Repository navigation
Add ASUS Zenbook A16 UX3607OA board support - #13820
FixItFoundry wants to merge 2 commits into
Conversation
[Medium impact] The changes reach boot and early hardware setup on one Qualcomm laptop model. The PR still needs the missing-kernel install failure and stale early-DSP settings addressed before merging. FindingsSummaryAdds ASUS Zenbook A16 setup for early keyboard, display, firmware, and DSP support.
The previous reset-unlock thread was manually resolved without explanation. The current hook refreshes both boot entries from the shared command line.
|
|
Thanks for this. I reviewed the diff against Pushed: 0c18f75 fixes the heredoc in Before this can go further, a few things need your answer:
For the maintainer to weigh: Nothing here can be reproduced on a VM. This waits on your answers above, then on someone with the hardware confirming the full head SHA on a Zenbook A16. |
|
Thanks for the careful bring-up; the board setup reads well and the test coverage is thorough. What we verified on this board: its device tree reserves 256 MiB via linux,cma-default at glymur-asus-zenbook-a16-ux3607oa.dts:279-284; the ISO orchestrator's finalize step runs limine-update after the hardware phase, picking up these drop-ins; and install/hardware/qualcomm/kernel-params.sh:11 already appends the ignore-unused flags. One correction: the firmware search's suffix loop sits outside the directory loop at zenbook-a16.sh:35-41, so an uncompressed copy in the base root wins over a compressed copy under updates/, and the test at lines 56-61 proves only that uncompressed updates/ beats a compressed base-root copy. That ordering is inherited from the Yoga script, so nothing changes there.
|
|
Thanks for working through those. I read e4fe087 against Two comments in I only read this revision. Nothing ran on a worker, and no second model looked at it: the reviewer, Codex Medium, is reachable but omabot's review budget for today is used up. So the classification of this pull request is unsettled for now, and it gets checked again on a later pass. Whatever happens there, the board-specific behaviour can only be confirmed by someone running the full head SHA on a Zenbook A16. |
|
@FixItFoundry , please rebase |
dcdea0b to
2f83505
Compare
e4fe087 to
15613fc
Compare
|
I rebased this onto today's |
|
@FixItFoundry thanks for the work on this. The layout fits Omarchy's Snapdragon image boots Arch Linux ARM's Which kernel did the verified install run? If it was your own build, I think this waits for 7.3 in Arch Linux ARM and then a test of this head on that kernel. It would help to know what works on a plain 7.3 without your series. A few smaller things.
I can make the smaller changes as a commit on top if that helps. |
|
A shared helper for the model check is up as #15064. With it the device tree part of the check becomes |
|
#15064 is merged, so |
|
A heads-up so it does not trip you up. The helper was renamed after my first note, so if your mail says |
15613fc to
c2a1e31
Compare
|
I read c2a1e31, the rebase onto One point nobody has raised yet: The classification is unsettled. I read this as an enhancement, because it adds support for a laptop Omarchy doesn't run on today. Codex Medium, asked separately, read it as a compatibility fix. Until that's settled the pull request carries no class label, and nothing ran on a worker this pass. This waits on you, @FixItFoundry, to answer the questions @birkskyum asked on 10 October. The main one is which kernel the verified install ran. The smaller ones are the device-tree check through |
|
Thanks @birkskyum and @omarchybot. I have read through the discussion, including the 15 October review. Here is the current status before I update the PR. The successful installation used my custom Following my separate discussions with the Omarchy team, the intended direction is the shared Omarchy 7.2.10 kernel with the necessary A16 backports, using the normal kernel packaging and ISO build flow. A local candidate has been built that way. It still needs full A16 hardware validation; I am not presenting it as merge-ready or as a reason to pin users to a private kernel indefinitely. The immediate installer blocker is reliable firmware delivery on a stock machine with a BitLocker-protected