You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Pixel 8a (akita / zuma, Tensor G3) port: offering hardware and help #1
Hi Derek, great work on Omandroid. I have a spare Pixel 8a (akita, Tensor G3 / zuma) that I'm happy to unlock, wipe and dedicate to testing, and I'd like to help bring up zuma support.
What I plan to do, following CONTRIBUTING.md:
Generalize bin/omandroid images and initramfs/init so device values (DTB name, soc_id, dtbo board ID, partition guard list, display init) come from a per-device config instead of being hard-coded for cheetah (per §5.3).
Capture akita board facts (bin/omandroid capture, fastboot getvar all, factory image) into device/akita/NOTES.md, with no serials or IMEI.
Start zuma groundwork in linux-omandroid, using gs201 as the pattern and Google's android-gs-akita-6.1 branch as reference: the SoC dtsi, clocks and pinctrl, then a minimal zuma-akita.dts aimed at first light (USB SSH).
Questions:
Is anyone already working on zuma, or do you have a preferred structure for multi-SoC support?
Do you want the device-config refactor as its own PR before any akita files land?
Thanks for offering hardware and time. A dedicated Pixel 8a is really useful.
Is anyone on zuma? No, you'd be the first. I've added the 8a to the Wanted table pointing here.
Device-config refactor: that part is already done, so no separate PR is needed. As of today, bin/omandroid reads every device value (DTB, soc_id list, dtbo board ID/rev, boot image list, partition guard allowlist, and so on) from device/<codename>/device.conf, and DEVICE=<codename> bin/omandroid … selects the device. Cheetah's images came out byte-identical. See device/cheetah/device.conf and the design in docs/RELEASES.md. The initramfs init still has cheetah's DECON trigger write and probe() reads, which must not run on another phone; RELEASES.md §3 lists them. A new device's copy of init drops them, as CONTRIBUTING §5.3 now says.
Suggested order for akita:
Capture first (§5.1): bin/omandroid capture akita-stock plus fastboot getvar all, with facts only in device/akita/NOTES.md (no serials, IMEI or MACs). Please confirm early which boot partitions akita has (init_boot, vendor_kernel_boot, dtbo), since the descriptor's image list depends on it.
device/akita/device.conf plus notes, as your first PR to this repo.
Zuma kernel groundwork in linux-omandroid: fork it, branch off omandroid, one subsystem per branch, and open PRs against omandroid. The SoC dtsi and board DTS go in arch/arm64/boot/dts/exynos/google/ (zuma.dtsi, zuma-akita.dts), in upstream order: bindings, then driver, then DTS. Google's GPL android-gs-akita-6.1 tree is fine as a reference: keep its copyright on anything imported, and port it to upstream APIs.
A heads-up on expectations: zuma is a new SoC for us. gs201 had a mainline gs101 to borrow from, and zuma has much less, so first light (USB SSH) is the realistic first milestone, and it will take longer than the Pixel 6 port in #2. The Pixel 6 is our preferred second device for proving the multi-device tooling, so zuma doesn't need to carry that.
Please keep CONTRIBUTING §3 and AGENTS.md's safety rules close: slot B and userdata only, never vbmeta, devinfo, radio or persist, and back up efs/persist/modem* off-device before any radio or PMIC work. Post what you find here, and I'm happy to review.
Reacted by William K⚡Santiago🔑☢️, Evan Eugene Pincham and ZapArcade
Heads-up: the project is now called Omarchy Mobile.
Omandroid was never Android, and the other Omarchy Mobile effort is merging into this one, so the project now carries that name. Everything moved today:
Kernel branches:omamobile (curated, upstream-ready) and omamobile-int (what the Pixel 7 Pro runs, as a snapshot). The old omandroid and omandroid-int branches stay a short while for anyone mid-clone, then go away.
Inside the code, the short name is omamobile, like Omarchy's omacalc and omawrite: bin/omamobile, omamobile-* packages and commands, the [omamobile] package repository and omamobile-keyring, and device/<codename>/device.conf as before. The release suffix is -omamobile.
CONTRIBUTING, DEVICES and the release design are updated. Nothing else about the process changed: fork linux-omarchy-mobile and open PRs against omamobile; device notes and device.conf go to this repo.
Thanks for bearing with the churn. Better now than after a release.
Reacted by William K⚡Santiago🔑☢️ and Evan Eugene Pincham
I have a retail Pixel 8 Pro (husky), also zuma, unlocked and dedicated to this, and I'd like to work on zuma with you, @VirtualMachinist, if you're still on it. I've finished stage 0 on husky. The facts are in device/husky/NOTES.md (#23), with no serials, IMEI or MACs. The findings that matter for every zuma phone:
1. zuma's ABL doesn't pick the base DTB by /soc_id. The four base DTBs in stock vendor_kernel_boot (there are none in vendor_boot, and no dt_table header) carry no /soc_id at all. Each has a /soc_compatible node with one child per silicon revision and package:
DTB
children
product_id
major
minor
pkg_mode
0
A0_FOPLP
0x9865
0
0
0
1
A0_IPOP
0x9865
0
0
1
2
B0_FOPLP, B1_FOPLP
0x9865
1
0 / 1
0
3
B0_IPOP
0x9865
1
0
1
A retail husky gets DTB 2 (the live DT's /soc_compatible holds B0 and B1 FOPLP). So SOC_IDS and the soc_id rewrite in bin/omamobile images don't fit zuma. The descriptor in the PR leaves SOC_IDS out on purpose, so the tools refuse it, and proposes:
images would then require that the DTB is built with a /soc_compatible node, write one child per entry, and refuse a descriptor that has both keys or neither. Would you take that as a separate tooling PR, or would you rather shape it yourself? I can write it with a test in tests/images-layout.
