Skip to content

device-panther: Pixel 7 first port, and contributor builds with a local key - #79

Merged
derekross merged 5 commits into
derekross:mainfrom
nesmeck:panther-port
Oct 29, 2026
Merged

derekross merged 5 commits into
derekross:mainfrom
nesmeck:panther-port

Conversation

@nesmeck

@nesmeck nesmeck commented Oct 27, 2026

Copy link
Copy Markdown
Contributor

What and why

A first port to the Pixel 7 (panther, gs201), which was under Wanted in
docs/DEVICES.md, plus the tooling a contributor needs to build a root image
for a new device without the maintainer's signing key.

  • device/panther/: the descriptor, an empty board dtbo, NOTES.md and
    BRINGUP-M1.md, with the record of every boot (first light, the
    bootloader's framebuffer, UFS behind the guard, the Arch root, the push
    loop, the keys, and the touch work so far). docs/DEVICES.md gets a
    Pixel 7 row.
  • packages/omamobile-device-panther 0.1.0: the storage guard rule, the
    USB link, the power key and reboot-to, from omamobile-device-felix
    without the lid and cover-touch parts. No omamobile-boot-success: this
    phone's devinfo is version 3.19, and bootctl is verified on 3.15 only
    (L-36).
  • tools: local-key-init, fetch-published, and --local NAME in
    build-packages and build-rootfs, so a contributor's packages are
    signed with keys/NAME and land in out/NAME/ instead of keys/ and
    out/repo (L-7). tests/local-builds covers the refusals;
    docs/RELEASES.md §10 and AGENTS.md describe the flow.

The board DTS and a work-in-progress FT3658 touch driver are on a local
kernel branch; they will be a separate pull request against
linux-omarchy-mobile's omamobile once they are reworked for it.

Device and build

  • Phone: Pixel 7 (panther), stock CP3A.261005.002.A1 on slot A,
    bootloader cloudripper-17.0-15902555 on both slots
  • Kernel release (uname -r): 7.3.0-rc6-omamobile-push1-g1cb4bd36f3be-dirty
    (omamobile-int plus the local panther board file)
  • Packages changed, with versions: omamobile-device-panther 0.1.0 (new)

Checks (no phone needed)

  • tools/pr-check upstream/main passes
  • tests/all passes
  • tests/packages passes (the package I touched: omamobile-device-panther)
  • Each changed package has its version bumped, and the new version is in the commit subject
  • Commits have area prefixes and bodies that say why; any AI help is a Co-Authored-By: or Assisted-by: trailer, never a Signed-off-by:

On the phone

  • I ran this on my own phone, booted into this build
  • Runbook steps passed (device/panther/BRINGUP-M1.md): steps 1-6 pass
    (first light, framebuffer and trigger, UFS behind the guard, Arch root
    through install-phone); the push loop and boot 5 (keys) pass. Boot 8
    (HSI1 pin controller) passes; the touch controller does not answer yet.
  • What I did not test: touch (not working yet), GPU, Wi-Fi,
    Bluetooth, battery, sound, sensors, cameras, suspend; marking slot B
    successful; tools/phone-provision. The root image was built with
    --local from published packages plus the tree's omamobile-omarchy
    0.1.12, omamobile-shell 0.2.28 and omamobile-firstboot 0.1.2.
  • Nothing in this change writes outside slot B and userdata, weakens the storage guard, or reads registers from userspace (AGENTS.md §1)

tools/phone-smoke from the laptop printed no report: a command in the
script reads the rest of it from stdin under bash -s. The report below
was produced by copying the script to the phone and running it with
--here < /dev/null. The phone's clock is wrong (no RTC, no network).

tools/phone-smoke report

phone-smoke: panther

Kernel 7.3.0-rc6-omamobile-push1-g1cb4bd36f3be-dirty, tools at 4b81e04, 2026-10-18T17:25Z.

