Skip to content
Closed
Changes from all commits
Commits
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
13 changes: 13 additions & 0 deletions device/husky/NOTES.md
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,19 @@ Sources:
- `P/`: postmarketOS pmaports `device/testing` (2026-10-21).
- `G/`: Google's GPL kernel for this phone, `android-gs-shusky-6.1-android16` ([source.android.com: Build Pixel kernels](https://source.android.com/docs/setup/build/building-pixel-kernels)).

## 2026-10-31: a second phone: Kioxia 256 GB, UFS 4.0

A second retail Pixel 8 Pro, first run through the same read-only steps. `R2/` below means read-only queries of this phone (`adb shell getprop` while booted, `fastboot getvar all`, the fastboot screen), facts only, no serial, IMEI or MACs. The kernel is `omamobile` at `1c5ba1aa2` unless noted.

- **Storage differs from the first phone** [V] (R2/ getvar): `storage-vendor:KIOXIA`, `storage-model:THGJFJT1E45BATPC`, `storage-capacity:256 GB`, `storage-protocol:4.0`, `storage-rev:0300`, `storage-power-mode:lanes=2 gear=4 mode=FAST hs_series=SERIES_B`. The device reports the UFS spec version `0x0400`, `manufacturer_id=0x0198`, `ufs_features=0xB3` in sysfs (`device_descriptor/`), and a `raw_device_capacity` of `0x1DCBC000` 512-byte blocks (499,892,224); `sda` shows 499,490,816 sectors, 401,408 fewer, and that difference was not looked into [I]. The first phone is Samsung 128 GB, UFS 3.1: same link shape, different vendor and spec version.
- **The U2 probe passes on it** [V] (UFS-PLAN U2: `omamobile.ufs-probe` and `omamobile.initramfs` on the command line, no hand-over to Arch, nothing written to UFS). The existing driver brings the link up 0.28 s into boot (`exynos-ufshc 13200000.ufs: Power mode changed to : FAST series_B G_4 L_2`); SCSI sees `KIOXIA THGJFJT1E45BATPC 0300` on three well-known LUNs and four disks, 62 partitions: `sda` has 34 (`persist` is `sda1`, `userdata` is `sda34`), `sdb` and `sdc` are the boot LUNs with 12 each (`bl1`, `pbl`, `bl2`, `dram_train`, `abl`, `bl31`, `tzsw`, `gsa`, `gsa_bl1`, `ldfw`, `dpm`, `gcf`), `sdd` has `devinfo`, `pinfo`, `blenv` and `mfg_data`. The storage guard held: the allowlist was `boot_b init_boot_b vendor_boot_b vendor_kernel_boot_b dtbo_b userdata`, the whole-disk node `/dev/sda` was removed, and only those six partitions were writable. Both watchdogs said "watchdog inactive, reset disabled" and the phone stayed up.
- **dmesg lines seen** [V]: `exynos-clkout … probe … failed with error -22` and the `vdd-hba`/`vccq`/`vccq2` supply lines from `ufshcd_populate_vreg` ("assuming enabled"), also present on the first phone, per its owner: zuma's PMU is not in `clk-exynos-clkout`'s match table, and the device tree does not describe UFS's rails yet.
- **Revision** [I]: the fastboot screen reads `Product revision: husky MP1.0 B1`, which may be the silicon revision. The live `/proc/device-tree/soc_compatible` holds both `B0_FOPLP` and `B1_FOPLP` (product_id 0x9865, major 1, minor 0 and 1, pkg_mode 0), exactly as on the first phone, so it cannot tell them apart (it is the property our own DTB carries); the chip-ID block would, and no zuma chip-ID driver is bound yet.
- **Starting state** [V] (R2/): GrapheneOS `2026102101` (`CP3A.261005.005`, security patch 2026-10-05), bootloader locked with its own verified boot key, current slot b and `slot-unbootable:a:yes`: what a GrapheneOS update leaves behind, as on raven. Bootloader `ripcurrent-17.0-15681527` and baseband `g5300i-260604-260807-B-16036236` (the first phone had `…250605-250630…`), anti-rollback `ap-ar-ns`, `ap-ar-s`, `gsa-bl1-ar` and `gsa-fw-ar` all 3, `hw-revision` `MP1.0`, `ro.boot.dtbo_idx` 8, `boot`, `init_boot`, `vendor_boot`, `vendor_kernel_boot` and `dtbo` the same sizes as the first phone, `userdata` 0x3935215000 bytes. GrapheneOS's own signed `flash-all.sh` for the same version (read, not guessed: it flashes the same bootloader to both slots, forces slot A, radio to slot A, erases `userdata`) put it back on slot A, which then booted and showed `slot-successful:a:yes`, `slot-unbootable:a:no`, with slot B free. It needs an unlocked bootloader, and the bootloader was not locked again.