2. Board dtbo: husky MP1.0 boots entry 8 of 19, board_id 0x40206, board_rev 0x10000 (androidboot.dtbo_idx = 8). The other zuma phones will have their own; akita's is worth checking the same way.
six 2 GiB RAM banks and 30 no-map reservations to copy;
uart0 = /uart@10870000 (samsung,exynos-uart).
4. Slot hygiene for anyone starting from a used phone: mine arrived with stock active on slot B, and the bootloader only reports the current slot's version. Two sideloads of the same full OTA (Google's own advice for the Pixel 8 family) put the identical build and bootloader on both slots, then --set-active=a. Root for the capture was a Magisk-patched copy of stock init_boot flashed to init_boot_b only, then put back, so slot A was never written.
Next: zuma kernel groundwork in a fork of linux-omarchy-mobile, branch omamobile-zuma-husky-dts off omamobile. It holds a minimal zuma.dtsi (CPUs, GIC, timer, the inherited framebuffer, UART) and zuma-husky.dts aimed at first light, with Google's android-gs-shusky-6.1 as the reference and gs101/gs201 as the pattern. If you've started on clocks or pinctrl for akita, tell me and I'll build on that instead of duplicating it.
Pixel 8 Pro (husky, Tensor G3 / zuma): mainline boots, and USB works.
On slot B, the phone boots our mainline-based kernel into the Omarchy initramfs. The host sees the CDC NCM + ACM gadget at high speed, and ssh root@172.16.42.1 works. All nine CPUs are online and 11.2 GiB of RAM is visible. Stock Android stays untouched on slot A. For now each boot lasts about 80 s: ABL leaves the CPU watchdogs running, and zuma has no watchdog driver yet. That's next.
What it took, beyond the minimal zuma DTB:
A base DTB laid out the way ABL expects:/soc_compatible rather than /soc_id, the ECT location in /ect, and a /memory node.
AVB hash footers on boot and init_boot. zuma's vbmeta chains them, and ABL rejects them without footers even when unlocked. Today I add these by hand with avbtool, so bin/omamobile images should learn to do it (an AVB_FOOTERS setting in device.conf?). I can send that with the /soc_compatible matching as one tooling PR, if that suits you.
The eUSB2 PHY. On a normal boot ABL leaves it off, so a bare DWC3 can't finish its soft reset. Mainline's Exynos 2200 drivers fit with small additions: the PMU, the control block and the eUSB2 PHY.
zuma keeps no ramoops across a warm reset, so the debugging was done with a stopwatch: a hang marker in the init crashes at a time that encodes how far boot got. Details are in device/husky/BRINGUP-M1.md (#23) and in derekross/linux-omarchy-mobile#13.
Next: the watchdog, then a framebuffer for a screen console (M1.4).
Congratulations, @greenbrettmichael: a mainline zuma kernel with USB and SSH up on slot B, and stock untouched on slot A, is a big milestone. The stopwatch hang marker is a neat trick for a SoC with no ramoops.
On the tooling: yes, one tooling PR for both /soc_compatible matching and the AVB footers suits me, separate from #23, as two commits (one per feature) with tests in tests/images-layout:
SOC_COMPATIBLE: a descriptor has exactly one of SOC_IDS or SOC_COMPATIBLE, the DTB must already have a /soc_compatible node, and there is one child per entry. Please keep the format general enough for zumapro (device-tegu: record Pixel 9a first light and staged descriptor #24 needs the same thing).
AVB_FOOTERS (or similar) in device.conf, naming the images that get a hash footer, with the partition sizes taken from the descriptor rather than guessed. avbtool becomes a build dependency only for devices that set it.
Cheetah's images must stay byte-identical with neither key set, and the test should check that.
My review of #23 is up. Since then I've also made tools/install-phone refuse a descriptor without SOC_IDS (and images built for another device), so once SOC_COMPATIBLE lands, the installer needs to accept it too. Please include that in the same PR.
On the kernel side, keep the watchdog, PMU, eUSB2 PHY and DTS work in separate subsystem branches, as in derekross/linux-omarchy-mobile#13. @VirtualMachinist, if you've started on akita clocks or pinctrl, please say so here so the two of you don't duplicate it.
Since my last comment, the watchdog driver has removed the 80 s reset. ABL's splash at 0xfac00000 now gives a console on the panel. And reboot-to bootloader works once the kernel resets warm, because a cold PSCI reset clears the reboot reason.
A correction to my milestone comment above: of the base-DTB items I listed, only /soc_compatible is shown to be required (test 5). The ECT location and /memory match stock and are in every image that booted, but test 2 showed they weren't why ABL refused our images; the missing AVB footers were. BRINGUP-M1.md now says so.
The tooling PR is next, as you describe:
two commits, SOC_COMPATIBLE and AVB_FOOTERS, with sizes from the descriptor;
tests/images-layout cases, including cheetah's images staying byte-identical with neither key set;
tools/install-phone accepting SOC_COMPATIBLE, on top of your change.
I'll keep the SOC_COMPATIBLE format general enough for zumapro (#24). The init's hang marker will go in a separate small PR, if you want it at all.
awesome stuff @greenbrettmichael -- my apologies, but I fear I will have to pull away from this mission, as I just got an offer for a (basically) free upgrade from pixel 8a to pixel 11 and from iphone 14 plus to 18 pro through my mobile carrier, and it requires me to trade them in.
sorry @derekross ! I do have a couple of old iphones that I can dedicate to the cause, as well as a pixel 11 that should be here within the next couple of days, but my pixel 8 is getting shipped away.