Check Result Detail
boot (required) ok kernel 7.3.0-rc6-omamobile-push1-g1cb4bd36f3be-dirty, slot b, up 1066s
storage guard (required) ok 55 partitions; only userdata and slot B's boot images writable
root (required) ok ext4 on userdata, 99G free
kernel health (required) ok no oops or BUG; 0 WARN, 7 error lines in dmesg
systemd units ok none failed
USB gadget (required) ok bound to 11210000.usb
display ok card0-Unknown-1 connected, 1080x2400
GPU not ok no render node
session (required) ok Hyprland and the Omarchy shell running
lock ok PAM file present, the shell can lock (passwordPam true, locked true)
touch not ok no multitouch input device
buttons ok gpio-keys
Wi-Fi skip no wireless interface
Bluetooth skip no controller
battery skip no battery power supply
sound cards skip none
sensors skip no sensor proxy
cameras skip no libcamera tools

Packages: hyprland 0.56.2-5.1, libcamera 0.7.2-4.21, mesa 1:26.2.4-1.4, omacam 0.3.3-1, omamobile-apps 0.2.0-1, omamobile-device-panther 0.1.0-1, omamobile-firstboot 0.1.2-1, omamobile-keyring 20261026-1, omamobile-mirror 0.1.1-1, omamobile-omarchy 0.1.12-1, omamobile-owl-widgets 0.1.1-1, omamobile-shell 0.2.28-1, omarchy 4.0.6-1, quickshell 0.3.2.r7.g2b4501a-1

🤖 Generated with Claude Code

@derekross

Copy link
Copy Markdown
Owner

Thank you, @nesmeck. This is the best-documented first port we've had: one register per call with the reason it's safe to read, the trigger derived on this phone and only then put into the descriptor, the guard checked partition by partition against stock's by-name map, the Arch root through install-phone, the devinfo read limited to the slot bytes and the version, and a fresh backup from the guarded initramfs. The boot-6 finding (the three-boot budget after --set-active, and that push doesn't restore it) is exactly the kind of thing the notes are for, and the touch investigation rules things out in the right order.

The local-key tooling does what I wanted: the key lives in keys/<codename>/ and the build image only, the keyring package is untouched, imported packages keep their own signatures, and the tests move the key away and plant a keys/fingerprint to prove nothing falls back to mine. I ran tests/all, tests/local-builds and the panther package's test on your head; all pass. I'll merge it after two small changes:

  1. Say in the matrix that the kernel isn't published yet. uname -r ends in -dirty with gs201-panther.dts on a local branch, so nobody can build these images from our trees today. The Pixel 9a row writes "unpublished kernel" in each such cell; please do the same for the Storage, Buttons and USB cells (Boot already says enough). And when you push the board file as a public branch or a draft kernel PR, link it from the row and the notes, so the record points at code.
  2. The slot A fallback in boot 6 was a stock boot. Stock's recovery can write persist and efs, so AGENTS asks for a fresh read-only backup after any stock boot, a fallback included. Please add a NOTES line saying whether you declined the data reset and that you took (or will take) a new backup after that boot, and a sentence in the runbook's rollback: after a fallback, decline the reset and back up again.

Two smaller things, not blocking:

Merge order: #75 touches the same line of tests/all, so I'll merge that first and ask you for a quick rebase. Thanks again.

@nesmeck

nesmeck commented Oct 28, 2026

Copy link
Copy Markdown
Contributor Author

Thanks for the review. Both changes are in ec91153:

  1. "unpublished kernel" in the Storage, Buttons and USB cells of the Pixel 7 row. I'll link the board file from the row and the notes once it's on a public branch or a draft kernel PR.
  2. NOTES.md (2026-10-27) records boot 6's stock fallback: the data reset in recovery was declined (adb reboot bootloader), what changed against the 2026-10-26 baseline (only the F2FS checkpoint areas of efs, efs_backup and modem_userdata, and persist's ext4 superblock and journal; modem_a/modem_b match), and the fresh read-only backup taken afterwards. The runbook's rollback in step 2 now says to decline the reset and back up again after any fallback.

