Skip to content

device-dodge 0.1.0: the OnePlus 13's device files, and qcom-abl boot images - #111

Merged
derekross merged 32 commits into
derekross:mainfrom
mrlund:dodge/pr-device-v1
Oct 31, 2026
Merged

derekross merged 32 commits into
derekross:mainfrom
mrlund:dodge/pr-device-v1

Conversation

@mrlund

@mrlund mrlund commented Oct 28, 2026 •

Copy link
Copy Markdown
Contributor

dodge: OnePlus 13 device files (device-only)

Base: main
Branch: dodge/pr-device-v1 on https://github.com/mrlund/omarchy-mobile
Companion kernel PR: derekross/linux-omarchy-mobile#115

Summary

Device-only first cut for the OnePlus 13 (dodge, SM8750). The guarded
inactive-slot installer and firmware-match helper are deferred to a later
PR (after the §3.1 call). How this phone was installed is a record in
BRINGUP, not a procedure for others.

  • device/dodge/ — descriptor, NOTES, BRINGUP, audio, Wi-Fi, Bluetooth,
    sensors, modem (opt-in units, no blobs), desktop, display, idle, timesync,
    chainload helpers, reboot wrapper
  • packages/omamobile-device-dodge
  • docs/DEVICES.md matrix row — statuses testing (kernel in review, #115)
    (no Wanted line, no README edit)
  • bin/omamobile — accepts BOOT_METHOD=qcom-abl (two-stage chain via
    tools/dodge-chain; flash refuses qcom-abl). UFS_RW still refuses *_a.
    install-phone still refuses qcom-abl until the installer PR
  • Shared initramfs/init (no dodge fork); UFS_RW="boot_b userdata",
    OMAMOBILE_SLOT=b

Companion kernel: derekross/linux-omarchy-mobile#115.

Feature matrix

Aligned with docs/DEVICES.md and
device/dodge/NOTES.md.

Block Status Notes
Boot testing Omarchy on boot_b; kernel #115; installer deferred
Storage testing boot_b + userdata allowlist; shared init
Display testing msm/dpu; warm-boot drop fixed
GPU testing Adreno 830 / gpu2; Vulkan untested
Touch testing
Buttons testing
USB net testing CDC NCM + SSH; gadget stuck after replug until reboot
Wi-Fi testing WCN7860 / ath12k
Bluetooth testing pairing; A2DP untested
Battery testing gauge; charging path not claimed
Speakers/mics testing speaker, earpiece, mics; headset plug untested
Sensors testing accel, prox, compass; autorotate
Modem testing opt-in flag; LTE data; calls/SMS untested; needs owner's firmware + EFS
Suspend WIP s2idle not default
Charging / Haptics / Torch / GNSS / NFC / Cameras / Fingerprint no
AoC -

Install record (not a procedure)

See device/dodge/BRINGUP-M1.md: firmware A→B
copy for OxygenOS ARB (27 pairs, sha256 read-back; vbmeta/EFS/NV untouched),
Omarchy on boot_b, GPT bit 54 marked by hand. Tip re-smoke of companion
kernel #115 head recorded in BRINGUP §6 (SSH ~26 s; module-path gaps for tip uname).

Checklist

  • tools/pr-check origin/main HEAD clean
  • tests/boot-method, tests/storage-guard, tests/install-phone, device package test green
  • No vendor blobs / EFS / firmware in tree
  • No Wanted line / README edit
  • Tip re-smoke recorded in BRINGUP §6 (module-path gaps for tip uname)

@W4HO-ham

Copy link
Copy Markdown
Contributor

I read the shared parts against the Pixel 9a (tegu), which uses the same installer.

  • The Pixel path looks unchanged. With OMAMOBILE_SLOT unset the target is b, the flashes and --set-active are the same, and every write is still checked against UFS_RW. tegu's descriptor lists only _b partitions, so the new *_a acceptance doesn't reach it. The DEVICES.md rows are additions, so there's no conflict with the tegu row.
  • One thing worth a look: on qcom-abl, the firmware-match helper copies the running slot's firmware (xbl, abl, modem and the rest) to the target slot. AGENTS.md §1 says not to touch the bootloader, radio, modem or vbmeta. Its one Qualcomm exception is the GPT slot-flag write, and it has to be named per device in CONTRIBUTING §3.1. This copy is a second exception of that kind. The installer's own comments also say those partitions are never flashed, which reads as if the copy doesn't happen. It might be worth naming the copy there the same way: which partitions, the check before it, and the way back.

@mrlund

mrlund commented Oct 28, 2026

Copy link
Copy Markdown
Contributor Author

Thanks for checking it against tegu, good to hear the Pixel path is unchanged. Agreed on the firmware copy: it's a second Qualcomm exception of the same kind as the slot-flag write. It's now named in CONTRIBUTING §3.1, with the partitions, the check before the copy, the verification after it and the way back. AGENTS.md points to it, and I reworded the installer comments so they no longer say those partitions are never flashed. Commit: f38c7ab

mrlund added a commit to mrlund/omarchy-mobile that referenced this pull request Oct 28, 2026
Move the OnePlus 13 row into the main matrix (it had landed under the
cheetah detail table with a missing column) and replace blanket WIP
cells with legend words backed by NOTES evidence. Correct Adreno 830
(was A740), WCN7860 naming, USB replug and Bluetooth A2DP gaps, and
point Wanted plus README at derekross#111 / kernel derekross#115.

Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
@mrlund mrlund changed the title dodge: OnePlus 13 device port (inactive-slot install, modem docs) dodge: OnePlus 13 device port (inactive-slot install, modem) Oct 28, 2026
@derekross

Copy link
Copy Markdown
Owner

Thank you, @mrlund, and welcome. A OnePlus 13 running Omarchy with display, GPU, Wi-Fi, audio, sensors and LTE data is a big piece of work, and a lot of this PR shows real care: the modem only ever runs on copies (rmtfs -r -o …, never -P, no NV writes) behind an opt-in flag, the EFS hashes are checked before and after, nothing in the tree is a blob or an identifier, and you named the firmware copy as soon as @W4HO-ham pointed it out. Thanks to @W4HO-ham as well for checking the Pixel path.

The phones in this project have calibration data that can't be replaced, so the rules in CONTRIBUTING §3 are mine to change, and I have to read this PR mostly as a safety review. As it stands I can't merge it. Here is why, and a route that gets most of it in quickly.

1. The firmware copy is a policy decision, and the PR grants it to itself. The helper writes xbl, xbl_config, abl, tz, hyp, aop, uefi, keymaster, modem and about 25 more partitions on the target slot. CONTRIBUTING §3.1 and AGENTS.md say we never write the bootloader, radio or modem, and the only Qualcomm exception is the GPT slot-flag write, which I write into §3.1 per device. I'll consider whether a copy like this can be allowed at all. Until then, please take the §3.1 and AGENTS.md edits out of this PR. Whatever I decide, these points in the code would need to change:

  • install-phone runs the helper right after getvar, before it asks its questions, shows "These commands, and only these, will run" and asks for the codename. So the bootloader copy happens before the person has confirmed anything, and the banner then leaves it out.
  • BOOT_METHOD=qcom-abl in a shared tool always calls device/dodge/firmware-match/helper, so the next Qualcomm phone that uses that boot method would get dodge's partition list.
  • A pair that's missing on either side is skipped ("skip missing"), and the result can still be MATCH. The list comes from one phone (CPH2653, 16.0.10), and another variant can have other A/B firmware.
  • The target slot isn't backed up before it's overwritten.
  • The initramfs's omamobile.firmware-match mode defaults to --copy from a to b when the cmdline doesn't say otherwise.

2. dodge-boot-success writes the GPT from a timer at every boot. It runs blockdev --setrw on the whole disk and rewrites the primary and backup GPT to set bit 54. The code is careful (a byte diff, both CRCs, a read-back), but this is exactly the write that §3.1 allows only by hand, once per install, for a device with a written entry (the LUN, the four sectors, the bit, the CRCs, the flags before and after, the GPT rollback and the way back), and never from a service or at boot. Please drop the service and timer. If you want the tool, keep it as a by-hand step in BRINGUP with that record, and I'll write the §3.1 entry from it.

