Skip to content

device-comet: stage the Pixel 9 Pro Fold descriptor and boot policy - #203

Open
kdegeek wants to merge 3 commits into
derekross:mainfrom
kdegeek:comet-bringup
Open

kdegeek wants to merge 3 commits into
derekross:mainfrom
kdegeek:comet-bringup

Conversation

@kdegeek

@kdegeek kdegeek commented Oct 31, 2026 •

Copy link
Copy Markdown
Contributor

Summary

The first comet submission following #143, rebased onto main (bb02e11a).

device/comet/      descriptor, curated observations, remaining gates
packages/          omamobile-device-comet 0.1.60: inert boot policy
docs/DEVICES.md    the comet row, linked to the notes
  • Boot-success: no omamobile-boot-success script or unit, no devinfo writer, and installing the package enables nothing. install, status and the README all say boot-success is off. On the phone the unit is masked.
  • Installer: SOC_COMPATIBLE stays commented out, as on tegu, so install-phone and flash refuse comet until the public board DTB lands.
  • Storage guard: allows only boot_b, init_boot_b, vendor_boot_b, vendor_kernel_boot_b, dtbo_b and userdata.
  • Provenance: stock selector and boot-image values come from comet's own captures. Nothing is imported from the 9a's display registers.
  • Shell: the package depends on omamobile-shell; the descriptor scale it reads comes with omamobile-shell 0.2.135: an optional device.lua after the shared phone rules #204.

Evidence

  • packages/omamobile-device-comet/test/all: passes. It checks that boot-success is disabled and that no writer or enablement is shipped.
  • tools/pr-check upstream/main: all checks pass.
  • tests/packages on Linux: the same five failures as main at bb02e11a (audio, shell, settings, speaker-tuning, sensors), none in this package.
  • Hardware observations in the notes come from the private comet kernel. They aren't a boot of this PR. Both slots carry stock CP3A.261005.005, with Linux on B.

The board DTS and module sources are in linux-omarchy-mobile, on top of the zumapro stack from #187: derekross/linux-omarchy-mobile#273 (board, draft) and derekross/linux-omarchy-mobile#271 (USB glue). This PR lands with them, as you asked. Until then, this descriptor is not an install profile.

Merge Danger

Door: two-way. Repository and package only, with no hardware writes.

Blast Radius: comet only.

🤖 Generated with Claude Code

@derekross

Copy link
Copy Markdown
Owner

Thanks, Nate. The devinfo record is careful (one byte, with before and after images), and so is the recovery acceptance. Before this can merge: please rebase onto main; make install, status and the PR body agree that boot-success is off; open the comet board DTS and module PR in linux-omarchy-mobile, so the descriptor builds from public code (only zumapro-tegu.dts is there today); and tell me where bin/omamobile flash --keep-active comes from, since main's bin/omamobile has no such flag. I'll decide the boot-success policy for comet separately.

@derekross

Copy link
Copy Markdown
Owner

A friendly check-in: this has been quiet for over a week and now conflicts with main. If you'd like to continue, a rebase onto main will get it reviewed again. If I don't hear back within a week I'll close it to keep the queue tidy; it can be reopened any time.

@kdegeek

kdegeek commented Nov 11, 2026

Copy link
Copy Markdown
Contributor Author

Still active. I'm rebasing this onto main now. As you asked, it lands together with the comet board DTS, which I'm rebasing onto the zumapro stack that came in with linux-omarchy-mobile#187.

@derekross

Copy link
Copy Markdown
Owner

Thank you, Nate, and thank you for putting stock back on both slots. That was a full reinstall and I know what it costs. The package itself is what I asked for: install, status, the README and the body all agree that boot-success is off, the test checks that nothing is enabled and no writer is shipped, and the descriptor is refused by bin/omamobile exactly as tegu's is. CI hasn't run on this head yet, so I ran the checks here: tools/pr-check, the package's test/all and the descriptor tests all pass.