No apology needed, @VirtualMachinist, that's a good deal to take. Thanks for the help while you had the 8a. If the Pixel 11 ends up free for experiments, #26 tracks Tensor G6, and stock captures from it (bin/omamobile capture, nothing flashed) would already help there.
@greenbrettmichael, akita is zuma like your husky, so the clocks, pinctrl, PMU, watchdog and eUSB2 work all carries over if someone else picks the 8a up later.
What it took: ABL hands UFS over powered down. VCC's GPIO and the reference-clock pin are both GPIO outputs driven low, so mainline needs the pin controller.
Arch:tools/install-phone installed a test root image, and the init handed over to Arch on userdata. The guard's flags hold under Arch, reboot-to bootloader works from there, and a second boot comes back cleanly.
The test image was Arch Linux ARM's base, its signature checked through tools/lib/alarm.sh, plus a new omamobile-device-husky package and the kernel's modules. The package is felix's minus the lid drop-in. It's on my husky-device-package branch with husky's UFS-PLAN (U1-U3), and I'll open the PR once #23 merges, since it builds on it.
The question: a full Omarchy image (U4).tools/build-rootfs installs our packages from out/repo, and the build container trusts only the project key in omamobile-keyring. As a contributor I can't sign that repository, and I'd rather not route around the trust model with a local key. How would you like to handle it? I see two options:
Once the package PR is in, you build and sign a repository with omamobile-device-husky in it, and I build the root image from that.
A documented contributor path for a development key that only ever goes onto the contributor's own phone.
Thanks, @greenbrettmichael. This is great progress: UFS behind the guard at HS-G4 with stock's partition names, and Arch booting from userdata, is the milestone that makes husky a real port. And thank you for not routing around the trust model.
Let's do option 1. Once #23 and your device-package PR are in, I'll build and sign a repository with omamobile-device-husky in it, and publish it so you and other contributors can build root images from it without my key. I'll post the URL here when it's up.
A contributor development key (your option 2) is worth having later, since every port hits this before its package merges, but it needs designing so a development image can never be mistaken for a release one. I'll come back to it separately.
Thanks for being straight about the levels. Stay at digital 737 / analog 10 or lower until husky has its own ceilings, and keep speaker_protection=1 off the boot unit. If either amp ever reports a fault or sounds wrong after those sessions, say so here.
I'll review #449, linux-omarchy-mobile#203 and #204 on their PRs.
On the System MMU exports: two modules can't export the same symbol, so don't duplicate them. Either make the existing gs201 driver handle zuma (a compatible match plus per-SoC data), which is what I'd prefer if the register layout is close, or, if it really differs, give the zuma file its own prefixed names and have the AoC core pick its ops through the match data. Send the first if it fits, and show in the PR that the gs201 objects don't change.
Hi all. I have a Pixel 8 Pro I can dedicate to husky as a third test unit.
Read-only so far, no serials:
husky, MP1.0, SKU G1MNW, Samsung 128 GB UFS, dtbo_idx 8 (SKU and UFS vendor and size match Brett's unit)
GrapheneOS 2026110201 (the security preview build of 2026110200) on CP3A.261005.005, security patch 2026-11-01, newer than device.conf's 2026-10-05 AVB footer
Bootloader ripcurrent-17.0-15681527 (getvar shows the current slot's only)
Locked getvar: current-slot a, slot-unbootable a:no b:yes (Virtual A/B after the 11-06 update merged, I think), ap-ar-ns/ap-ar-s/gsa-bl1-ar/gsa-fw-ar 3/3/3/3, baseband g5300i-260604-260807-B-16036236, storage KLUDG4UHGC-B0E1 UFS 3.1 rev 3203
Carrier-unlocked, OEM unlocking already allowed in Developer options, no SIM in it
Nothing is unlocked or flashed, and won't be until my written steps check out against CONTRIBUTING §3. Two questions first, @derekross:
Slot A. Your 10-27 note is stock on both slots via two full-OTA sideloads. To get stock onto slot A in the first place, I'd run only fastboot -w update with Google's CP3A.261005.005 image, skipping flash-all's bootloader and radio lines, since both versions already match what's installed. That's an unsuffixed flash of slot A's OS images, and a lower patch level (2026-10-05) than GrapheneOS's 2026-11-01 on the same bootloader. OK, or would you rather I use GrapheneOS's flash-all, as Lef did?
Backup. ACPM and the S2MPG14 are built in for USB, so every boot of our kernel touches the PMIC. Is a U2-style omamobile.initramfs boot the right place for the first read-only backup, before Arch loads anything? I've built one by adding omamobile.initramfs to the DTB's /chosen/bootargs with fdtput, in its own KERNEL_OUT (ABL ignores the vendor_boot cmdline); only dtb.img and vendor_kernel_boot.img differ from the install set. Is there a preferred way?
I'll add the unit to device/husky/NOTES.md as a third phone, facts only.
At the lock screen, what helps the next kernel round more from a second Samsung G1MNW unit: #199/#200 on Lef's husky-test-20261106-pr199-200, or the display and GPU stack (Brett's husky-test-20261101-e1, or the rebased PRs when he's back)? Firmware will come from my own vendor image, with a full tools/phone-smoke report per run.
Host: Ubuntu 26.04 in WSL2 with usbipd. kernel-2026.11.01, both image sets and the root image build there (Omarchy edge, because of #433). Fastboot through usbipd works (WSL's getvar matches Windows'); the USB gadget network through it is still untested. I use Claude Code on the PC; every unlock, flash and reboot is my call. @Somnius@greenbrettmichael
I'm back. Thank you all for a very productive week on the 8 Pro. @Somnius, Bluetooth, haptics, the torch and then the AoC and sound in a few days is great work, and so is the packet-ID fix on the Wi-Fi.