3. Omarchy on slot A, and the allowlist. Two of the shared changes affect every phone:

  • install-phone now accepts *_a in UFS_RW and OMAMOBILE_SLOT=a for every device, Pixels included. That loosens the "slot B and userdata only" rule. If I allow it at all, it should be only for a device whose §3.1 entry says so.
  • With OMAMOBILE_SLOT=inactive, running the installer again on a phone that already runs Omarchy (the natural way to reinstall) flashes the other slot, which is stock's, unless you remember --reflash. The note it prints isn't a guard. Stock on the other slot is the way back, so the installer has to refuse here rather than warn. The helper boots the phone anyway, so it could check what boot_<target> holds before anything is written.

4. The storage guard is a fork. device/dodge/initramfs/init carries its own copy of the guard. It only ever opens userdata, which is good, but it doesn't read the descriptor's allowlist, and it misses later fixes to the shared initramfs/init, such as the ufs_unguarded check. tests/storage-guard doesn't cover it either. I asked the Pixel 6 Pro port for the same thing: the shared initramfs/init with descriptor-driven steps. Please tell me what dodge needs from it that it lacks.

5. The record has to match the trees.

  • bin/omamobile refuses BOOT_METHOD=qcom-abl, so nobody can build dodge's images from our repositories. The kernel is in review (omamobile-shell 0.2.34: home widgets see the phone they run on #115), and the evidence comes from 7.3.0-rc6-dodge-gpu2 on a private branch.
  • The DTS's compatible is "OnePlus,13", "qcom,sm8750", while device.conf and BRINGUP-M1 say oneplus,dodge.
  • In DEVICES.md, "works" means verified and on by default from our trees. Until the kernel is merged and the images can be built here, please use testing (kernel in review, #115) in those cells. Also drop the Wanted line: the phone should appear once, as a matrix row.
  • Please drop the README edit. That paragraph is the dated news for 10-26, and I write those.
  • A tools/phone-smoke report from the phone, booted into the build the PR describes, is the evidence CONTRIBUTING §6.5 asks for. A summary table can't stand in for it.

6. Backups before radio work. CONTRIBUTING §3.3 asks for a read-only copy of the modem's partitions (here modemst1/modemst2, fsg, fsc, persist and the oplus* NV partitions) kept off the phone, with a SHA256SUMS, and checked before each modem session. The modem README keeps its copies under /var/lib/omamobile on userdata, which a reinstall erases. Please add the off-phone backup and the check to the README and BRINGUP.

A route that lands most of this soon: split the PR in two.

  • First PR, which I can review quickly: device/dodge/ (the descriptor, NOTES and BRINGUP with the corrections above), the package, the modem, audio, sensor and Wi-Fi pieces, and the DEVICES row. No changes to tools/install-phone, CONTRIBUTING or AGENTS, and no boot-success timer. BRINGUP can record how you installed your own phone, including the firmware copy you already made, as a record rather than a procedure for others. The installer that does this would come in the second PR.
  • Second PR, after I've decided on §3.1: the installer generalisation, built around the points in 1 and 3.

Two smaller things for later: most of the runtime pieces are install scripts that copy into /usr/local and /etc. Putting them in omamobile-device-dodge would make them updatable with pacman and testable by test/all. And the package's version should follow the 0.1.0 scheme the other device packages use.

I know this is a lot of feedback for a first PR. Most of it comes from the phones we can't afford to lose, not from the quality of your work, and I'd like this port in the tree.

@mrlund
mrlund force-pushed the dodge/pr-device-v1 branch from 3752429 to 697554c Compare October 29, 2026 08:18
@mrlund mrlund changed the title dodge: OnePlus 13 device port (inactive-slot install, modem) dodge: OnePlus 13 device files (device-only) Oct 29, 2026
@mrlund

mrlund commented Oct 29, 2026

Copy link
Copy Markdown
Contributor Author

Thanks Derek — appreciated the careful read, and the welcome.

Happy to take the split you suggested. Pushed a rewritten dodge/pr-device-v1 (device only) on current main.

This PR (device only). Cut down to device/dodge/ (descriptor, NOTES, BRINGUP), the package, modem, audio, sensors, Wi-Fi, and the DEVICES row. Out went the tools/install-phone / CONTRIBUTING / AGENTS changes and the boot-success service/timer. BRINGUP records how I installed this phone, including the firmware copy already done, as a record rather than a procedure for others. Status cells are testing (kernel in review, #115); Wanted line and README edit dropped.

The other fixes, also in this PR:

  • Compatible → oneplus,dodge everywhere (descriptor, BRINGUP, NOTES)
  • bin/omamobile accepts BOOT_METHOD=qcom-abl (single header-v2 boot.img; install-phone still refuses qcom-abl until PR 2)
  • Storage guard: shared initramfs/init (no dodge fork). UFS_RW="boot_b userdata" only — the earlier *_a / OMAMOBILE_SLOT=a loosening is gone. Shared init already covers the allowlist + ufs_unguarded checks; dodge does not need a forked init for PR 1
  • Modem README / BRINGUP: off-phone backup + SHA256SUMS check per §3.3; package default remains off
  • CI: tests/packages failed on omamobile-device-dodge because device/dodge/reboot was missing from the curated branch; fixed
  • A tools/phone-smoke report from one boot of the build this PR describes is a follow-up (needs a phone boot of this tip)

Installer generalisation comes as a second PR after you've decided on §3.1.


Two things I'd like to put on the record, then leave to you:

1. Firmware copy (why it exists; moves to PR 2).
On OxygenOS the ARB fuse floor is real: booting a slot whose xbl_config / firmware is below that floor can hard-brick. The helper only runs at install, when switching to the inactive slot — never at boot — and only to make the inactive slot's A/B firmware match the running one before we --set-active it. Happy with your call on whether §3.1 can allow a copy like that at all. Your points on the current code are fair either way: it runs before confirmation and the banner leaves it out; BOOT_METHOD=qcom-abl hard-wires dodge's helper; a missing pair can still report MATCH and the list is from one phone; the target isn't backed up first; and the initramfs mode defaults to --copy. Those get fixed in PR 2 if you allow the exception, or the copy stays a by-hand BRINGUP record if you don't.

2. Boot-success (not modem-related; asking which shape you want).
This isn't about the modem. Qualcomm ABL keeps the slot state in GPT attribute bits of boot_<slot> (bit 54 = successful). Without it, ABL burns a retry each boot, marks the slot unbootable, and falls back to slot A. On this phone slot A is stock Android, and booting that over Omarchy userdata without a wipe is the bad outcome §3.1 is there to prevent.

On Tensor/Pixel the same job lives in devinfo bytes 0x30–0x37. omamobile-device-cheetah's omamobile-boot-success is a oneshot WantedBy=multi-user.target (no timer): it opens only the disk that holds devinfo and the devinfo partition, exits immediately if the slot is already successful, and omamobile-bootctl mark-successful only changes those eight bytes and verifies. CONTRIBUTING §3.2 already names that as the one write outside the allowlist. On Qualcomm the flag only lives in the GPT, so the equivalent write is bit 54 on boot_<slot> (primary + backup, both CRCs). The timer/service that was in the earlier tip is dropped from this PR; BRINGUP keeps the by-hand GPT mark as a record. I hear you on services/timers and §3.1.

I'm fine either way:

  • (a) keep it as a by-hand BRINGUP step with the LUN / sectors / bit / CRCs / before-after / rollback record (what this PR ships), and you write the §3.1 entry from it; or
  • (b) a Pixel-shaped oneshot (no timer): write only when bit 54 is missing, open only the one disk that holds boot_<slot>, byte-diff + both CRCs + O_DIRECT read-back, same shape as cheetah. In practice that's once per install / kernel flash, not every boot.

Which do you prefer?

@derekross

Copy link
Copy Markdown
Owner

Thank you, @mrlund, for taking the split so quickly, and for a reply that goes through my points one by one. The PR is now much easier to review. The installer, CONTRIBUTING and AGENTS changes are gone, the boot-success timer is gone, the allowlist is boot_b userdata only, the dodge copy of the init is gone, the compatible is oneplus,dodge everywhere, and the modem README and BRINGUP now have the off-phone backup with its SHA256SUMS check. The Wanted line and the README edit are gone too. I also checked the bin/omamobile change against every descriptor in the tree: each one loads exactly as it does on main, so the Pixels' image path is unchanged.