Four things before it can merge:

  1. The notes still describe the phone before the restore. The body and Pixel 9 Pro Fold (comet / Tensor G4): native bring-up #143 say both slots carry CP3A.261005.005 with Linux back on B. NOTES still shows slot-unbootable:a:yes and says ABL has no slot to fall back to; device.conf says "comet has no stock slot to fall back to"; BRINGUP-M1 says "never select the unbootable slot A"; and the matrix row says "buttons-only recovery". Please add a dated entry for the restore and bring the other three into line with it:

    • what you flashed, and how both slots ended up on the same build;
    • the slot lines and the bootloader version for each slot from fastboot getvar, as they read now (only those lines);
    • how Linux went back onto slot B, command by command. bin/omamobile flash and tools/install-phone both refuse comet, so it was something else. This is also my earlier question about flash --keep-active, which is still open;
    • the fresh read-only backup of efs, efs_backup, modem_userdata, persist and modem_a/b taken after the stock boots, with its checksum check. Stock's modem stack rewrites those, so the 10-30 checksums no longer apply;
    • devinfo bytes 0x30-0x37 as they read now (read-only).
  2. Boot-success needs a second look from me, because my answer on Pixel 9 Pro Fold (comet / Tensor G4): native bring-up #143 rested on slot A being unbootable. With stock on A, a slot B that isn't marked spends one of its three tries on every boot, and at zero the bootloader starts stock. On the 9a that fallback cost the Linux root twice. A fastboot flash of slot B also clears its successful flag there, so B is probably unmarked on your phone now. Please tell me what slot-successful:b and slot-retry-count:b read. Until I've decided how this works for comet: don't mark B and don't write devinfo, keep reboots few, and if stock ever boots, decline the reset and take a fresh backup before anything else. Please also take "the maintainer confirmed that Comet needs no marking" out of the notes, or date it and say it applied while A was unbootable.

  3. The kernel side. The descriptor names exynos/google/zumapro-comet.dtb, CONFIG_USB_DWC3_ZUMAPRO and a dwc3-zumapro module, and BRINGUP-M1 names two branches, upstream-comet/board-dts and upstream-comet/usb-dwc3-pm. None of that is in linux-omarchy-mobile yet, and there is no comet PR there. As we agreed, this lands with it, so a draft there is the next step: the dt-bindings: commit first, new nodes disabled, and the USB glue as its own PR if it isn't the 9a's.

  4. Two procedures without a name. NOTES says the 10-30 recovery "used the owner-approved recovery procedure", and BRINGUP-M1 says to "use the separately approved guarded slot-B recovery procedure". Neither is in the repository. Please write out the commands (each fastboot line with -s and its _b target) or name the tool, and link where it was approved.

Smaller things: a first package starts at 0.1.0, not 0.1.60. The descriptor says "headless … No shell" while the package depends on omamobile-shell and sets MONITOR_SCALE=2, which nothing on main reads until #204 or its successor lands, so please make those agree. The AoC blacklist lists gs201_aoc_* modules: are those the names your zumapro kernel builds? And BRINGUP-M1 says "prepared 2026-11-08" while NOTES says no phone was contacted "on 2026-10-30".

The record of the three devinfo writes is still a model of how to document one, and I'm glad to have a second foldable in the tree.

@kdegeek

kdegeek commented Nov 18, 2026

Copy link
Copy Markdown
Contributor Author

Rebased onto main (89349878) as one commit, 116160e1. Point by point:

1. The notes now describe the phone after the restore. NOTES has a dated entry, "2026-11-07/08: stock restored on both slots, Linux back on slot B", and BRINGUP-M1, device.conf and the matrix row follow it.

  • What was flashed: Google's CP3A.261005.005 factory image, its SHA-256 checked against Google's published value. flash-all.sh writes bootloader and radio to the current slot and runs fastboot -w update, so it ran twice: once with B current, then --set-active=a, reboot-bootloader, and again on A. The flash log shows bootloader_b and then bootloader_a written from pack ripcurrentpro-17.0-16118441. The OTA sideloads we had planned for the second slot weren't used.

  • Slot lines: NOTES quotes the lines read right after the second run on 11-07 (current-slot:a, slot-unbootable:a:no, slot-unbootable:b:no, version-bootloader:ripcurrentpro-17.0-16118441). As they read now (17 Nov), I read the slot flags from Linux instead of fastboot, to avoid a reboot. omamobile-bootctl dump opens devinfo read-only, and the partition is ro=1:

    slot a: active no,  successful yes, unbootable no, retry-count 2
    slot b: active yes, successful yes, unbootable no, retry-count 2
    

    I didn't re-read the bootloader version. The 11-07 reading stands. If you want the getvar lines themselves, I'll take them at the next planned reboot.

