38,11 @@ It differs from cheetah's in two ways the bootloader forces:
lynx's unlocked ABL never shows the orange warning and never boots slot B
(black screen, nothing on USB, three silent retries, then slot A).

The shared init writes no display register (`SIMPLEFB_TRIG_CON` is unset)
and reads no display DMA register (`BRINGUP_PROBE` is unset).
At first light, `SIMPLEFB_TRIG_CON` and `BRINGUP_PROBE` were unset. The
current descriptor enables the board-specific trigger write described in
the console section below; `BRINGUP_PROBE` remains unset.

## First boot
## First boot (historical, before Arch owned userdata)

Flash only slot B, from fastboot: `DEVICE=lynx bin/omamobile images`, then
`DEVICE=lynx bin/omamobile flash`. Slot A stays stock Android. Confirm the
Expand All @@ -55,34 +62,34 @@ Expect, in this order:
`dmesg | grep watchdog` shows both `10060000.watchdog` and
`10070000.watchdog` as `watchdog inactive, reset disabled`.

Stop if no USB device appears within a minute, if the phone resets by itself
at about 60 s (a watchdog was left running), or if it does not come back to
slot A by itself after three failed boots. Do not flash again until you know
why.
Stop if no USB device appears within a minute or the phone resets by itself
at about 60 s (a watchdog was left running). Do not flash again until the
cause is understood. On the current Arch installation, do not wait for a
working Android fallback; follow the physical recovery procedure below.

## How to get back

Slot B has three retries after `fastboot --set-active=b`, and every boot
that is not marked successful uses one. After the third, ABL boots slot A on
its own. While stock is on slot A that is the fallback; just let it boot.
Slot B was marked successful after Arch installation. The old fallback via
three unsuccessful boots relied on Android being able to boot slot A.
With Arch on `userdata`, Android stops at recovery instead. Likewise,
`bin/omamobile fastboot` cannot complete its crash-to-Android-to-adb path.
Do not use either route as the recovery plan for this installation.

- **Force restart:** hold Power for about 30 s. The phone restarts into the
active slot. Both watchdogs are stopped at probe, so a hang that does not
panic only ends this way.
- **Fastboot by hand:** hold Volume Down and Power together, from off or
through a force restart, until the bootloader menu appears. Then
`fastboot --set-active=a` (or `b`). Never pick "Factory data reset" in
that menu or in Android recovery: it wipes `userdata`.
- **Fastboot without buttons**, from our Linux: `DEVICE=lynx bin/omamobile
fastboot`. It crashes our kernel, ABL falls back to slot A, and the tool
runs `adb reboot bootloader` from Android. Holding Volume Down during a
`reboot-to` is too late by the orange warning.
- **Power off:** `poweroff` from our Linux stops the SoC but leaves the
panel lit and the phone off USB (no PMIC driver). Use the bootloader
menu's "Power off" instead.

Nothing here writes outside slot B's five boot images. Stock's `init_boot_a`
stays as it was, so slot A can go back to plain stock with the factory image.
through a force restart, until the bootloader menu appears (about 20 s).
Never choose "Factory data reset" in that menu or Android recovery: it
erases the Arch root on `userdata`.
- **Force restart:** hold Power for about 30 s if the phone is unresponsive.
Use the button recovery procedure with the human present; do not keep
sending commands to a phone that has stopped answering. An armed capture
watchdog may recover a test hang, but recovery must be confirmed.
- **Power off:** the earlier `poweroff` test stopped the SoC but left the
panel lit and USB absent. Use the bootloader menu's "Power off".

Only slot B's five boot images and `userdata` belong to this work. Keep the
storage guard intact. Slot A's untouched boot images do not make Android
bootable with the current userdata. Reboots and flashes still require the
human's explicit approval and physical recovery readiness.

## The on-screen console (2026-10-24)

Expand All @@ -102,3 +109,43 @@ Stop if the panel goes black and stays black, shows a torn or shifted
picture, or if the phone resets right after the write. Roll back by
building and flashing slot B from a descriptor without `SIMPLEFB_TRIG_CON`;
the phone then boots as at first light, with the bootloader logo on screen.

## devinfo against fastboot (2026-10-27, `g38b5a04d67e0`, device-lynx 0.2.0)

For Derek's #88 question: lynx's bootloader is a different build from the
Pixel 7 Pro's (it needs AVB footers), so the slot bytes are checked against
what the bootloader itself reports. devinfo is `/dev/sdd1`, version 3.15 in
every reading. Only bytes 0x30-0x37 are shown (slot a 0x30-0x33, slot b
0x34-0x37), from `omamobile-bootctl dump`, next to `fastboot getvar
slot-successful`, `slot-unbootable` and `slot-retry-count` for the same
state:

| State | a: bytes | a: fastboot | b: bytes | b: fastboot |
| --- | --- | --- | --- | --- |
| B marked by `omamobile-boot-success` | `00 0a 00 00` | successful yes, unbootable no, retry 0 | `02 06 00 00` | successful yes, unbootable no, retry 2 |
| `fastboot --set-active=b`, before booting | (unchanged) | successful yes, unbootable no, retry 0 | (not readable in fastboot) | successful no, unbootable no, retry 3 |
| after one unmarked boot (`omamobile-boot-success` masked) | `00 0a 00 00` | successful yes, unbootable no, retry 0 | `02 04 00 00` | successful no, unbootable no, retry 2 |

So byte 0 is the retry count and byte 1 holds the flags, as on the
Pixel 7 Pro and Pixel 6 Pro: 0x04 active, 0x02 successful, 0x08 what the
tool calls fastboot-ok (set on slot a only here); no reading had
unbootable set. Afterwards the unit was unmasked, `fastboot
--set-active=b` and one boot left slot b at `02 04` with the bootloader's
retry spent, and the marker wrote 0x35 `04 -> 06`: `02 06 00 00` again.

A forced restart (Volume Down + Power, released too early, so it booted
slot B rather than stopping in the menu) did not spend a retry on the
marked slot: `02 06 00 00` before and after, retry 2 in fastboot.

**The first marks were made by hand, outside the storage guard's rules.**
On 2026-10-24, over SSH from the running Arch system, before the package
shipped the marker: `/dev/sdd` and `/dev/sdd1` were set writable with
`blockdev --setrw`, a trap set both back read-only on exit, and the
Pixel 7 Pro package's `omamobile-bootctl mark-successful /dev/sdd1 b
--i-understand` (copied to `/tmp`) changed byte 0x35 `04 -> 06` and
verified the block. On 2026-10-27, after the reinstall from the stock
afternoon, the root image still carried 0.1.0: 0.2.0's
`omamobile-boot-success` and `omamobile-bootctl` were copied in over SSH
and the marker was run once by hand, 0x35 `04 -> 06`, verified (slot b
`02 04 00 00` before, `02 06 00 00` after). Since then only the packaged
`omamobile-boot-success` writes devinfo.
Loading