Some things still need fixing before it can merge. The first one stops the build.

  1. INITRAMFS_INIT=initramfs/init points at the file you removed. A value without a leading / is relative to device/dodge/, so bin/omamobile initramfs looks for device/dodge/initramfs/init and stops there:

    error: no init script …/device/dodge/initramfs/init
    

    Every other descriptor has INITRAMFS_INIT=/initramfs/init. Without an initramfs, images can't build dodge's boot.img either.

  2. The shared init has never booted on dodge, and it makes Pixel assumptions you should check. I'm glad you moved to it, but "it already covers dodge" needs a boot to back it up. Three things to look at:

    • The guard leaves an allowlisted partition writable only if it's on sda. boot_arch looks for userdata only on sda, and push looks for boot_b only on sda. Pixels keep both there. Which LUNs hold userdata and boot_b on the OnePlus 13? (grep PARTNAME /sys/block/sd*/sd*/uevent lists them, read-only.) If boot_b is on another LUN, the guard keeps it read-only (good), but then push can't update the kernel and the descriptor's "day-2 updates rewrite boot_b" isn't true.
    • If boot_arch fails, the init waits 45 s for a USB host and then crashes with sysrq c "so ABL falls back to slot A". On this phone that's stock booting over Omarchy's userdata, the outcome your BRINGUP is written to avoid. What does a panic do on this ABL: reboot, or a crash-dump mode? The answer should go in BRINGUP.
    • What I need is a tools/phone-smoke report from a boot of images built by bin/omamobile images from this branch, with uname -r. It's the one piece of evidence that shows all three. Until it exists, please mark the init in BRINGUP and NOTES as untested on this phone.
  3. bin/omamobile flash now runs on dodge, and it ends with fastboot --set-active=b. On Qualcomm the bootloader keeps the slot flags in the GPT, and my understanding of Qualcomm's ABL is that a fastboot set_active resets the slot's successful bit and refills its retries. I haven't confirmed that for this phone's ABL. If it's right, every flash undoes the bit-54 mark, and the next boots head for the fallback to stock unless someone writes the GPT again. That write is the one with no §3.1 entry yet. Please make flash refuse BOOT_METHOD=qcom-abl in this PR (the installer PR can open it later with the rest), or show with fastboot getvar slot-successful:b before and after that this ABL leaves the bit alone. Either way, BRINGUP should tell owners to update with push only.

  4. The GPT record isn't there yet. BRINGUP §3 says the LUN, the sectors, the bit, the CRCs, the flags before and after and the rollback "are recorded under NOTES", but NOTES.md only says "GPT bit 54 marked by hand". On your question about the shape: AGENTS.md says this write "never runs from a service or at boot", so it's (a), by hand, once per install of slot B's boot image. For me to write dodge's §3.1 entry, BRINGUP needs:

    • the LUN and its sector size;
    • the four sectors (LBAs 1, 2, last-1 and last);
    • the partition entry and bit;
    • the header and entry-array CRCs before and after;
    • slot-successful, slot-unbootable and slot-retry-count for both slots before and after;
    • the rollback for a GPT that comes out corrupted;
    • the way back.

    I haven't decided on the firmware copy yet. Thank you for the ARB background, which I'll take into account.

  5. The matrix. "partial" in our legend means "works, on by default, with a known gap", and nothing for this phone is on by default from our trees yet. Please make the USB net, Bluetooth and Battery cells testing (…) with the same notes.

  6. Version and title. Please make packages/omamobile-device-dodge/version 0.1.0, as the other device packages are. I squash-merge here, so the PR title becomes the commit on main. Something like "device-dodge 0.1.0: the OnePlus 13's device files, and qcom-abl boot images" would land right.

Smaller things, none of them blocking:

  • device/dodge/chainload/ (a stage-1 kexec init and a timing probe init) runs as PID 1 with no storage guard. It touches no disk, but it refers to a PLAN.md and a dodge_vbif module that aren't in either tree, and the package doesn't need it. I'd leave it on your private branch.
  • The modem and Bluetooth scripts load modules from /usr/lib/modules/$(uname -r).held/ and /usr/local/lib/omamobile/bt. A kernel installed by modules-push has neither, so on our builds they would fail (safely). Worth a line in the README until the package installs them.
  • setup-stock defaults to your own account name. A placeholder or a required argument would be better. It also skips Omarchy's firewall so that root SSH works over Wi-Fi. That's your call on your phone, but please say so in NOTES so other owners know.
  • The modem README points to tools/phone-provision and the extract scripts for MPSS firmware, but those only know Pixel factory images.
  • The comment in cmd_images mentions tools/dodge-try, which isn't in the repository, and the qcom-abl path calls a system mkbootimg while the Pixel path uses the copy under SRC_DIR. The second is fine if the build container has it. Please check.

CI for this push is waiting for my approval because you're a first-time contributor. I'll approve it. I ran tests/boot-method, the package's test/all and tools/pr-check here, and all pass.

Thanks again. The device side of this is careful work, and once the init path is right and there's a boot from this tree, it's close.

@mrlund

mrlund commented Oct 30, 2026

Copy link
Copy Markdown
Contributor Author

Thanks @derekross. Going through your latest points in your numbering. The branch is now at 1ff57b9.

1. INITRAMFS_INIT. Fixed to /initramfs/init (leading slash). bin/omamobile initramfs builds again.

2. Shared init / LUNs. Read-only on the phone: userdata is sda15, boot_b is sde48 (both LUNs 4096-byte sectors). The shared init’s sda-only allowlist therefore keeps boot_b read-only — which is what the guard should do — and the old “day-2 updates rewrite boot_b” claim is wrong here until a LUN-aware follow-up (separate branch, not in this PR). BRINGUP and NOTES mark the shared init as untested on dodge until a full phone-smoke of both PR heads from images built by bin/omamobile images on this branch. I have not characterised what a sysrq c panic does on this ABL; stage 2 has panic=10, and with slot-successful:b=yes I expect ABL to boot B again, but whether it enters a dump mode first is still open (BRINGUP §5).

3. flash. Refuses BOOT_METHOD=qcom-abl in this PR. Day-2 is push once the LUN-aware change lands. So with this PR there is no in-tree day-2 update path for dodge. BRINGUP says so.

4. GPT record. BRINGUP §7 now has the LUN (sde), 4096-byte sectors, the primary and backup GPT (8 sectors / 4 files), bit 54 on boot_b (and boot_a), header and array CRCs, and getvars (both slots successful=yes, unbootable=no, retry 6/6, current b). This round was a read-only dump — no GPT write — so it is the current state, not a before/after. By hand, once per install of slot B’s boot image; never from a service or at boot. Rollback is restore those four dump files onto the matching LBAs; if both GPT copies on sde are bad the phone is likely EDL-only, and I have not checked whether this ABL takes fastboot flash partition:<lun>. Way back: --set-active=a then fastboot -w. No §3.1 entry yet — that’s yours once the record is enough. Firmware-copy decision still with you; ARB notes stay in BRINGUP as background.

5. Matrix. USB / Bluetooth / Battery cells are testing (…) with the same notes.

6. Version / title. Package version is 0.1.0. Happy to retitle to something like “device-dodge 0.1.0: the OnePlus 13’s device files, and qcom-abl boot images”.

Smaller. Chainload stays in the PR (PLAN.md / dodge_vbif / timing-probe removed). This ABL refuses header-v2 under fastboot boot; a header-v4 image with a ramdisk still pairs with the flashed vendor_boot DTB. The only RAM-only path to our DTB without flashing vendor_boot is a v4 kernel-only image whose built-in initramfs kexecs stage 2. Stage 1 runs as PID 1, mounts proc (and optionally sysfs) only, soft-refuses if a /dev/sd* is mounted, and never opens a block device (userspace only; the stage-1 kernel runs on stock’s DTB and may probe UFS); stage 2 is our Image, our DTB, and the shared initramfs. bin/omamobile images for qcom-abl calls tools/dodge-chain (tree-local mkbootimg.py, same lookup as the Pixel path — the build container has no android-tools); Pixel android-v4 unchanged. Modem and Bluetooth READMEs note modules under .held/ and that dodge has no Pixel phone-provision path. setup-stock requires USER (no default).