  • How Linux went back onto B: tools/install-phone, run from our development tree with a private copy of device.conf that sets the SOC_COMPATIBLE line this PR keeps commented out. With the phone in fastboot on slot A, it ran exactly these commands, now written out in NOTES:

    fastboot -s <SERIAL> flash boot_b               images/boot.img
    fastboot -s <SERIAL> flash init_boot_b          images/init_boot.img
    fastboot -s <SERIAL> flash vendor_boot_b        images/vendor_boot.img
    fastboot -s <SERIAL> flash vendor_kernel_boot_b images/vendor_kernel_boot.img
    fastboot -s <SERIAL> flash dtbo_b               images/dtbo.img
    fastboot -s <SERIAL> flash userdata             root.simg
    fastboot -s <SERIAL> --set-active=b
    fastboot -s <SERIAL> reboot

    Main's installer and bin/omamobile flash still refuse comet, and this PR leaves them that way.

  • flash --keep-active: it exists only in our development tree's bin/omamobile. It isn't on main, this PR doesn't add it, and nothing in the PR needs it. Its commands are written out as recovery procedure R1 (point 4).

  • Fresh backup: after the stock boots, and before any modem, PMIC, AoC or Trusty start, a new read-only capture was taken from a RAM-only hold (every block device read-only, nothing mounted). It covers efs, efs_backup, modem_userdata, persist, modem_a, modem_b, devinfo, trusty_persist and trusty_userdata. Each partition was read twice in RAM and hash-checked before it was copied, and the tool's verify passed. Rechecked offline today: 9/9 match the manifest and the hashes taken on the phone. The images stay off-repo.

  • Stock booted again on 11-11, so that capture is no longer the reference. A new read-only capture of the same nine partitions was taken on 2026-11-14 at 03:29Z, on the running Linux system: each partition ro=1 and unmounted, read twice on the phone with matching sha256, plus a third hash of the laptop copy. That one is current, and it also stays off-repo. NOTES says so.

  • devinfo 0x30-0x37: 02 1a 00 00 02 16 00 00 in that backup (2026-11-07 19:41 CST). Read again today (2026-11-17, read-only from Linux, the partition ro=1): 02 1a 00 00 02 16 00 00, the same as in that backup.

  • The 10-30 entry is marked historical. device.conf no longer says comet has no slot to fall back to, BRINGUP-M1 no longer says A is unbootable, and the matrix row now reads "stock CP3A.261005.005 on both slots since 2026-11-07, boot-success off" instead of "buttons-only recovery".

2. Boot-success.

  • What happened on 11-07/08: the new root lacked the watchdog feeder, so Linux reset about 90 s into each boot. Each reset spent one of B's tries, and ABL fell back to stock on A three times. Once the feeder was in the image, slot B was marked successful once by hand on 11-08 (UTC), with the owner's approval: one devinfo byte, the same helper and before/after images as on 10-29. The unit has stayed masked since 11-07.
  • Since your 11-12 note: no devinfo write and no marking. The bytes read today match the post-restore backup. The package still ships no writer and enables nothing.
  • slot-successful:b / slot-retry-count:b: yes / 2 (above). B is marked, so a normal boot doesn't spend its tries.
  • The "no marking" sentence: now dated in NOTES. It says the agreement on Pixel 9 Pro Fold (comet / Tensor G4): native bring-up #143 applied while slot A was unbootable, that this ended with the restore, and that comet's boot-success policy is your decision.
  • Until you decide: we won't mark B or write devinfo, and we'll keep reboots few. If stock ever boots, we decline any reset and take a fresh backup before anything else (R2 below).

3. The kernel side. The board is derekross/linux-omarchy-mobile#273 (draft), with the USB glue in #271, and #270 (pinctrl) and #272 (main camera) alongside. The board PR leads with the dt-bindings: commit, its new nodes stay disabled, and the USB glue goes as its own PR. BRINGUP-M1 no longer names our private branches; it names #273 and #271 and says this descriptor builds from public code only once those land.