@gitColinX welcome, and thanks for the careful read-only start. The slot A and backup questions are Derek's call. On what to test: #199 and #200 are merged, so on the kernel side the open husky work is the display and GPU stack. I rebased both today onto omamobile: linux-omarchy-mobile#160 (gs-decon, DSIM0 and the HK3 at 120 Hz) and #158 (which carries #157: the GPU on panthor, CPU DVFS and the thermal zones). A boot of those two on a second Samsung G1MNW unit, with a tools/phone-smoke report and uname -r, would help most. They conflict only in zuma.dtsi, where both add nodes at the same place; keep both, in address order. If you'd rather wait, #160 lands first and #158 gets one more rebase after it.
Thanks, @greenbrettmichael. Yes, I'll boot the display and GPU stack on this G1MNW unit as soon as it's running from slot B, and post the tools/phone-smoke report and uname -r here.
I see #160 merged today, so I'll build omamobile with #158 on top once it's rebased. If it isn't by then, I'll resolve zuma.dtsi the way you said, both sets of nodes in address order.
Not installed yet: I'm waiting on Derek's answers about slot A and the backup. The kernel, both image sets and the root image are built on kernel-2026.11.08 in the meantime.
@greenbrettmichael, the boot you asked for, on my G1MNW unit (Samsung 128 GB UFS 3.1): your husky-test-20261109-t1 tag (54e10dccc, omamobile with #160 and #158), built here with the pinned clang, modules installed with modules-push, slot B written with omamobile flash. The root image is a fresh install from main (2026.11.09, omamobile-device-husky 0.1.8), first booted on kernel-2026.11.08 an hour earlier.
uname -r:7.3.0-rc6-omamobile-push1-g54e10dcccd6b
Display (docs/review: lessons from review round 29 #160):card2-DSI-1 at 1344x2992 through gs-decon. 3166 gs-decon interrupts at 100 s, no DSI, underrun or flip-timeout lines. The HK3's refused brightness write (sending dcs data 51 05 49 failed: -1) is there too, as on yours.
GPU: panthor's render node; devfreq idles at 150 MHz.
Thermal: big, mid, little and gpu read 53-54 °C at 100 s with the screen on and the phone charging from a PC port, warmer than your 48 °C. I'll read them again after a longer idle.
Battery: only max77759 and max17042_battery load, as with the 0.1.8 blocklist. The gauge reads 96 % at 36 °C.
No failed units. Wi-Fi and sound are skipped because I haven't run phone-provision yet, so there's no firmware on the root.
tools/phone-smoke on husky-test-20261109-t1
Check
Result
Detail
boot (required)
ok
kernel 7.3.0-rc6-omamobile-push1-g54e10dcccd6b, slot b, up 18s
storage guard (required)
ok
62 partitions; only userdata and slot B's boot images writable
root (required)
ok
ext4 on userdata, 98G free
kernel health (required)
ok
no oops or BUG; 0 WARN, 23 error lines in dmesg (13 known, 10 new)
new dmesg errors
not ok
Bluetooth: N: BCM: '…' x2; exynos-clkout exynos-clkout: probe with driver exynos-clkout failed with error N x1; exynos-clkout exynos-clkout: cannot match parent device x1
systemd units
ok
none failed
USB gadget (required)
ok
bound to 11210000.usb
display
ok
card2-DSI-1 connected, 1344x2992
GPU
ok
render node on panthor
session (required)
ok
Hyprland and the Omarchy shell running
lock
ok
PAM file present, the shell can lock (passwordPam true, locked true)
touch
ok
SALIXP Touchscreen
buttons
ok
gpio-keys
Wi-Fi
skip
no wireless interface
Bluetooth
ok
hci0, powered true
battery
ok
96% Charging, 36 °C
sound cards
skip
none
sensors
skip
no sensor proxy
cameras
skip
no libcamera tools
thermal
ok
hottest zone 63 °C
The new dmesg lines are the BCM4398 ID lines and the exynos-clkout probe that #209 fixes; none come from these PRs.
How I got here, for the notes: GrapheneOS to Google's CP3A.261005.005 on slot A with fastboot -w update (no bootloader or radio write), then the full OTA sideloaded twice, and the read-only NV backup from the guarded initramfs before Arch ever ran. One thing that may help other Windows hosts: through Windows' own UsbNcm driver, the initramfs link stalled for 30 s at a time under nv-backup until the phone sent something (ARP stayed Incomplete). A ping -i 0.2 from the phone to the host, started on the ACM console, kept it moving, and the backup then ran clean. I'll open an issue with the details.
Now that Anthropic shared a reset that we can use until Dec 10.. I'll use the ultra plan to fix all I can with Pixel 8 pro 😈
“ψ(`∇´)ψ
Haha I feel the same, want to help push this Pixel 8 Pro as much forward as I can over the finish line and try using it as a daily driver phone finally, so help us Claude!
Now that Anthropic shared a reset that we can use until Dec 10.. I'll use the ultra plan to fix all I can with Pixel 8 pro 😈
“ψ(`∇´)ψ
Haha I feel the same, want to help push this Pixel 8 Pro as much forward as I can over the finish line and try using it as a daily driver phone finally, so help us Claude!
Haaaa fellow brother in arms @gitColinX 😜 (also researched a portable monitor, I got a wireless keyb+mouse [bt] that is why I fixed BT first 🫡)
Pixel 8a display (akita M6): a design question before any code.
gs-decon's zuma modes are the HK3's (1344×2992), and its probe refuses any other scanout size ("unexpected scanout size"). So on the 8a it stops safely at probe. The 8a has an ak3b at 1080×2400, on the same DECON0 → DSIM0 path, with DSC.