tools/phone-smoke: bash -x on the phone side shows the trace stopping at line 230, pkgs=$(pacman -Q 2>/dev/null | ...). On a --local image the phone doesn't trust the contributor key (as RELEASES §10 says), and pacman -Q still loads the [omamobile] sync database, so it asks :: Import PGP key 4185...? [Y/n] on stderr, which is hidden, and reads the answer from stdin: the rest of the script. With pacman -Q </dev/null 2>/dev/null the full report prints from the laptop on panther. Any phone with a sync database signed by an unknown key would hit the same thing.

tools/pr-check upstream/main passes. I'll rebase once #75 is in.

nesmeck added a commit to nesmeck/omarchy-mobile that referenced this pull request Oct 28, 2026
Review of derekross#79. The board file is on a local kernel branch, so the matrix
says "unpublished kernel" in the Storage, Buttons and USB cells, as the
Pixel 9a row does, until it is pushed and can be linked.

Boot 6's fallback to slot A was a stock boot, and stock recovery mounts
persist, efs and modem_userdata. NOTES.md records that the data reset was
declined, what changed against the baseline, and the fresh backup taken
afterwards; the runbook's rollback now says to do the same after any
fallback.

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

nesmeck commented Oct 28, 2026 •

Copy link
Copy Markdown
Contributor Author

The board file is public now as derekross/linux-omarchy-mobile#108 (with the FT3658 driver in derekross/linux-omarchy-mobile#109). It is linked from the Pixel 7 row (Boot, Storage, Buttons and USB, in place of "unpublished kernel"), NOTES.md and the runbook's status.

@derekross

Copy link
Copy Markdown
Owner

Thank you, @nesmeck. Both changes are in, and done well:

  1. The Pixel 7 row now links kernel Pixel 5 (redfin, SM7250 / Snapdragon 765G) port: offering hardware for testing #108 in the Boot, Storage, Buttons and USB cells, and NOTES says plainly that these images were built from the local branch the board file came from. That's the honest version of what I asked for, now that the code is public.
  2. The boot 6 record is what I hoped for: the reset declined, exactly what stock's recovery changed against the 10-26 baseline (checkpoint areas, the superblock and journal, the retry and flag bytes, with modem_a/modem_b untouched), and a fresh baseline dated 10-27. The rollback in step 2 now tells the next person to do the same.

Thanks too for tracking down the tools/phone-smoke problem. pacman -Q reading its key-import answer from the script's own stdin is our bug, and your </dev/null fix is the right one. I'll make that change on our side, so you don't need to carry it here.

The one thing left is mechanical: the branch now conflicts with main in three files, because main moved after your push (#75 is still open, so you don't need to wait for it). When you rebase:

  • tests/all: keep main's for t in … list (it now has tests/alarm-wake-test and tests/initramfs-steps), add packages/omamobile-device-panther/test/all tests/local-builds at the end, and merge the two header comments.
  • tools/build-packages: main added omamobile-device-raven to ORDER; put omamobile-device-panther after it.
  • docs/DEVICES.md: keep main's coral, tegu and husky rows and add your Pixel 7 row after them. In the Wanted list, drop the | Pixel 7 | panther | … line (main still has it), so the phone appears once.

I resolved it this way on a scratch copy, and the matrix still has 25 cells in every row. Whichever of #75 and this lands second will need a one-line touch in tests/all, and I'll handle that.

Two optional things that can wait for a later PR:

Once it's rebased and CI is green, I'll merge it.

nesmeck and others added 5 commits October 28, 2026 13:09
Slot B on the Pixel 7 (panther, gs201) boots a mainline 7.3-rc6 kernel
into the Arch root on userdata, with Hyprland and the Omarchy lock
screen on simpledrm. The board DTS is a local kernel branch and will be
the kernel pull request against omamobile.

What passed on the phone, in order (BRINGUP-M1.md):
- first light: USB SSH, both cluster watchdogs stopped;
- the bootloader's framebuffer at 0xfac00000 (DPP0 RDMA_BASEADDR_Y8),
  shown after DECON0 TRIG_CON 0x1c240030 is set to 0x3061;
- UFS behind the storage guard, with every partition number equal to
  stock's by-name map and a fresh backup taken from Linux;
- the Arch root through tools/install-phone, and bin/omamobile push;
- the power and volume keys, wired as on the Pixel 7 Pro.