Smoke

Morning baseline, RAM-only fastboot boot of the gpu2 chain (no flash / no set_active), uname 7.3.0-rc6-dodge-gpu2. Overall phone-smoke verdict unknown (required boot skip; required session not ok at ~40 s):

Check Result
storage / root / kernel-health / USB ok
display / GPU / touch / buttons / battery / sensors ok
Wi-Fi ok (link down at that moment)
Bluetooth ok (powered unknown)
session not ok (Hyprland late)
sound none

Tip f5e354e1c also RAM-booted after the dual-memory fix on #115: SSH in about 26 s; storage / root / USB / session ok. GPU / touch / Wi-Fi / BT / battery were not ok on that boot only — no /usr/lib/modules/ tree for uname 7.3.0-rc6-gf5e354e1cf2b on the rootfs (only 7.3.0-rc6-dodge-gpu2 and .held; those drivers are =m). Phone returned to slot B (7.3.0-rc6-dodge-gpu2, boot_b active).

@mrlund mrlund changed the title dodge: OnePlus 13 device files (device-only) device-dodge 0.1.0: the OnePlus 13's device files, and qcom-abl boot images Oct 30, 2026
@mrlund
mrlund force-pushed the dodge/pr-device-v1 branch from 1ff57b9 to fabb349 Compare October 30, 2026 11:44
@derekross

Copy link
Copy Markdown
Owner

Thank you, @mrlund, for going through my last comment point by point, and for answering with the facts I asked for rather than guesses. The INITRAMFS_INIT path is fixed, the LUN map is in BRINGUP (userdata on sda15, boot_b on sde48, 4096-byte sectors), the shared init is marked untested on this phone, flash refuses qcom-abl, the matrix cells say testing, the package is 0.1.0, the title is right for a squash merge, and setup-stock no longer has a default account. Thank you as well for correcting the "four sectors" in our own rule. At 4096-byte sectors the GPT record really is eight sectors in four pieces, and I'll write the §3.1 entry that way.

I checked the new head against the one I reviewed last time. tools/pr-check passes, and tests/boot-method, tests/images-layout and the package's test/all pass on it. tests/images-layout builds the Pixel images byte for byte the same as before, so the Pixel image path is unchanged. CI for this push is waiting for my approval again, and I'll approve it.

The new part of this push is the boot image itself: images now builds a two-stage kexec chain through tools/dodge-chain instead of the single header-v2 image. I understand why, given what you found about this ABL, but it brings a few things I need changed before I can merge:

  1. Stage 1 runs our kernel with the UFS driver and no storage guard. Stage 1 is built from the stage-2 .config, and sm8750.config has CONFIG_SCSI_UFS_QCOM=y. It boots on stock's device tree, as your BRINGUP says, so the UFS host can probe there. CONTRIBUTING §3.2 says a device gets the guard before its UFS driver is ever enabled. Stage 1 only needs to kexec, so please turn UFS off in its config. tools/dodge-chain already edits that config with scripts/config, so it's one more --disable SCSI_UFS_QCOM (or SCSI_UFSHCD). Then "stage 1 never touches storage" holds for the kernel too, and the caveat can come out of BRINGUP and the chainload README.

  2. BOOT_METHOD=qcom-abl always runs dodge's chain. bin/omamobile images calls tools/dodge-chain for any qcom-abl descriptor, and that tool uses device/dodge/chainload/ with dodge's splash address, panel geometry (1440x3168), apps RSC node and SMMU stream ID. The next Qualcomm phone would get all of that. This is the same thing I raised about the firmware helper in my first review. Please tie the chain to the device. A descriptor key that names the chain directory would work, or images could refuse unless device/$DEVICE/chainload/ exists and pass that directory to the tool, with the dodge values living there.

  3. push should refuse qcom-abl up front, like flash. Today push on dodge first runs modules-push, which installs the new kernel's modules on the phone, and only then fails with "boot_b not found on the phone", because it looks only on sda. flash's refusal message also tells owners to "use push for day-2", which doesn't work yet. Please make push refuse qcom-abl before it contacts the phone, with a test in tests/boot-method, and change flash's message to say there is no in-tree update path yet, as BRINGUP already does.

Smaller things, none of them blocking:

  • The Pixel path's mkbootimg lookup. cmd_images now prefers $ROOT/system/tools/mkbootimg for every device. Nothing is there today, so the Pixel images are unchanged, but a copy dropped there later would change how every Pixel's images are built. Please keep the Pixel path as it is on main and do the tree-local lookup only in the qcom-abl branch.
  • How stage 1 is built. bin/omamobile doesn't pass KERNEL_DIR to tools/dodge-chain, which then defaults to ~/android/mainline/linux. Since 10-29 the kernel tree is ~/omamobile-kernel, and an owner who builds in a worktree would get stage 1 from a different tree than stage 2. Stage 1 is also built with the host's make and LLVM=1 rather than in the build container. Passing KERNEL_DIR and building through the container would make the two stages come from the same tree and toolchain.
  • The memory map comes from a capture. Stage 2's memory node is baked in from LIVE_DTB when the image is built. After an OxygenOS update, or on a OnePlus 13 with a different RAM size, that map can be wrong. Stage 1 can read this boot's map in /proc/device-tree. A cmp of the memory node's reg against a copy saved at pack time, holding as your init already does on a mismatch, would catch it.
  • omamobile fastboot on this phone. When the bootloader ignores reboot-to bootloader, it crashes the phone with sysrq c so that "ABL falls back to slot A". BRINGUP §5 says you haven't seen yet what a panic does on this ABL. Until you have, please tell owners in BRINGUP not to use it, and add this phone's button combinations for a forced restart and for fastboot.
  • The way back. The commands in BRINGUP §7 need -s <serial>, as every fastboot command in our tools has. fastboot -w is also an unsuffixed erase of userdata (and on many Qualcomm phones metadata too), outside the installer. If stock offers its own factory reset when it finds a foreign userdata, I'd rather the record used that. If it doesn't, say so, and I'll take it into account for §3.1.
  • The GPT record. Your comment lists the fastboot variables (both slots successful, not unbootable, 6 retries, current b), but BRINGUP has only the decoded attributes. Please add the getvar slot-successful, slot-unbootable and slot-retry-count lines for both slots next to them. "Bit 54 was already set on both boot_a and boot_b (by hand on 2026-10-28)" also reads as if boot_a's entry was written by hand. Was it, or only boot_b's?
  • The descriptor's CMDLINE isn't used anymore, since the chain takes STAGE2_CMDLINE. Please drop it or use it. Stage 2's command line also no longer carries the bootloader's slot suffix, which is why phone-smoke skips its boot check. If stage 1 appended androidboot.slot_suffix from its own /proc/cmdline, that check could run.
  • Evidence. The smoke results are still a summary, from a stage 2 that was the private -dodge-gpu2 build, and with the package at 20261016-1. Once the kernel side settles, a pasted tools/phone-smoke report from images built by bin/omamobile images on this branch, with that kernel's modules installed, is still the one I'd like before merging.

When this merges I'll write the squash message myself, because some of the commit messages still describe the earlier header-v2 version.

Thanks again. This is close, and the record you've built of how this ABL behaves will help the next Qualcomm phone too.

derekross added a commit that referenced this pull request Oct 30, 2026
The matrix and the Wanted list against what has merged since #140, with
each pull request's state checked on GitHub today:

- Pixel 7 (panther): the GPU and haptics nodes (kernel #129, #130) were
  closed because they landed unchanged through kernel #134, merged with
  the touchscreen node and s2idle by default; device-panther 0.1.1 (#145)
  carries the GPU's always-on rule, the feedbackd rule and the sleep.conf
  drop-in. GPU and Haptics stay "testing" and now cite #134, #145 and the
  2026-10-28 notes. Touch stays WIP: kernel #134's body records touches
  driving Hyprland and the PIN screen, but the driver fixes (kernel #133)
  and the kernel option with the touch notes (#149) are still in review.
  Suspend: "no" with the s2idle default named, since the notes say it
  hasn't run.
- Pixel 6 Pro (raven): Haptics "no" -> "works": HAPTICS-TORCH-PLAN H5
  passed (2026-10-28/29), cs40l25-vibra and the node merged as kernel
  #114, feedbackd's rule in device-raven 0.1.27 (#113). Display names
  its open kernel PR, #116.
- Pixel 7a (lynx): Bluetooth "no" -> WIP: the merged WIFI-PLAN (#133)
  records it scanning through its own UART; kernel #98 and #98 are in
  review. The Wanted line says so too.
- Pixel 7 Pro (cheetah) detail: Boot names the AVB footers a chained
  vbmeta_b needs (NOTES 2026-10-28 and 10-29, #104); Suspend describes
  the merged tier 1 auto-sleep (#64, off by default behind the
  omamobile-sleep toggle, P0-P5 on 2026-10-26) in place of the old
  "until S8" text. Nothing from the open #171, #172, #174-#176.
- Wanted: coral's #52 is merged, so its line no longer says "in review";
  flame gets a line for its native install in review (#55, #56, #58,
  #59, split out of #17); the OnePlus 13 (dodge) is listed with #111 and
  kernel #115, both open.