1. Modes. To support both zuma phones, would you rather gs-decon:
(a) carried a list of mode sets per SoC, and picked the one matching the size the bootloader left; or
(b) took the modes from the panel node or the panel driver?
2. The panel driver. For the 8a's panel, would you rather have:
a new panel-google-ak3b.c, with an ID check for the production module 0x0080801E, sending only sleep, 60/120 Hz and brightness from Google's GPL driver; or
the HK3 driver generalised?
Nothing has run on the 8a yet. The next step there is a read-only boot: gs-decon at stage 0, and gs-dsim-cmd with allow_tx off, to log what the bootloader leaves.
Activity
DEVICES, CONTRIBUTING: Pixel 6 and 8a ports in progress; the fork-and…
derekross commentedon Oct 13, 2026
Thanks for offering hardware and time. A dedicated Pixel 8a is really useful.
Is anyone on zuma? No, you'd be the first. I've added the 8a to the Wanted table pointing here.
Device-config refactor: that part is already done, so no separate PR is needed. As of today,
bin/omandroidreads every device value (DTB,soc_idlist, dtbo board ID/rev, boot image list, partition guard allowlist, and so on) fromdevice/<codename>/device.conf, andDEVICE=<codename> bin/omandroid …selects the device. Cheetah's images came out byte-identical. Seedevice/cheetah/device.confand the design indocs/RELEASES.md. The initramfsinitstill has cheetah's DECON trigger write andprobe()reads, which must not run on another phone; RELEASES.md §3 lists them. A new device's copy ofinitdrops them, as CONTRIBUTING §5.3 now says.Suggested order for akita:
bin/omandroid capture akita-stockplusfastboot getvar all, with facts only indevice/akita/NOTES.md(no serials, IMEI or MACs). Please confirm early which boot partitions akita has (init_boot,vendor_kernel_boot,dtbo), since the descriptor's image list depends on it.device/akita/device.confplus notes, as your first PR to this repo.omandroid, one subsystem per branch, and open PRs againstomandroid. The SoC dtsi and board DTS go inarch/arm64/boot/dts/exynos/google/(zuma.dtsi,zuma-akita.dts), in upstream order: bindings, then driver, then DTS. Google's GPLandroid-gs-akita-6.1tree is fine as a reference: keep its copyright on anything imported, and port it to upstream APIs.A heads-up on expectations: zuma is a new SoC for us. gs201 had a mainline gs101 to borrow from, and zuma has much less, so first light (USB SSH) is the realistic first milestone, and it will take longer than the Pixel 6 port in #2. The Pixel 6 is our preferred second device for proving the multi-device tooling, so zuma doesn't need to carry that.
Please keep CONTRIBUTING §3 and AGENTS.md's safety rules close: slot B and
userdataonly, nevervbmeta,devinfo, radio orpersist, and back upefs/persist/modem*off-device before any radio or PMIC work. Post what you find here, and I'm happy to review.derekross commentedon Oct 13, 2026
Heads-up: the project is now called Omarchy Mobile.
Omandroid was never Android, and the other Omarchy Mobile effort is merging into this one, so the project now carries that name. Everything moved today:
omamobile(curated, upstream-ready) andomamobile-int(what the Pixel 7 Pro runs, as a snapshot). The oldomandroidandomandroid-intbranches stay a short while for anyone mid-clone, then go away.omamobile, like Omarchy's omacalc and omawrite:bin/omamobile,omamobile-*packages and commands, the[omamobile]package repository andomamobile-keyring, anddevice/<codename>/device.confas before. The release suffix is-omamobile.omamobile; device notes anddevice.confgo to this repo.Thanks for bearing with the churn. Better now than after a release.
greenbrettmichael commentedon Oct 21, 2026
I have a retail Pixel 8 Pro (husky), also zuma, unlocked and dedicated to this, and I'd like to work on zuma with you, @VirtualMachinist, if you're still on it. I've finished stage 0 on husky. The facts are in
device/husky/NOTES.md(#23), with no serials, IMEI or MACs. The findings that matter for every zuma phone:1. zuma's ABL doesn't pick the base DTB by
/soc_id. The four base DTBs in stockvendor_kernel_boot(there are none invendor_boot, and nodt_tableheader) carry no/soc_idat all. Each has a/soc_compatiblenode with one child per silicon revision and package:product_idmajorminorpkg_modeA0_FOPLPA0_IPOPB0_FOPLP,B1_FOPLPB0_IPOPA retail husky gets DTB 2 (the live DT's
/soc_compatibleholds B0 and B1 FOPLP). SoSOC_IDSand thesoc_idrewrite inbin/omamobile imagesdon't fit zuma. The descriptor in the PR leavesSOC_IDSout on purpose, so the tools refuse it, and proposes:imageswould then require that the DTB is built with a/soc_compatiblenode, write one child per entry, and refuse a descriptor that has both keys or neither. Would you take that as a separate tooling PR, or would you rather shape it yourself? I can write it with a test intests/images-layout.2. Board dtbo: husky MP1.0 boots entry 8 of 19,
board_id 0x40206,board_rev 0x10000(androidboot.dtbo_idx = 8). The other zuma phones will have their own; akita's is worth checking the same way.3. Same shape as cheetah otherwise:
boot init_boot vendor_boot vendor_kernel_boot dtbo, header v4, page size 0x800;#address-cells = <2>/#size-cells = <1>;/ectat0x90000000, size0x14000;no-mapreservations to copy;uart0=/uart@10870000(samsung,exynos-uart).4. Slot hygiene for anyone starting from a used phone: mine arrived with stock active on slot B, and the bootloader only reports the current slot's version. Two sideloads of the same full OTA (Google's own advice for the Pixel 8 family) put the identical build and bootloader on both slots, then
--set-active=a. Root for the capture was a Magisk-patched copy of stockinit_bootflashed toinit_boot_bonly, then put back, so slot A was never written.Next: zuma kernel groundwork in a fork of linux-omarchy-mobile, branch
omamobile-zuma-husky-dtsoffomamobile. It holds a minimalzuma.dtsi(CPUs, GIC, timer, the inherited framebuffer, UART) andzuma-husky.dtsaimed at first light, with Google'sandroid-gs-shusky-6.1as the reference and gs101/gs201 as the pattern. If you've started on clocks or pinctrl for akita, tell me and I'll build on that instead of duplicating it.greenbrettmichael commentedon Oct 22, 2026
Pixel 8 Pro (husky, Tensor G3 / zuma): mainline boots, and USB works.