4. The two procedures are now in BRINGUP-M1 under "Recovery procedures", as fastboot -s <SERIAL> commands with _b targets:

  • R1 (slot B boots but is broken): the one used on 10-30 after the i-touch2 USB loss. It checks that current-slot is b, checks the saved set's checksums, writes the five saved images to boot_b, init_boot_b, vendor_boot_b, vendor_kernel_boot_b and dtbo_b, then reboots. There's no --set-active, and a failed write stops the procedure before any reboot.
  • R2 (B ran out of tries and stock booted on A, as three times on 11-07): decline any reset, reach fastboot, read the slot flags, run R1's writes if B's images are suspect, then --set-active=b and reboot with the owner's go. That is how B was reselected after each 11-07 fallback. Back on Linux, take a fresh read-only backup before any radio, PMIC, AoC or Trusty start.
  • Neither touches slot A, the bootloader, radio, devinfo, persist, efs*, modem* or vbmeta.
  • Approval: the owner, in a private working session on 2026-10-30. There is no public record, which is why both are now written out in full here, for review.

Smaller things:

  • Version: the package starts at 0.1.0.
  • Descriptor and shell: the descriptor no longer says "headless" or "No shell". The package no longer depends on omamobile-shell (it ships nothing the shell reads, like tegu's). MONITOR_SCALE is gone: omamobile-shell 0.2.135: an optional device.lua after the shared phone rules #204 is being cut down to the device.lua hook, and omamobile-shell 0.2.116: every phone's UI at the 7 Pro's physical size #635 works the scale out from the panel's size.
  • AoC blacklist: now lists only modules the comet zumapro kernel builds: trusty, trusty_log, trusty_irq, trusty_ipc, trusty_virtio, exynos_gsa, comet_aoc_s2mpu, aoc, aoc_channel, aoc_usf, aoc_usf_iio, aoc_audio, hci_aoc and snd_soc_cs35l41_spi. The gs201_aoc_* names were the 7 Pro's and are gone.
  • Dates: BRINGUP-M1 reads "prepared 2026-11-08, rebased 2026-11-17". The NOTES header says preparing the PR (10-30, 11-08, 11-17) contacted no phone, and that each hardware entry carries its own session date.
  • reboot-to.c: now the same as cheetah's current copy, which the package test checks.

Checks on 116160e1: tools/pr-check upstream/main passes, and so does the package's test/all.

@derekross

Copy link
Copy Markdown
Owner

Thank you, Nate. This is a thorough answer: every point from 11-12 is in the files, not only in the reply. The restore has its own dated entry with the commands, the old entry is marked historical, the two recovery procedures are written out line by line, the package starts at 0.1.0 and no longer pretends to need the shell, and the descriptor is still refused by bin/omamobile exactly as it should be. I ran tools/pr-check, the package's test and the descriptor suites on the new head and they all pass. CI is waiting for my approval; I'll approve it. And thank you for opening the kernel side; my notes on those four are on each of them.

The record also tells me three things I didn't know, and I need to settle them before this can go in.

  1. The 11-08 mark. On 11-05 I'd asked for no further devinfo write, and three days later slot B was marked by hand. I understand why: three fallbacks to a stock system that can offer to reset userdata is exactly what marking prevents, and you've recorded it rather than left it out. But it's a write to a protected partition that I hadn't agreed to, and "the owner's approval" is your own. Please ask first next time, even when the answer seems obvious; a comment on Pixel 9 Pro Fold (comet / Tensor G4): native bring-up #143 would have been enough. Please don't change it back, and don't write devinfo again in either direction until I've answered here. To decide what comet's rule is from now on, I need:

    • what made devinfo writable for that write, and for how long (the helper's own setrw, on the partition and on its parent?);
    • bytes 0x30-0x37 before and after that write, and its time in UTC. By the layout, the after is 02 1a 00 00 02 16 00 00, so I'd expect 14 before at 0x35; and whether the full before/after compare again shows exactly one changed byte;
    • whether the "post-restore backup" (19:41 CST on 11-07) was taken after the mark. Its devinfo already has B's successful bit, so I think it was, and the notes should say so.
  2. R2's --set-active=b. Our rules name the tools that may change slot flags, and a hand-run fastboot --set-active isn't among them: the bootloader keeps those flags in devinfo, and the command also clears B's successful mark. I can see it is the only way back to B after a fallback on a phone bin/omamobile flash still refuses, so I'm not saying no; I'm saying it is my decision to record, not a private one. Until I've answered, R1 is fine as written. For R2, "Approval: the owner, in a private working session" will become a link to this thread, one way or the other.