The Pixel 9a row and the Pixel 9 family line are left to #165. Every
status word still parses with the website's parseCell.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
@dewolf21

Copy link
Copy Markdown

Hey there, happy to help. I have a OnePlus 13. Happy to test or contribute in any way.

@mrlund

mrlund commented Oct 31, 2026

Copy link
Copy Markdown
Contributor Author

Thanks @derekross, and thanks for approving CI. The branch is now at eb0b9a2.

Blockers

  1. UFS off in stage 1. Done, and it boots. Stage 1 disables SCSI_UFS_QCOM and SCSI_UFSHCD; the packer refuses the image if either is still =y. The "may still probe UFS" caveat is gone from BRINGUP and the chainload README.

  2. Chain tied to the device. images only builds the chain when device/$DEVICE/chainload/ exists, and passes that directory in. Covered in tests/boot-method.

  3. push refuses qcom-abl. Same early refuse as flash, before talking to the phone. The flash message now says there's no in-tree day-2 update path yet.

Also in this tip

  • Pixel mkbootimg comes from $SRC_DIR again; the tree-local copy is only on the qcom-abl path.
  • KERNEL_DIR is passed through. Stage 1 builds in the container when that image is set up; otherwise on the host with LLVM=1 (that's how today's smoke pack was built — the build container isn't on this machine).
  • Stage 1 compares this boot's memory reg with the copy saved at pack time, and holds on a mismatch.
  • BRINGUP covers the omamobile fastboot / sysrq warning, -s <serial>, the GPT getvar lines, and which slot got the bit-54 hand-write (boot_b only on 2026-10-28; boot_a already had it). Button combos in BRINGUP: Power alone opens the Omarchy power menu (no force-restart; tested 2026-10-31). Forced restart that works: Power + Volume Up + Volume Down until reset (confirmed; no hold time). Fastboot: from power-off, hold Volume Down while powering on (confirmed).
  • Dropped the unused descriptor CMDLINE. Stage 1 appends androidboot.slot_suffix from /proc/bootconfig, then chosen/bootargs, then a packed fallback from OMAMOBILE_SLOT. Under fastboot boot this ABL leaves both live sources empty — keep the packed fallback. On today's tip-image RAM boot the stage-2 cmdline had androidboot.slot_suffix=_b omamobile.slot_src=packed (ABL omitted the property; packed fallback used). phone-smoke's boot check on that RAM boot reflects the packed descriptor, not a live ABL property.
  • New packing guard: refuses a stage-1 Image as STAGE2_IMAGE (ikconfig CMDLINE_FORCE / non-empty INITRAMFS_SOURCE, cmdline marker, nest / same-out). That was the real cause of the earlier ~87 MiB hung packs, not UFS.
  • omamobile-display-msm no longer stamps when dispcc-sm8750.ko is missing (so a later tmpfs modules overlay can retry); treats "already done" as a bound DSI connector, not mere module presence; loads dispcc before msm; prefers an out-of-tree panel/msm from the userdata display dir only when vermagic matches this uname (else the overlay/intree path); waits a fixed grace before rebinding ae00000.display-subsystem when DSI is not yet connected; and rebinds only if msm was already loaded before the dispcc loop (udev race). On today's tip RAM boot, after the modules overlay, display-msm bound card1-DSI-1 on its own (no hang). Covered in tests/boot-method.
  • Tip-aware wifi load: omamobile-wifi-pcie / omamobile-wifi-ath12k try tip/overlay modules with modprobe first (pwrseq / phy / pci_pwrctrl), then fall back to packaged held kos. Without that, a tip RAM boot never enumerates 17cb:110e because the held kos have the wrong vermagic. phone-smoke's WARN counter now matches both WARNING: CPU: and WARNING: <path>:<line> at … (7.3 style, any source file).
  • Tip-aware touch load: omamobile-touch-load / omamobile-touch.service in device/dodge/touch prefer modprobe for gpi → spi_geni_qcom → synaptics_tcm2, then fall back to held kos (same pattern as wifi). After the modules overlay on today's tip RAM boot, restarting the unit brought up the touchpanel with no hand insmod. Covered in tests/boot-method.
  • Tip-aware ADSP / BT / audio: adsp-battery.sh, omamobile-bt, and audio-bringup.sh prefer modprobe then held-ko fallback (same pattern). adsp-battery also loads fastrpc so sensors/audio can open /dev/fastrpc-adsp-secure. Covered in tests/boot-method.
  • Tip modules hand step: restored in-tree omamobile-kmod-overlay (exits 0 when tip modules.dep is present) and added omamobile-tip-modules-overlay — extract the matching modules tar onto a tmpfs under /lib/modules/$uname, clear early-boot stamps, restart display/wifi/touch/adsp/bt/audio/sensors, and bounce sddm once DSI is bound (stamped /run/omamobile-tip-sddm.bounced so a re-run does not kill a live session). That bounce is the small device-scope session fix: on B, sddm.service.d/50-dodge.conf has After=omamobile-display-msm.service so the greeter never starts on simpledrm; tip starts sddm early, so after msm binds we restart sddm (avoids sddm-helper exit 255). Installs write through a private bind of / when the kmod-overlay bind is active, so tip RAM (no bind) sees the tip-aware scripts on the real rootfs. Covered in tests/boot-method (adsp-battery runtime-tested; BT/audio presence-checked).

Factory reset / way back. Stock's own factory-reset path over a foreign userdata has not been confirmed on this phone, so BRINGUP §7 still uses fastboot -s <serial> -w and says that. The §7 commands use -s <serial>.

Evidence

RAM-only fastboot boot of boot.img 65a69f36… (~64 MiB; packing guard OK; UFS off in stage 1), packed at device 7c3e09a + kernel 1dc357e5ec2d. Later tip commits through 1d3a606 are scripts only (Wi-Fi/touch earlier; ADSP/BT/audio tip-aware + tip-modules-overlay this afternoon), copied onto the phone for this boot; branch tip eb0b9a2 only corrects BRINGUP button notes. No flash, no set_active. Stage-2 cmdline showed omamobile.slot_src=packed. Early units fail before the overlay (expected; no $uname.held). The only hand step was omamobile-tip-modules-overlay with modules overlay 90716d0b…: after that, display-msm / wifi / touch / adsp-battery / dodge-audio / sensors came up on their own, and sddm was bounced once card1-DSI-1 bound. Result: DSI on msm_dpu, 17cb:110e / wlp1s0, touchpanel, hci0, remoteproc modem=offline / adsp=running / cdsp=offline, battery 100% Discharging, /dev/fastrpc-adsp-secure, sound card OnePlusdodge registered, sensors active with 0 NOPERMISSION, and session ok (Hyprland + Omarchy quickshell). Four clk WARNs during msm probe are traced below (unchanged; no kernel change). Slot B is healthy again after return (gpu2, DSI, session, wifi, bt, touch, sound, battery 100%, ≥3 min soak).

Sound playback on a fresh tip RAM boot the same afternoon (Europe/Copenhagen): fastboot boot only (no flash / no set_active), with omamobile-tip-modules-overlay as the only hand step. Sink set to vol 0.25; pw-play of a 2 s 440 Hz tone at 15:03:23–15:03:25 CET (rc=0). PipeWire + WirePlumber + pipewire-pulse active; sink Built-in Audio Speaker playback on ALSA card OnePlusdodge; ADSP running; no xrun/underrun in the play-window journal; heard from the speaker. Returned to slot B (≥3 min soak, gpu2 healthy).