On slot B, the phone boots our mainline-based kernel into the Omarchy initramfs. The host sees the CDC NCM + ACM gadget at high speed, and
ssh root@172.16.42.1works. All nine CPUs are online and 11.2 GiB of RAM is visible. Stock Android stays untouched on slot A. For now each boot lasts about 80 s: ABL leaves the CPU watchdogs running, and zuma has no watchdog driver yet. That's next.What it took, beyond the minimal zuma DTB:
/soc_compatiblerather than/soc_id, the ECT location in/ect, and a/memorynode.bootandinit_boot. zuma's vbmeta chains them, and ABL rejects them without footers even when unlocked. Today I add these by hand withavbtool, sobin/omamobile imagesshould learn to do it (anAVB_FOOTERSsetting indevice.conf?). I can send that with the/soc_compatiblematching as one tooling PR, if that suits you.zuma keeps no ramoops across a warm reset, so the debugging was done with a stopwatch: a hang marker in the init crashes at a time that encodes how far boot got. Details are in
device/husky/BRINGUP-M1.md(#23) and in derekross/linux-omarchy-mobile#13.Next: the watchdog, then a framebuffer for a screen console (M1.4).
derekross commentedon Oct 22, 2026
Congratulations, @greenbrettmichael: a mainline zuma kernel with USB and SSH up on slot B, and stock untouched on slot A, is a big milestone. The stopwatch hang marker is a neat trick for a SoC with no ramoops.
On the tooling: yes, one tooling PR for both
/soc_compatiblematching and the AVB footers suits me, separate from #23, as two commits (one per feature) with tests intests/images-layout:SOC_COMPATIBLE: a descriptor has exactly one ofSOC_IDSorSOC_COMPATIBLE, the DTB must already have a/soc_compatiblenode, and there is one child per entry. Please keep the format general enough for zumapro (device-tegu: record Pixel 9a first light and staged descriptor #24 needs the same thing).AVB_FOOTERS(or similar) indevice.conf, naming the images that get a hash footer, with the partition sizes taken from the descriptor rather than guessed.avbtoolbecomes a build dependency only for devices that set it.My review of #23 is up. Since then I've also made
tools/install-phonerefuse a descriptor withoutSOC_IDS(and images built for another device), so onceSOC_COMPATIBLElands, the installer needs to accept it too. Please include that in the same PR.On the kernel side, keep the watchdog, PMU, eUSB2 PHY and DTS work in separate subsystem branches, as in derekross/linux-omarchy-mobile#13. @VirtualMachinist, if you've started on akita clocks or pinctrl, please say so here so the two of you don't duplicate it.
greenbrettmichael commentedon Oct 22, 2026
Thanks, @derekross. On the kernel side, it's now split by subsystem in derekross/linux-omarchy-mobile:
Since my last comment, the watchdog driver has removed the 80 s reset. ABL's splash at
0xfac00000now gives a console on the panel. Andreboot-to bootloaderworks once the kernel resets warm, because a cold PSCI reset clears the reboot reason.A correction to my milestone comment above: of the base-DTB items I listed, only
/soc_compatibleis shown to be required (test 5). The ECT location and/memorymatch stock and are in every image that booted, but test 2 showed they weren't why ABL refused our images; the missing AVB footers were.BRINGUP-M1.mdnow says so.The tooling PR is next, as you describe:
SOC_COMPATIBLEandAVB_FOOTERS, with sizes from the descriptor;tests/images-layoutcases, including cheetah's images staying byte-identical with neither key set;tools/install-phoneacceptingSOC_COMPATIBLE, on top of your change.I'll keep the
SOC_COMPATIBLEformat general enough for zumapro (#24). The init's hang marker will go in a separate small PR, if you want it at all.VirtualMachinist commentedon Oct 23, 2026
awesome stuff @greenbrettmichael -- my apologies, but I fear I will have to pull away from this mission, as I just got an offer for a (basically) free upgrade from pixel 8a to pixel 11 and from iphone 14 plus to 18 pro through my mobile carrier, and it requires me to trade them in.
sorry @derekross ! I do have a couple of old iphones that I can dedicate to the cause, as well as a pixel 11 that should be here within the next couple of days, but my pixel 8 is getting shipped away.
derekross commentedon Oct 23, 2026
No apology needed, @VirtualMachinist, that's a good deal to take. Thanks for the help while you had the 8a. If the Pixel 11 ends up free for experiments, #26 tracks Tensor G6, and stock captures from it (
bin/omamobile capture, nothing flashed) would already help there.@greenbrettmichael, akita is zuma like your husky, so the clocks, pinctrl, PMU, watchdog and eUSB2 work all carries over if someone else picks the 8a up later.
device/husky: stage 0 notes and descriptor for the Pixel 8 Pro
omamobile images, install-phone: SOC_COMPATIBLE for zuma's base-DTB m…
omamobile images: AVB_FOOTERS for bootloaders that chain boot partitions
omamobile images: SOC_COMPATIBLE and AVB_FOOTERS for zuma (and zumapr…
greenbrettmichael commentedon Oct 26, 2026
@derekross, a status update on the Pixel 8 Pro, and one question about signing.