Windows partition. The earlier installer treated extraction as optional and could complete without the required firmware. The newer local preflight stops before disk cleanup when the required stage is incomplete. That explains the apparent install regression; it does not prove a firmware ABI mismatch or establish the cause of every missing device. I also checked the bring-up repository: its live-image scripts copy a prepared Linux firmware tree into the image, rather than require extraction from Windows during installation. Its firmware documentation describes generic upstream Glymur/QCC2072 files, but the working A16 device tree requests ASUS-specific DSP files, and the local preflight additionally requires a Wi-Fi calibration file. We have not yet established that the generic upstream files are equivalent, or the original provenance of every file on the working Fedora system. I am resolving that discrepancy before choosing the permanent delivery method. We are evaluating upstream/vendor delivery and a guided, read-only recovery-key import; requiring users to turn off BitLocker is not the only option. The smaller review points are handled in the local changes: the shared The next acceptance step is an install and boot on the complete proposed revision, checking DSP startup, battery, Wi-Fi, audio including idle recovery, and a real kernel update/rollback. I will report the exact source SHAs and results before asking for another merge review. We are still settling the firmware delivery plan. |
|
I switched the base to |
c2a1e31 to
a34db8a
Compare
| if [[ -n $a16_kernel ]]; then | ||
| source "$OMARCHY_INSTALL/hardware/asus/zenbook-a16-early-dsp.sh" || return 1 | ||
| else | ||
| # Start the board's DSP remote processors. | ||
| mkdir -p "$systemd_dir" | ||
| cat >"$systemd_dir/zenbook-a16-remoteprocs.service" <<'UNIT' |
There was a problem hiding this comment.
Early DSP settings survive downgrade
If an A16 previously configured for the audited kernel later loses that kernel or its DSP ABI, setup takes this late-start branch without removing the early initramfs profile or boot-entry drop-in. A subsequent boot-image rebuild can still use the early DSP settings with a kernel that failed the capability check, defeating the separation between the two startup paths.
Knowledge Base Used: Install stages and package sets
There was a problem hiding this comment.
The late-start fallback is removed. Failed package/ABI checks stop setup before it changes the boot configuration. The initramfs profile and build hook also reject an incompatible kernel, and pacman enforces the ABI dependency. An existing profile link cannot bypass those checks. Forced dependency removal is not a supported downgrade path.
|
Want your agent to iterate on Greptile's feedback? Try greploops. |
a34db8a to
7164dbe
Compare
|
Thanks for the review. The unified ISO selects I've removed the late-start fallback and its unused helper. A16 setup now stops before changing configuration if the required kernel capability or support profile is missing, preserving the existing DSP guard and boot settings. Sleep now uses The ISO already rebuilds the boot images after hardware setup. Focused regression tests pass; fresh-install and hardware upgrade/rollback validation of the complete revision remain pending. |
| if [[ -z $a16_kernel ]]; then | ||
| echo "A16 setup requires linux-omarchy with DSP ABI 1 and omarchy-a16-support; refusing unsafe late DSP startup." >&2 | ||
| return 1 |
There was a problem hiding this comment.
On a normal Zenbook A16 install, the Qualcomm package set supplies omarchy-a16-support but not linux-omarchy. This new check therefore returns 1 before applying the board settings. The failure aborts the system hardware stage; on an image install, the deferred step stays queued and fails again on the next run.
Knowledge Base Used: Install stages and package sets
There was a problem hiding this comment.
Agreed. The runtime/ISO changes stay blocked on package publication. The proposed support package pulls in a shared kernel providing omarchy-a16-dsp-abi=1 and its release metadata. The private firmware-package dependency is removed. The replacement package set has passed build, boot and hardware validation on my A16 with revision 24.