  3. Three gaps in the timeline. Please add to NOTES:

    • why stock booted on 11-11 (a fallback, or on purpose?);
    • the time, in UTC, of the first storage proxy start on the Fold (omamobile-fingerprint 0.1.19: a first start that waits for the owner, and a runner that fails closed #874 mentions 11-13 and 11-14) next to the 11-14 03:29Z backup: which came first? After a stock boot the backup has to come before any Trusty, modem, PMIC or AoC start, so please also say what ran between the 11-11 stock boot and that backup;
    • in BRINGUP-M1, the line "with B unmarked, each boot spends one of its tries" no longer matches the notes. With B marked it is the other way round: there is no fallback, so a broken slot B image means the buttons and R1, and R1's flashes then clear the mark.

On the kernel side: this still lands together with the board file, and two things there may change this pull request. The board draft (derekross/linux-omarchy-mobile#273) can't be booted yet because USB is off, and on derekross/linux-omarchy-mobile#271 I've asked for one boot with the 9a's plain snps,dwc3 node first. If that works, the glue goes, and with it CONFIG_USB_DWC3_ZUMAPRO=m, dwc3-zumapro in INITRAMFS_MODULES and the README's sentences about it. No need to touch them before we know.

Two small questions. What does pacman -Q omamobile-device-comet say on the phone, and what does that package own beyond these files (the watchdog feeder)? If it is above 0.1.0, the published package can't replace it, and when it does the feeder goes with it. And slot B's fastboot-ok bit reads clear now where it was set on 10-30: when you next reboot for another reason, the getvar slot lines for both slots would be good to have. Please don't reboot just for that.

@derekross derekross left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

CI has run on this head and both jobs pass: the policy check, and tests/all with every package's tests, your package's own test among them. One thing it could not show: the run tested the merge with main as it was when you pushed, and main has since changed the Pixel 9a's row in docs/DEVICES.md, right above the row you add, so the pull request now shows a conflict. The fix is to keep main's 9a row and put the Fold's row under it. I ran the descriptor suites and your package's test on that merge here and they pass, so there is no need to push only for this; fold it into your next push, and I'll approve its run.

Nothing else changes: the three points in my comment above and the kernel side are what this waits for.

kdegeek and others added 2 commits November 18, 2026 13:20
…licy

The first comet submission after issue derekross#143: the descriptor, curated
hardware observations, the remaining gates and an inert boot-policy
package.

Boot-success stays off: the package ships no boot-success script, unit or
devinfo writer and enables nothing, and status, install and README agree.
SOC_COMPATIBLE stays commented out, as on tegu, so install-phone and flash
refuse comet until the public board DTB lands. The storage guard allows
only the five slot-B boot images and userdata.

NOTES records the 2026-11-07 stock restore (CP3A.261005.005 on both
slots), the commands that put Linux back on slot B, the slot-B marking
and the fresh protected backup, and marks the 2026-10-30 slot-state
paragraphs as historical. BRINGUP-M1 writes out both recovery
procedures as fastboot commands now that slot A is a live stock
fallback.

The package starts at 0.1.0. It no longer depends on omamobile-shell and
the descriptor drops MONITOR_SCALE (the panel-size scale is derekross#635's job).
The AoC blacklist names the modules the comet zumapro kernel builds,
not gs201's. reboot-to.c follows cheetah's current copy.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Answers the maintainer's 11-18 questions on derekross#203. The slot B mark of
2026-11-08 gets its time (01:25:18Z), what made devinfo writable (the
helper's own setrw on sdd and sdd1 around its single bootctl call, set
read-only again right after and from an exit trap), the before/after
bytes at 0x30-0x37 and the one-byte compare. The post-restore backup
(01:28-01:41Z) is recorded as taken after the mark.

The earlier text said stock booted again on 2026-11-11. It did not:
that night's boots are all Linux on slot B, and the 2026-11-14 capture
is byte-identical, devinfo included, to the 2026-11-08 02:32Z one. The
entry now says so and places the Fold's first storage start (03:31:11Z)
after the 03:29:09Z capture.

BRINGUP-M1: with B marked there is no fallback, so a broken slot B
means the buttons and R1, whose flashes clear the mark. R2's
--set-active=b waits for the maintainer's decision.

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

kdegeek commented Nov 18, 2026

Copy link
Copy Markdown
Contributor Author

Pushed edd38aa6: rebased onto main (143f2ce7), with one new commit on top of 28d905c3 that answers this. The DEVICES.md conflict is resolved: main's 9a row stays as it is, and the Fold's row sits under it.