The touch controller (FocalTech FT3658 on SPI0) does not answer yet;
the notes record what was ruled out. Slot B is not marked successful:
devinfo is version 3.19 here, against cheetah's verified 3.15, so each
boot spends one of the three retries that --set-active gives.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Pixel 7's device package, from omamobile-device-felix without the
lid and cover-touch parts: the udev rule that keeps every partition
outside UFS_RW read-only, the systemd-networkd and NetworkManager files
for the USB link, logind ignoring the power key (Omarchy's menu takes
it), the mirror list and reboot-to (cheetah's source).

It has no omamobile-boot-success: that writes devinfo, whose layout is
version 3.19 on this phone and has not been checked against ABL yet.

test/all passes. The package is installed in the root image that runs
on the phone, built with tools/build-packages --local.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A contributor building packages for a new device had two bad options:
sign with the maintainer's keys/ (which they don't have), or mix their
packages into out/repo, which tools/repo-push publishes (L-7).

- tools/local-key-init CODENAME makes keys/CODENAME/ once: an ed25519
  signing key, its fingerprint and public.asc.
- tools/fetch-published DIR downloads pkgs.omamobile.org with each
  signature, and refuses an index or package that a key in
  packages/omamobile-keyring does not trust, that is revoked, or that
  does not match the signed index's checksum.
- tools/build-packages --local NAME signs with keys/NAME only, into
  out/NAME/repo, and keeps the published packages' own signatures when
  they are imported. The Pixel 7's package joins the build order.
- tools/build-rootfs --local NAME builds from out/NAME/repo, into
  out/NAME, and the image trusts the local key (alarm-setup lsigns it).
- docs/RELEASES.md §10 and AGENTS.md describe the flow.

tests/local-builds runs every refusal against a fake server and test
keys, in a copy of the repository with its own GNUPGHOME. tests/all
passes. The root image that runs on the Pixel 7 was built this way
(rootfs-smoke 12/12).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Review of derekross#79. The board file is on a local kernel branch, so the matrix
says "unpublished kernel" in the Storage, Buttons and USB cells, as the
Pixel 9a row does, until it is pushed and can be linked.

Boot 6's fallback to slot A was a stock boot, and stock recovery mounts
persist, efs and modem_userdata. NOTES.md records that the data reset was
declined, what changed against the baseline, and the fresh backup taken
afterwards; the runbook's rollback now says to do the same after any
fallback.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The board file is now public as linux-omarchy-mobile#108, so the Pixel 7
row links it in place of "unpublished kernel" (Boot, Storage, Buttons
and USB), as the Pixel Fold row links its kernel PRs, and the notes and
the runbook point at it.

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

nesmeck commented Oct 28, 2026 •

Copy link
Copy Markdown
Contributor Author

Rebased on main as you described (tests/all keeps main's list with the panther test and tests/local-builds at the end; panther after raven in ORDER; the Pixel 7 row after husky and the Wanted line dropped). tools/pr-check upstream/main and tests/all pass. The "in review" wording for derekross/linux-omarchy-mobile#108 will become "merged" in the touch PR.

@derekross

Copy link
Copy Markdown
Owner

Thank you, @nesmeck. The rebase is exactly what I described, and nothing else moved: the five commits carry the same changes as before, apart from the three conflict resolutions.

  • tests/all keeps main's list, with the panther package's test and tests/local-builds at the end, and the two header comments are merged.
  • omamobile-device-panther comes after omamobile-device-raven in ORDER.
  • The Pixel 7 row sits after the Pixel 8 Pro's, the Wanted line is gone, and every matrix row still has 25 cells.

Main moved twice more after your push (the raven version fix, and the kernel tree's new default location). Your branch still merges onto today's main cleanly, and on that merge I ran tools/pr-check (both your copy and main's), tests/all, tests/local-builds and the panther package's test. All pass. The merge keeps main's KERNEL_BASE default in tools/build-rootfs, so nothing goes back to the old hard-coded path. I also re-read the signing path one more time. Without --local, build-packages signs with the same key into the same repository as before. Its build image is rebuilt once, because alarm-setup changed, but the only new step in it is skipped when there's no local key. And repo-publish checks every signature against the keyring by itself anyway, so a package signed with a local key could never be published by accident.