- **U4 on it** [V]: the signed development channel (39 packages, `omamobile-device-husky` 0.1.1) and Omarchy 4.0.6 stable, `tools/install-phone` with `--bootloader-slots-match ripcurrent-17.0-15681527`: boots, SDDM logs in, Hyprland and the Omarchy shell run, the lock screen shows. First boot hit #188: the clock started at 2026-10-18 and gpg skipped the release key (`key 8632C8075EB94673 was created 8 days in the future`); setting the clock over ssh and `pacman-key --populate omamobile` fixed it. `tools/phone-smoke` passes every required check (boot, storage guard, root, kernel health, USB gadget, session).
- **The display goes off about 43 s into the session; it is the lock screen's idle blank, and Power wakes it** [V]: Omarchy's lock service (`plugins/lock/Service.qml`) has an `idleBlankTimer` that runs `omarchy-brightness-display off` while the session is locked and idle, whose `off` branch is `hyprctl dispatch 'hl.dsp.dpms({ action = "disable" })'`. Caught with kernel audit watch rules on `omarchy-brightness-display` and `hyprctl`: the script ran at uptime 1100.657 s, `hyprctl` 26 ms later, and the DRM connector read `disabled`/`Off` by 1101.4 s (43.2 s after `systemctl restart sddm`). By hand the same dispatch with `action = "enable"` brings the display back at once and it stays on. On the test tag below, pressing Power wakes the blanked screen, and so does a tap on it (both reported by the phone's owner; the other phone's owner reports only Power on older shell packages), and the key reaches the system (`KEY_POWER`); the lock screen then took a swipe up and the on-screen PIN pad by touch and unlocked (`unlocked` in the lock service's log). The `omamobile` head (`1c5ba1aa2`) has no `gpio-keys` node in `zuma-husky.dts` (the test tag has one, from #148), so on the head kernel no key can wake a blanked screen and there is no touch either [V: the device trees]; that is probably why the first boots looked like a dead panel [I]. Hyprland's log has `CDRMRenderer(drm): Can't create renderer, no matching devices found` and `drm: Failed to update renderer state … on applyCommit` from start-up on (the DRM device is `simple-framebuffer`, `/dev/dri` has only `card0`); the first DPMS enable after a disable worked regardless, so these lines are not what blanks it. Why the blank fires at about 43 s and not 5 s after the lock is not known [I].
- **Test tag `husky-test-20261031-c6`** (greenbrettmichael/linux-omarchy-mobile, kernel `7.3.0-rc6-omamobile-push1-gc7a0b2ef2f66`, the repository's husky config unchanged, then `modules-push` and a reboot) [V]: `dmesg | grep salixp` shows `sysinfo fw 0083 cxver ffff`, `afe cfg 03 cx 03 panel 03`, `sense echo yes events 2` and an input device `SALIXP Touchscreen`; the raw events of two rounds of hand taps (top-left, top-right, bottom-left, bottom-right, centre) decode to corners at x 57 to 1255 of 0..1343 and y 37 to 2962 of 0..2991, axes not swapped or flipped; `gpio-keys` gives `KEY_VOLUMEUP`, `KEY_VOLUMEDOWN` and `KEY_POWER`. If the modules are brought up by hand before a reboot, `acpm-protocol` can probe before `exynos-mailbox` (`no mailbox channel (-EPROBE_DEFER), skipped`, `16 channels, 0 usable`) and `salixp` then fails with `-ENODEV: enabling vdd`; after a reboot with the modules installed it says `16 channels, 16 usable` and the touch driver binds at about 3.7 s.

## 2026-10-21: stage 0, part 4: root capture, backup, and how ABL picks the DTB

Root for the capture: the stock `init_boot` from the CP3A.261005.005 OTA, extracted with `payload-dumper-go` 2.1.0 (hashes checked against the signed payload) and patched with Magisk v31.0, the first release with Android 17 support. It was flashed to **`init_boot_b` only**, with slot B booted. Slot A was not written. The capture and backup were read-only (`su -c cat`), into `~/android/reference/husky-root-*` (never the repository).
Expand Down