UFS and Arch work on husky now.
UFS_RWis writable. The kernel side is linux-omarchy-mobile#49 (PMU tables), DEVICES, README, ROADMAP, CONTRIBUTING: status as of 2026-10-26 #50 (zuma pinctrl, reusing @W4HO-ham's G4 tables) and Pixel 9a (tegu): which SIM or eSIM plan to get for modem, data and SMS testing? #51 (UFS bindings, falling back to the G4 UFS data from the same CAL), plus a DT follow-up to device-oriole: use the shared initramfs init #13 that I'll open once device-oriole: use the shared initramfs init #13 merges.tools/install-phoneinstalled a test root image, and the init handed over to Arch onuserdata. The guard's flags hold under Arch,reboot-to bootloaderworks from there, and a second boot comes back cleanly.The test image was Arch Linux ARM's base, its signature checked through
tools/lib/alarm.sh, plus a newomamobile-device-huskypackage and the kernel's modules. The package is felix's minus the lid drop-in. It's on myhusky-device-packagebranch with husky's UFS-PLAN (U1-U3), and I'll open the PR once #23 merges, since it builds on it.The question: a full Omarchy image (U4).
tools/build-rootfsinstalls our packages fromout/repo, and the build container trusts only the project key inomamobile-keyring. As a contributor I can't sign that repository, and I'd rather not route around the trust model with a local key. How would you like to handle it? I see two options:omamobile-device-huskyin it, and I build the root image from that.Either works for me.
🤖 Generated with Claude Code
derekross commentedon Oct 26, 2026
Thanks, @greenbrettmichael. This is great progress: UFS behind the guard at HS-G4 with stock's partition names, and Arch booting from
userdata, is the milestone that makes husky a real port. And thank you for not routing around the trust model.Let's do option 1. Once #23 and your device-package PR are in, I'll build and sign a repository with
omamobile-device-huskyin it, and publish it so you and other contributors can build root images from it without my key. I'll post the URL here when it's up.A contributor development key (your option 2) is worth having later, since every port hits this before its package merges, but it needs designing so a development image can never be mistaken for a release one. I'll come back to it separately.
45 remaining items
derekross commentedon Nov 7, 2026
Thanks for being straight about the levels. Stay at digital 737 / analog 10 or lower until husky has its own ceilings, and keep
speaker_protection=1off the boot unit. If either amp ever reports a fault or sounds wrong after those sessions, say so here.I'll review #449, linux-omarchy-mobile#203 and #204 on their PRs.
On the System MMU exports: two modules can't export the same symbol, so don't duplicate them. Either make the existing gs201 driver handle zuma (a compatible match plus per-SoC data), which is what I'd prefer if the register layout is close, or, if it really differs, give the zuma file its own prefixed names and have the AoC core pick its ops through the match data. Send the first if it fits, and show in the PR that the gs201 objects don't change.
device/husky: AOC-PLAN.md, level tests as written on derekross#1, acd…
device/husky: AOC-PLAN.md, the AoC and sound results so far (#450)
gitColinX commentedon Nov 9, 2026
Hi all. I have a Pixel 8 Pro I can dedicate to husky as a third test unit.
Read-only so far, no serials:
husky,MP1.0, SKUG1MNW, Samsung 128 GB UFS,dtbo_idx8 (SKU and UFS vendor and size match Brett's unit)2026110201(the security preview build of2026110200) onCP3A.261005.005, security patch 2026-11-01, newer than device.conf's 2026-10-05 AVB footerripcurrent-17.0-15681527(getvar shows the current slot's only)getvar:current-slota,slot-unbootablea:no b:yes (Virtual A/B after the 11-06 update merged, I think),ap-ar-ns/ap-ar-s/gsa-bl1-ar/gsa-fw-ar3/3/3/3, basebandg5300i-260604-260807-B-16036236, storageKLUDG4UHGC-B0E1UFS 3.1 rev3203Nothing is unlocked or flashed, and won't be until my written steps check out against CONTRIBUTING §3. Two questions first, @derekross:
fastboot -w updatewith Google'sCP3A.261005.005image, skipping flash-all's bootloader and radio lines, since both versions already match what's installed. That's an unsuffixed flash of slot A's OS images, and a lower patch level (2026-10-05) than GrapheneOS's 2026-11-01 on the same bootloader. OK, or would you rather I use GrapheneOS's flash-all, as Lef did?omamobile.initramfsboot the right place for the first read-only backup, before Arch loads anything? I've built one by addingomamobile.initramfsto the DTB's/chosen/bootargswithfdtput, in its ownKERNEL_OUT(ABL ignores the vendor_boot cmdline); onlydtb.imgandvendor_kernel_boot.imgdiffer from the install set. Is there a preferred way?I'll add the unit to
device/husky/NOTES.mdas a third phone, facts only.At the lock screen, what helps the next kernel round more from a second Samsung G1MNW unit: #199/#200 on Lef's
husky-test-20261106-pr199-200, or the display and GPU stack (Brett'shusky-test-20261101-e1, or the rebased PRs when he's back)? Firmware will come from my own vendor image, with a fulltools/phone-smokereport per run.Host: Ubuntu 26.04 in WSL2 with usbipd.