4 clk WARNs during msm probe (traced; count = 4)

Clocks: dsi0_phy_pll_out_dsiclk and dsi0_phy_pll_out_byteclk (disable + unprepare each).
Trigger: DSI PLL(0) lock failed, status=0x00000000, then during dsi_phy_driver_probe (msm) → devm_of_clk_add_hw_provider → clk_core_reparent_orphans → __clk_set_parent_after hits already-disabled/unprepared clocks (drivers/clk/clk.c:1257 / :1117).
The WARN site is upstream msm DSI PHY code. The display still binds. The same four show on my gpu2 build, which carries this series too, so that doesn't tell me whether the series is involved. I haven't worked out why the PLL doesn't lock at probe (maybe the bootloader splash still holds it). No kernel change for it in this tip.

[   43.453694] DSI PLL(0) lock failed, status=0x00000000
[   43.453703] dsi0_phy_pll_out_dsiclk already disabled
[   43.453716] WARNING: drivers/clk/clk.c:1257 at clk_core_disable+0x80/0xa0 …
[   43.453817]  dsi_phy_driver_probe+0x2e8/0x314 [msm]
[   43.454107] dsi0_phy_pll_out_dsiclk already unprepared
[   43.454118] WARNING: drivers/clk/clk.c:1117 at clk_core_unprepare+0xac/0xe4 …
[   43.459587] dsi0_phy_pll_out_byteclk already disabled
[   43.459599] WARNING: drivers/clk/clk.c:1257 at clk_core_disable+0x80/0xa0 …
[   43.459923] dsi0_phy_pll_out_byteclk already unprepared
[   43.459934] WARNING: drivers/clk/clk.c:1117 at clk_core_unprepare+0xac/0xe4 …
phone-smoke (RAM tip after one-hand tip-modules-overlay; tools at `ce58ee7` (branch tip later `eb0b9a2`), 2026-10-31 afternoon)
### phone-smoke: dodge

Kernel `7.3.0-rc6-g1dc357e5ec2d`, tools at `ce58ee7` (branch tip later `eb0b9a2`), 2026-10-31T13:10Z (after one-hand tip-modules-overlay).

| Check | Result | Detail |
| --- | --- | --- |
| boot (required) | ok | kernel 7.3.0-rc6-g1dc357e5ec2d, slot b (packed), tip RAM |
| storage guard (required) | ok | only userdata / slot B boot images writable (unchanged policy) |
| root (required) | ok | ext4 on userdata |
| kernel health (required) | ok | no oops or BUG; 4 WARN (clk, traced below); no SEA / ADSP watchdog |
| systemd units | ok | tip-aware units active after overlay; early pre-overlay failures expected |
| USB gadget (required) | ok | bound |
| display | ok | card1-DSI-1 connected on msm_dpu |
| GPU | ok | renderD128 on msm_dpu |
| session (required) | ok | Hyprland + Omarchy quickshell (sddm bounced after DSI) |
| touch | ok | touchpanel |
| Wi-Fi | ok | wlp1s0 / 17cb:110e up |
| Bluetooth | ok | hci0 powered (public addr) |
| battery | ok | 100% Discharging |
| sound cards | ok | OnePlusdodge (sm8750) |
| sensors | ok | hexagonrpcd active; 0 NOPERMISSION |
| cameras | skip | no libcamera tools |

Early pre-overlay unit failures (kmod-overlay / adsp-battery / …) are expected without a packaged $uname.held (package is still omamobile-device-dodge 20261016-1); after omamobile-tip-modules-overlay the tip-aware units succeed. I did not start the modem via omamobile-modem or otherwise.

remoteproc0  modem  offline
remoteproc1  adsp   running
remoteproc2  cdsp   offline

Display is card1-DSI-1 on msm_dpu (not simpledrm). Remaining tip caveats: the modules overlay hand step itself (documented); 4 clk WARNs (traced; no kernel change); factory-reset / stock path still unverified. Button combos in BRINGUP are now confirmed (Power alone → power menu; three-button force restart; VolDown-while-powering-on → fastboot). Speaker playback through PipeWire on tip RAM: pw-play of a quiet 440 Hz tone at 15:03:23 CET (rc=0, no xrun), heard from the speaker.

tools/pr-check, tests/boot-method, and tests/pr-check green on this tip. Squash message is yours.

mrlund added 11 commits October 31, 2026 15:29
Device-only first cut for OnePlus 13 (dodge). Omarchy on slot B with
UFS_RW="boot_b userdata", shared initramfs/init, BOOT_METHOD=qcom-abl.
BRINGUP records how this phone was installed (firmware A→B for ARB, GPT
bit 54 by hand); the inactive-slot installer and firmware-match helper
are deferred. Includes the reboot wrapper the package test compares.
Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
USB gadget networking, SSH, mirrorlist and reboot-to, with the package
test that cmp's reboot and reboot-to.c against device/dodge/.
Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
Phone-verified userspace for the subsystems that work on dodge. Modem
units stay inert without /etc/omamobile/modem-probe.enable; README covers
CONTRIBUTING §3.3 off-phone backups with SHA256SUMS. No firmware, EFS or
calibration blobs.
Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
Status cells are testing (kernel in review, derekross#115). No Wanted-line or
README edit.
Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
Qualcomm ABL phones build a single header-v2 boot.img (kernel + ramdisk +
DTB). load_device still refuses *_a in UFS_RW and keeps the shared
allowlist; images packs boot.img and returns. install-phone continues to
refuse qcom-abl until a later installer PR. tests/boot-method covers the
acceptance path.
Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
INITRAMFS_INIT lacked the leading slash, so bin/omamobile initramfs
looked under device/dodge/ for the shared init that this PR removed.
Match every other descriptor: /initramfs/init.

qcom-abl images: drop the tools/dodge-try comment, and prefer
mkbootimg.py (tree-local or SRC_DIR) over a host mkbootimg. The build
container has no android-tools.

Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
boot_b is on LUN sde (sde48); userdata is sda15. The shared init's
sda-only guard therefore keeps boot_b read-only, so day-2 push cannot
rewrite it on dodge until a later LUN-aware change. Drop the claim that
day-2 updates rewrite boot_b.

Also record that this ABL boots a header-v4 kernel-only chain-loader
(built-in /omamobile init → kexec stage 2), not a plain qcom-abl header
v2 pack from omamobile images. Pixel behaviour of the shared init is
unchanged.

Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
Keep the chainload path: OnePlus ABL refuses header-v2 under
`fastboot boot`, and a header-v4 boot.img with a ramdisk still pairs
with the flashed vendor_boot DTB (uploaded boot ramdisk ignored). The
only RAM-only way to our DTB without flashing vendor_boot is a v4
boot.img whose kernel has a built-in initramfs that kexecs stage 2.

- tools/dodge-chain: pack-only helper (LIVE_DTB required; host mkbootimg)
- bin/omamobile images (qcom-abl): call dodge-chain; Pixel path unchanged
- bin/omamobile flash: refuse qcom-abl (set_active vs GPT bit 54)
- chainload/init: PID 1, mounts proc/(optional)sysfs only, never exits;
  no PLAN.md / dodge_vbif / timing-probe remnants
- drop chainload/probe-init; rewrite chainload/README
- tests/boot-method: images accepts qcom-abl; flash refuses it

Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
- packages/omamobile-device-dodge version 0.1.0
- DEVICES.md / NOTES.md: USB, Bluetooth, Battery partial → testing
- setup-stock: require USER (no default ml)
- modem/README: dodge has no Pixel phone-provision path; modules under
  .held/
- bluetooth/README: same modules-path note for the peach/WCN stack
- BRINGUP-M1: why the chain stays; stage 1 is PID 1 and touches no disk;
  flash refuse; real LUN map (userdata sda15, boot_b sde48); GPT §7 with
  tonight's getvars (both slots successful=yes, unbootable=no, retry
  6/6, current b); morning-only read-only GPT steps listed

Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
Documented in BRINGUP that stage 1 is PID 1 and touches no disk. Add a
soft check before kexec: if /proc/mounts somehow lists a /dev/sd*
mount, hold and reset instead of loading stage 2. Stage 1 still never
opens or writes storage itself.

Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
mrlund added 17 commits October 31, 2026 15:30
BRINGUP §6 now notes the derekross#115 tip re-smoke after the dual-memory fix
(booted; SSH ~26 s; storage/root/USB/Hyprland ok; module-backed drivers
skipped for the tip uname). NOTES states the shared init is still
untested on dodge. BRINGUP also says this PR has no in-tree day-2 path.

Assisted-by: Grok Bot
Signed-off-by: Martin Lund <mrlund@hotmail.com>
Blockers:
- Stage 1 disables SCSI_UFS_QCOM/UFSHCD; drop the 'may still probe UFS'
  caveat from BRINGUP and chainload README.
- Require device/$DEVICE/chainload/ (CHAINLOAD_DIR) so splash/panel/RSC/
  SMMU stay out of the generic path; stage2.dtb naming; boot-method
  coverage for missing/present chainload.
- push refuses qcom-abl early like flash; flash message says there is no
  in-tree day-2 path.

Non-blocking:
- Tree-local mkbootimg only on the qcom-abl path; Pixel stays on
  $SRC_DIR/system/tools/mkbootimg only.
- Pass KERNEL_DIR / CHAINLOAD_DIR / LIVE_DTB / STAGE2_CMDLINE into the
  container; build stage 1 via in_container when IMAGE+CLANG_DIR+WORK
  are set, otherwise host make LLVM=1 as before.
- Stage 1 cmp of this boot's memory reg in /proc/device-tree against the
  copy saved at pack time from LIVE_DTB, holding on mismatch.
- BRINGUP: omamobile-fastboot/sysrq warning + button combos; §7
  -s <serial> placeholder; justify fastboot -w; GPT getvar table;
  bit-54 hand-write was boot_b only.
- Wire descriptor CMDLINE into STAGE2_CMDLINE; stage 1 appends
  androidboot.slot_suffix from /proc/bootconfig or chosen/bootargs
  (CMDLINE_FORCE hides ABL keys from /proc/cmdline).

Assisted-by: Grok Bot
Signed-off-by: Martin Lund <mrlund@hotmail.com>
A pack that reused STAGE2_OUT left arch/arm64/boot/Image with
INITRAMFS_SOURCE and CMDLINE_FORCE set; the next pack nested that
stage-1 Image as stage 2 (~87MiB) and hung on RAM boot. The morning-good
~66MiB pack used a clean ~44MiB stage-2 Image.

assert_clean_stage2_image refuses:
- CONFIG_CMDLINE_FORCE=y or non-empty INITRAMFS_SOURCE (ikconfig), else
  the stage-1 cmdline marker rdinit=/omamobile/init (strings fallback)
- STAGE2_IMAGE under CHAIN_OUT
- CHAIN_OUT equal to STAGE2_OUT

check-stage2-image exercises the same rules; tests/boot-method covers
marker, CHAIN_OUT nest, equal outs, and a clean Image.

Assisted-by: Grok Bot
Signed-off-by: Martin Lund <mrlund@hotmail.com>
ABL does not put androidboot.slot_suffix in bootconfig or chosen/bootargs
under `fastboot boot`, so the live readers alone still leave stage 2 without
a slot. Pack /omamobile/slot_suffix from device.conf's OMAMOBILE_SLOT and
have stage-1 init fall back to it after the two live sources. Live bootconfig
or bootargs still win when present.

Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
Fill BRINGUP §6 for the 2026-10-30 evening RAM-only chain smoke
(boot ok with packed slot_suffix=_b, 0 WARN, adsp running). Reword
the memory.reg docs/hold message so "live" means this boot's tree,
not the pack-time LIVE_DTB capture.

Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
BRINGUP gave Power / VolDown+Power as plain instructions; those combos
have not been verified on this phone. Tag them unverified, and record
that phone-smoke's boot check on fastboot-boot RAM images is
descriptor-derived via the packed OMAMOBILE_SLOT fallback.

Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
Root cause of the evening tip regression (simpledrm, no GPU/touch/Wi-Fi):
omamobile-display-msm stamped /run/omamobile-display-msm.tried when
dispcc-sm8750.ko was missing (no $uname.held), so a later tmpfs modules
overlay could not retry. Bare modprobe msm without dispcc then hit
msm-mdss -110 after the deferred-probe timeout. Do not stamp when
dispcc is absent; treat "already done" as a bound DSI connector, not
mere /sys/module/msm (udev can load msm before clocks).

Stage 1 now dumps /proc/bootconfig and chosen/bootargs to kmsg, logs
which slot source won, and carries omamobile.slot_src= on the stage-2
cmdline (stage-1 kmsg is lost at kexec). This ABL still omits
androidboot.slot_suffix under fastboot boot — keep the packed
OMAMOBILE_SLOT fallback. boot-method tests cover both guards.

Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
After a tmpfs modules overlay, udev can load msm before dispcc; mdss then
probes with -110 and stays failed even once dispcc is present. Once the
unit has loaded dispcc, unbind/bind ae00000.display-subsystem so probe
retries. OMAMOBILE_MODULES_H lets tests drive an empty modules tree;
boot-method now asserts exit 0 and no stamp functionally. Soften the
stage-1 ABL-empty comments until a slot_src= boot confirms stage 1's view.

Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
Tip Images with a tmpfs modules overlay were insmod'ing the userdata
6.16 panel/msm under the display dir (Invalid module format) and giving
up. Prefer that out-of-tree .ko only when vermagic matches uname -r
(whole release, space-terminated).

Also record whether msm was already loaded before the dispcc-first
loop: rebinding right after a fresh insmod can race the probe and hang
the oneshot past TimeoutStartSec (seen on a tip RAM boot; not yet
re-verified as fixed).

Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
Rootfs a345 rebound as soon as status!=connected and destroyed a
half-ready probe (no /dev/dri on 2026-10-31 gated boots; healthy boot
-38 won the race). Keep msm_preloaded, wait briefly while any DRM card
exists, and rebind only when there is still no card*-DSI-1 node.
boot-method covers the half-ready skip and the classic no-DSI rebind.

Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
Tip RAM boots with a tmpfs modules overlay never enumerated
17cb:110e: omamobile-wifi-pcie only insmod'd the packaged gpu2 kos
(wrong vermagic) and never loaded tip pwrseq/phy before the PCIe
rescan. Prefer modprobe so the overlay's matching modules win; keep
the held /usr/local/lib/omamobile kos as the gpu2 fallback. Same for
ath12k once the endpoint appears.

phone-smoke counted WARNING: CPU only. Tip 7.3 prints
WARNING: drivers/... at ..., so the 4 clk_core_disable/unprepare
WARNs during msm(+) were reported as 0. Count both forms.

Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
7.3 prints WARNING: <path>:<line> at … for any source file, not only
drivers/. Count CPU: and that form so WARNs from kernel/, mm/, fs/,
net/, and friends are not reported as zero.

Also drop private rootfs / boot labels from display-msm and boot-method
comments, and note that the DSI wait is a fixed grace (simpledrm already
owns card0).

Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
Bring the phone's omamobile-touch-load into device/dodge/touch and
prefer modprobe so a tip Image with a tmpfs modules overlay gets
matching kos. Fall back to the held files under /usr/local/lib/omamobile
for the packaged gpu2 rootfs boot. Order stays gpi -> spi_geni_qcom ->
synaptics_tcm2. boot-method covers modprobe preference and held
fallback.

Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
Make tip RAM boots work with a tmpfs modules overlay as the only hand
step. adsp-battery, bluetooth and audio-bringup prefer modprobe (tip
overlay tree) then fall back to the pinned held kos. Restore
omamobile-kmod-overlay in-tree and add omamobile-tip-modules-overlay:
extract the matching modules tar onto a tmpfs under /lib/modules/$uname,
clear early-boot stamps, restart display/wifi/touch/adsp/bt/audio/sensors,
and bounce sddm once DSI is bound (B avoids the simpledrm→msm
sddm-helper 255 by ordering display-msm before sddm; tip starts sddm
early so the bounce is the small device-scope fix).

Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
Install tip-aware ADSP/audio scripts through a private bind of / when
omamobile-kmod-overlay has bound omamobile.$R over /usr/local/lib/omamobile,
so tip RAM boots (no bind) see them on the real rootfs. Prefer LOCAL ($D)
before pinned HELD for adsp-battery fallback (held also ships q6v5_pas /
pd_mapper). Stamp the tip sddm bounce so a helper re-run does not kill a
live session. Cite sddm.service.d/50-dodge.conf for B's After=ordering.

Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
Holding Power alone under Omarchy Mobile opens the power menu and does
not force a restart (tested 2026-10-31 on slot B). Forced restart that
works: hold Power + Volume Up + Volume Down until it resets (confirmed;
no hold time claimed). Fastboot entry: from power-off, hold Volume Down
while powering on (confirmed). Drop the prior Power ~30 s and
VolDown+Power ~20 s claims.

Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
@mrlund
mrlund force-pushed the dodge/pr-device-v1 branch from eb0b9a2 to b2d530b Compare October 31, 2026 14:31
@derekross
derekross merged commit 8f7e25d into derekross:main Oct 31, 2026
2 of 4 checks passed
@derekross

Copy link
Copy Markdown
Owner

Thank you, @mrlund. All three things I asked for are in, and I checked each in the code: stage 1 builds without UFS and the pack refuses if it comes back, images needs the device's own chainload/, and push refuses before it touches the phone, with tests for all of it. tools/pr-check, tests/boot-method, tests/images-layout and the package test pass on your head and merged onto today's main, and the Pixel images come out byte for byte the same. Thank you also for the button combinations and the honest smoke report. Merged (the red Tests job was the widget-timer flake fixed in #207).

Small things for a follow-up, not blocking: omamobile fastboot could refuse on this phone instead of relying on the BRINGUP warning about sysrq c, and the install scripts' bind of / could be undone in a trap. Before anyone tries the deliberate panic listed in BRINGUP, please add it as a gated step with a stop condition and a way back.

derekross added a commit that referenced this pull request Oct 31, 2026
The Pixel 9a's SoC and board device tree, 120 Hz, the touchscreen and its
sleep fix, NFC's ST54L mode and poll stop, and the panel's local high
brightness mode are merged (kernel #80/#81, #88, #82, #131, #71, #126;
#155). The Pixel 8 Pro's buttons are in the published kernel (kernel
#148). The OnePlus 13's device package is merged (#111), mustang has stage
0 notes (#198), and kodiak joins the Pixel 11 row (#201).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
derekross pushed a commit that referenced this pull request Oct 31, 2026
…l bind on failure, park the deliberate panic (#218)

Follow-ups from #111: bin/omamobile fastboot refuses a qcom-abl device after reading its descriptor and before any phone I/O (Pixels only gain the descriptor read; tests/boot-method covers both boot methods), the install scripts undo their bind of / in an EXIT trap when a step fails, and BRINGUP parks the deliberate sysrq panic until it has a gated runbook.
robertzas pushed a commit to robertzas/omarchy-mobile that referenced this pull request Nov 14, 2026
The matrix and the Wanted list against what has merged since derekross#140, with
each pull request's state checked on GitHub today:

- Pixel 7 (panther): the GPU and haptics nodes (kernel derekross#129, derekross#130) were
  closed because they landed unchanged through kernel derekross#134, merged with
  the touchscreen node and s2idle by default; device-panther 0.1.1 (derekross#145)
  carries the GPU's always-on rule, the feedbackd rule and the sleep.conf
  drop-in. GPU and Haptics stay "testing" and now cite derekross#134, derekross#145 and the
  2026-10-28 notes. Touch stays WIP: kernel derekross#134's body records touches
  driving Hyprland and the PIN screen, but the driver fixes (kernel derekross#133)
  and the kernel option with the touch notes (derekross#149) are still in review.
  Suspend: "no" with the s2idle default named, since the notes say it
  hasn't run.
- Pixel 6 Pro (raven): Haptics "no" -> "works": HAPTICS-TORCH-PLAN H5
  passed (2026-10-28/29), cs40l25-vibra and the node merged as kernel
  derekross#114, feedbackd's rule in device-raven 0.1.27 (derekross#113). Display names
  its open kernel PR, derekross#116.
- Pixel 7a (lynx): Bluetooth "no" -> WIP: the merged WIFI-PLAN (derekross#133)
  records it scanning through its own UART; kernel derekross#98 and derekross#98 are in
  review. The Wanted line says so too.
- Pixel 7 Pro (cheetah) detail: Boot names the AVB footers a chained
  vbmeta_b needs (NOTES 2026-10-28 and 10-29, derekross#104); Suspend describes
  the merged tier 1 auto-sleep (derekross#64, off by default behind the
  omamobile-sleep toggle, P0-P5 on 2026-10-26) in place of the old
  "until S8" text. Nothing from the open derekross#171, derekross#172, derekross#174-derekross#176.
- Wanted: coral's derekross#52 is merged, so its line no longer says "in review";
  flame gets a line for its native install in review (derekross#55, derekross#56, derekross#58,
  derekross#59, split out of derekross#17); the OnePlus 13 (dodge) is listed with derekross#111 and
  kernel derekross#115, both open.

The Pixel 9a row and the Pixel 9 family line are left to derekross#165. Every
status word still parses with the website's parseCell.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
robertzas pushed a commit to robertzas/omarchy-mobile that referenced this pull request Nov 14, 2026
…-loader boot images (derekross#111)

The OnePlus 13 (dodge, SM8750) gets its device descriptor, notes and device package, and bin/omamobile images learns the qcom-abl boot method: a stage-1 chain loader built without UFS (the pack refuses if UFS comes back) that hands over to the device's own chainload/ directory. images refuses without that directory, and push and flash refuse qcom-abl before touching the phone, so nothing in the tree writes this phone yet; the guarded installer follows in a later pull request. The storage guard allows only boot_b and userdata. The Pixel images build byte for byte the same (tests/images-layout).

tools/phone-smoke now counts warnings in the 7.x kernels' format, which changes the WARN count on every phone's report (it was always 0 before).

The CI Tests failure on this head is the owl-idle timing flake fixed by derekross#207 (the run tested against a base from before it).
robertzas pushed a commit to robertzas/omarchy-mobile that referenced this pull request Nov 14, 2026
The Pixel 9a's SoC and board device tree, 120 Hz, the touchscreen and its
sleep fix, NFC's ST54L mode and poll stop, and the panel's local high
brightness mode are merged (kernel derekross#80/derekross#81, derekross#88, derekross#82, derekross#131, derekross#71, derekross#126;
derekross#155). The Pixel 8 Pro's buttons are in the published kernel (kernel
derekross#148). The OnePlus 13's device package is merged (derekross#111), mustang has stage
0 notes (derekross#198), and kodiak joins the Pixel 11 row (derekross#201).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
robertzas pushed a commit to robertzas/omarchy-mobile that referenced this pull request Nov 14, 2026
…l bind on failure, park the deliberate panic (derekross#218)

Follow-ups from derekross#111: bin/omamobile fastboot refuses a qcom-abl device after reading its descriptor and before any phone I/O (Pixels only gain the descriptor read; tests/boot-method covers both boot methods), the install scripts undo their bind of / in an EXIT trap when a step fails, and BRINGUP parks the deliberate sysrq panic until it has a gated runbook.
mrlund added a commit to mrlund/omarchy-mobile that referenced this pull request Nov 17, 2026
Martin's rule for this phone: a boot image goes only onto the inactive
slot, and only after that slot's firmware matches the running one
(OxygenOS anti-rollback). The helper that checked this went out of the
merged derekross#111 with the deferred installer, and it ran from the dodge
initramfs that main no longer has.

device/dodge/firmware-match/check is its read-only half, as a script for
the booted Omarchy root (or piped over tools/phone-ssh): sha256 of the 33
A/B firmware pairs, the same list as the 2026-10-28 copy, without
vbmeta, qweslicstore, oplusstanvbk or the boot images. It prints MATCH,
DIFF or MISSING per pair and one RESULT line, exits 0 only when every
pair matched, and never writes: no setrw, no dd, no copy mode. A copy
stays a by-hand step the owner approves.

Signed-off-by: Martin Lund <mrlund@hotmail.com>
Assisted-by: Grok Bot
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.

4 participants