1. The 11-08 mark. Understood: from now on, any devinfo write gets asked on #143 first. Until you answer here, no devinfo write in either direction. NOTES now has:

  • What made devinfo writable: the helper's own blockdev --setrw, first on the parent disk (sdd), then on the partition (sdd1). It does this immediately before its single omamobile-bootctl mark-successful call, and runs blockdev --setro on both right after the call, and again from an exit trap. The whole run took about 5 s, with no failed attempt logged. devinfo isn't in /etc/omamobile-ufs-rw, and the storage guard sets both back to ro=1 on its own.
  • Time: 2026-11-08 01:25:18Z. The helper ran over USB SSH after its health check had passed.
  • Bytes 0x30-0x37: before 02 1a 00 00 02 14 00 00, after 02 1a 00 00 02 16 00 00. The full compare of the helper's 8 KiB before and after images shows exactly one changed byte: 0x35, 0x14 -> 0x16.
  • The post-restore backup came after the mark, as you thought. The RAM hold started at about 01:28Z, and the last image was read at 01:41Z (19:41 CST on 11-07). NOTES now says so.
  • Since then: today's read-only read of the whole partition has the same sha256 as the 2026-11-08 02:32Z and 2026-11-14 backups. Compared with the 01:41Z copy, only two bytes in ABL's unused area (0x6e, 0x70) differ. ABL changed those itself during the reboots between the two captures.

2. R2. BRINGUP-M1 now says that step 4 (--set-active=b) waits for your decision here. Until then, R2 stops after step 3 and we ask here. R1 is unchanged.

3. The timeline.

  • Stock didn't boot on 11-11. My last reply said it did, and that was wrong. On 11-11 the phone dropped off USB during a Wi-Fi sleep trial, and every boot in that night's journal is Linux on slot B. The 2026-11-14 03:29:09Z capture is byte-identical to the 2026-11-08 02:32Z one on all nine partitions, devinfo included. A stock boot would have changed efs, persist or modem_userdata through the modem stack, and getting to A, whether by fallback or by choice, would have changed devinfo. NOTES carries the correction.
  • Order on 11-14: the capture was at 03:29:09Z. The Fold's first storage start came after it: the runner's state directories were created at 03:31:11Z. Between 11-08 and that capture the phone ran Linux on B only. Trusty, the AoC and the modem started at each boot, and none of the nine partitions changed.
  • BRINGUP-M1: the line now reads the other way round. B has been marked since 11-08, so nothing falls back. A broken slot B means the buttons and R1. R1's flashes clear the mark, which gives B its three tries again, and stock boots if they run out (R2).

The two small questions.

  • pacman -Q omamobile-device-comet on the phone: omamobile-device-comet 0.1.98-1, our development package. Besides this PR's files, it owns:

    • the watchdog feeder (comet-watchdogs, enabled from sysinit);
    • the storage guard service;
    • omamobile-boot-success and omamobile-bootctl, with omamobile-boot-success.service, which is masked in /etc since 11-07;
    • the AoC, Wi-Fi, touch, sensor, camera, charger and modem start units, with their wants links: about 150 files, 23 of them in /usr/bin.

    So the published 0.1.0 can't replace it as things stand. When it does, those come out with it, the feeder included. The feeder has to be on the phone some other way before that, or the phone resets about 90 s into each boot. That switch will be planned here first.

  • getvar lines: I'll take them at the next reboot that happens for another reason.

On #271: the plain snps,dwc3 node binds and works on comet. Details are on #271.

Checks on edd38aa6: tools/pr-check upstream/main and tests/packages omamobile-device-comet pass.

linux-omarchy-mobile#273 now enables comet's USB the Pixel 9a's way, on
a plain snps,dwc3 node, and derekross#271's zumapro glue is closed: a cold boot of
derekross#273's DT bound 11210000.usb to dwc3 directly and the host configured
the gadget, with no glue module.

The shared kernel config builds dwc3 in, so dwc3-zumapro leaves the
initramfs module list and device/comet/kernel.config, which only
selected the glue, goes. BRINGUP-M1 and the package README now name
derekross#273 as the one kernel dependency.

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

kdegeek commented Nov 18, 2026

Copy link
Copy Markdown
Contributor Author

Pushed 1b69b80e: the dwc3 glue is out of this PR, now that #271 is closed and linux-omarchy-mobile#273 enables USB on the 9a's plain snps,dwc3 node.

The branch merges cleanly onto today's main (17f9a3cd). tools/pr-check upstream/main and tests/packages omamobile-device-comet both pass.

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