kernel-2026.11.01, both image sets and the root image build there (Omarchy edge, because of #433). Fastboot through usbipd works (WSL's getvar matches Windows'); the USB gadget network through it is still untested. I use Claude Code on the PC; every unlock, flash and reboot is my call. @Somnius @greenbrettmichaeldevice/husky: NOTES, a third phone (Samsung 128 GB, UFS 3.1, locked) (#…
greenbrettmichael commentedon Nov 9, 2026
I'm back. Thank you all for a very productive week on the 8 Pro. @Somnius, Bluetooth, haptics, the torch and then the AoC and sound in a few days is great work, and so is the packet-ID fix on the Wi-Fi.
@gitColinX welcome, and thanks for the careful read-only start. The slot A and backup questions are Derek's call. On what to test: #199 and #200 are merged, so on the kernel side the open husky work is the display and GPU stack. I rebased both today onto
omamobile: linux-omarchy-mobile#160 (gs-decon, DSIM0 and the HK3 at 120 Hz) and #158 (which carries #157: the GPU on panthor, CPU DVFS and the thermal zones). A boot of those two on a second Samsung G1MNW unit, with atools/phone-smokereport anduname -r, would help most. They conflict only inzuma.dtsi, where both add nodes at the same place; keep both, in address order. If you'd rather wait, #160 lands first and #158 gets one more rebase after it.gitColinX commentedon Nov 10, 2026
Thanks, @greenbrettmichael. Yes, I'll boot the display and GPU stack on this G1MNW unit as soon as it's running from slot B, and post the
tools/phone-smokereport anduname -rhere.I see #160 merged today, so I'll build
omamobilewith #158 on top once it's rebased. If it isn't by then, I'll resolvezuma.dtsithe way you said, both sets of nodes in address order.Not installed yet: I'm waiting on Derek's answers about slot A and the backup. The kernel, both image sets and the root image are built on
kernel-2026.11.08in the meantime.gitColinX commentedon Nov 11, 2026
@greenbrettmichael, the boot you asked for, on my G1MNW unit (Samsung 128 GB UFS 3.1): your
husky-test-20261109-t1tag (54e10dccc,omamobilewith #160 and #158), built here with the pinned clang, modules installed withmodules-push, slot B written withomamobile flash. The root image is a fresh install frommain(2026.11.09,omamobile-device-husky0.1.8), first booted onkernel-2026.11.08an hour earlier.uname -r:7.3.0-rc6-omamobile-push1-g54e10dcccd6bcard2-DSI-1at 1344x2992 through gs-decon. 3166gs-deconinterrupts at 100 s, no DSI, underrun or flip-timeout lines. The HK3's refused brightness write (sending dcs data 51 05 49 failed: -1) is there too, as on yours.EM: created perf domainfor cpu0, cpu4 and cpu8, noinvalid perf. state.max77759andmax17042_batteryload, as with the 0.1.8 blocklist. The gauge reads 96 % at 36 °C.phone-provisionyet, so there's no firmware on the root.tools/phone-smokeonhusky-test-20261109-t1The new dmesg lines are the BCM4398 ID lines and the exynos-clkout probe that #209 fixes; none come from these PRs.
How I got here, for the notes: GrapheneOS to Google's
CP3A.261005.005on slot A withfastboot -w update(no bootloader or radio write), then the full OTA sideloaded twice, and the read-only NV backup from the guarded initramfs before Arch ever ran. One thing that may help other Windows hosts: through Windows' own UsbNcm driver, the initramfs link stalled for 30 s at a time undernv-backupuntil the phone sent something (ARP stayed Incomplete). Aping -i 0.2from the phone to the host, started on the ACM console, kept it moving, and the backup then ran clean. I'll open an issue with the details.Somnius commentedon Nov 13, 2026
Now that Anthropic shared a reset that we can use until Dec 10.. I'll use the ultra plan to fix all I can with Pixel 8 pro 😈
“ψ(`∇´)ψ
gitColinX commentedon Nov 13, 2026
Haha I feel the same, want to help push this Pixel 8 Pro as much forward as I can over the finish line and try using it as a daily driver phone finally, so help us Claude!
device/husky: ROADMAP, from boots to an everyday phone (derekross#213)
device/husky: AOC-PLAN.md, the AoC and sound results so far (derekros…
device/husky: NOTES, a third phone (Samsung 128 GB, UFS 3.1, locked) (d…
Somnius commentedon Nov 14, 2026
Haaaa fellow brother in arms @gitColinX 😜 (also researched a portable monitor, I got a wireless keyb+mouse [bt] that is why I fixed BT first 🫡)
aninmucek commentedon Nov 15, 2026
Pixel 8a display (akita M6): a design question before any code.
gs-decon's zuma modes are the HK3's (1344×2992), and its probe refuses any other scanout size ("unexpected scanout size"). So on the 8a it stops safely at probe. The 8a has an ak3b at 1080×2400, on the same DECON0 → DSIM0 path, with DSC.
1. Modes. To support both zuma phones, would you rather gs-decon:
2. The panel driver. For the 8a's panel, would you rather have:
panel-google-ak3b.c, with an ID check for the production module0x0080801E, sending only sleep, 60/120 Hz and brightness from Google's GPL driver; orNothing has run on the 8a yet. The next step there is a read-only boot: gs-decon at stage 0, and gs-dsim-cmd with
allow_txoff, to log what the bootloader leaves.🤖 Generated with Claude Code