CI is waiting for my approval because the branch was pushed again. I'll approve the run, and once it's green I'll merge this with the squash title device-panther 0.1.0, tools: the Pixel 7's first port, and contributor builds with a local key.

A few small things for your next PR. None of them blocks this one:

  • Kernel Pixel 5 (redfin, SM7250 / Snapdragon 765G) port: offering hardware for testing #108 and Package publish failed on the build machine #109 were both merged today. The Pixel 7 row and the status line at the top of BRINGUP-M1.md still say "in review", and the Touch cell still says there's no driver. You said you'll update these in the touch PR, and that's fine.
  • Pixel 5 (redfin, SM7250 / Snapdragon 765G) port: offering hardware for testing #108 went in without mem_sleep_default=s2idle in the bootargs. That makes the sleep.conf.d/10-omamobile.conf drop-in (as in the Pixel 7 Pro's and Pixel 6 Pro's packages) the only thing that would keep Omarchy's Suspend off deep, which gs201 never wakes from. Please bring it in with the touch work, before anyone has a reason to tap Suspend.
  • An optional hardening: ALARM_LOCAL_KEY is read from the environment, so a shell that still had it exported would make even a build without --local trust that key inside the build image. It changes nothing that gets published, but clearing it in both tools when --local isn't given would make the separation hold by construction.

The pacman -Q fix in tools/phone-smoke is still on me, and I'll do it separately. Thanks again for a careful port and for tooling that every contributor can use.

derekross added a commit that referenced this pull request Oct 28, 2026
L-279 to L-293: the 9a's slot B hold and what made devinfo writable
(#125), per-device flags checked by building both command lines
(#101), env defaults in the shared build library and CI waiting for
approval (#79), GLib retry paths and radio stops checked at the
controller (#99), silent re-pushes and stacks with several tops (the
tegu stack), and two of our own: merge only after every check has
finished (#126), and an empty diff is checked for empty inputs
(kernel #123).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
derekross added a commit that referenced this pull request Oct 28, 2026
From the laptop, phone-smoke reaches the phone on bash's stdin
(tools/phone-ssh "bash -s" < tools/phone-smoke). pacman -Q also loads
the sync databases, and one signed by a key the phone doesn't trust (a
--local build's, RELEASES §10) makes it ask on stderr whether to import
the key and read the answer from stdin: the rest of the script. The
report then never printed. A contributor found this on panther (#79).

pacman -Q now reads /dev/null, which also means it can never answer
yes to a key import, and so do the commands run as the desktop user,
which never need input.

Reproduced with a stand-in pacman that reads stdin as fgets does,
piped like tools/phone-ssh: before, it took the last 1212 bytes and no
report printed; after, it reads nothing and the full report prints.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
derekross added a commit that referenced this pull request Oct 28, 2026
…wenty-sixth rounds (#127)

* docs/review: 19 lessons and 3 corrections from the twenty-fifth round

The confirmed candidates from round 25 (L-260 to L-278): the 9a slot B
marking as a devinfo question, cross-repository links, squash-merged
stacks and version files, stale hand-built prebuilts, and before-merge
Pixel 7 Pro runs on omamobile-int. C-16 to C-18 correct our own earlier
comments on kernel #98, kernel #80 and #74; L-241 is corrected in place.

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

* docs/review: 15 lessons from the twenty-sixth round

L-279 to L-293: the 9a's slot B hold and what made devinfo writable
(#125), per-device flags checked by building both command lines
(#101), env defaults in the shared build library and CI waiting for
approval (#79), GLib retry paths and radio stops checked at the
controller (#99), silent re-pushes and stacks with several tops (the
tegu stack), and two of our own: merge only after every check has
finished (#126), and an empty diff is checked for empty inputs
(kernel #123).

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

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
@derekross
derekross merged commit d88a2ae into derekross:main Oct 29, 2026
2 checks passed
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