|
Thanks @FixItFoundry. Two things block this on
The A16 step reads |
|
I read 7164dbe against I can confirm both blockers @birkskyum raised on 16 October:
One smaller point that nobody has raised: Codex also suggested that the step reads the root command line from the wrong file. That is not the case. The installer writes The classification is still unsettled. I read this as an enhancement, because it adds support for a laptop Omarchy does not run on today. Codex Medium, asked separately, read it as a compatibility fix. Until that is settled, the pull request carries no class label. This waits on you, @FixItFoundry. The package recipes and the kernel marker need to land in omarchy-pkgs first, and then the hardware validation you described on the 15th needs to run on this head. |
7164dbe to
937b15e
Compare
|
Thanks @birkskyum and @omarchybot. I agree the package recipes and kernel capability need to land in omarchy-pkgs and reach edge before the runtime and ISO changes merge. The revised proposal follows Omarchy's shared 7.2.8 base. I built and installed 7.2.8-16 with support 20261014-12 on the A16. The package update rebuilt and verified both normal and rescue boot images. A normal-entry boot now passes, with audio playback, the built-in keyboard, touchpad and battery indicator working. Wi-Fi is connected, and both DSPs run. Revision 16 adds the upstream block-mq fix already in 7.2.10. Our source-level check reproduces missing scheduler initialization without it and passes with it. The fix addresses a known Kyber queue-stall bug. Whether it explains the A16 freezes is still a working theory; long-term stability hasn't been validated. Additionally, my test environment runs off a USB-C SSD, so I suspect the stability issues might be related to that. The local runtime follow-up removes the redundant default Limine drop-in and explains the temporary Plymouth setting beside the code. The model/kernel capability checks remain in place, and sleep is still disabled. Next step would be to work on sleep/suspend. I noted a possible charging regression in this Omarchy build that I didn't see in my Fedora project. Our existing charger-notification patch also overlaps omacom/omarchy-pkgs#1030 and needs coordination. |
|
I read 937b15e, the rebase onto Against 7164dbe, the pull request's own files change in two places, and both are correct. Both blockers from 16 October still stand. Codex found one smaller point, and I confirmed it. The classification is still unsettled. I read this as an enhancement, because it adds support for a laptop Omarchy does not run on today. Codex Medium, asked separately, read it as a compatibility fix. Until that is settled, the pull request carries no class label. This waits on you, @FixItFoundry: the package recipes and the kernel marker in omarchy-pkgs, then the hardware validation of this head on a Zenbook A16. |
|
The follow-up targets linux-omarchy 7.2.8-24. After testing and debugging, I rebuilt it, installed it and completed hardware validation on my A16. It's still running. Audio playback, keyboard, touchscreen, charger transitions, and Bluetooth calls and audio with my phone passed. Both DSPs start before root mount, and the early modprobe and battery uevent errors are gone. The update removes the private firmware-package dependency, checks prerequisites before changing board settings, and refreshes normal/rescue command lines from Limine before each build. Packages still need to publish before the runtime/ISO changes merge. The known PMIC supplier-link warnings remain. Camera and sleep are disabled, but we have a potential path forward for camera. |
|
I read b86c35c against Against 937b15e, this PR's own files changed in three ways, and all three hold up:
On the package blockers: the kernel marker the A16 step reads ( The classification is still unsettled at this head. I read this as an enhancement, because it adds support for a laptop Omarchy doesn't run on today. Codex Medium, asked separately, read it as a compatibility fix. Codex was asked only for the classification this pass and did not review the code, so the pull request carries no class label for now. This waits on the maintainer to settle the classification, and on |
|
Closing this older 7.2.8-based proposal in favor of the current A16 setup in #17083 and omacom/omarchy-pkgs#1875. This branch depends on the earlier support-package/DSP-ABI design. The 7.2.8-24 test results remain historical; they do not validate the new 7.3 changes. |
Adds model-gated ASUS Zenbook A16 UX3607OA support using shared linux-omarchy 7.2.8-24 and omarchy-a16-support.
Setup checks the kernel DSP ABI and root command line before changing board settings, enables early DSP/audio and keyboard/display support, and removes the late-start fallback. Normal and rescue entries refresh from Limine's current default before each build, preserving factory-reset unlock settings. Required OEM firmware comes from the owner's Windows installation.
Revision 24 is built, installed, fully tested for this bring-up scope and still running on my A16. Native build, all 8,487 module checks, packaged-header compilation, source replay and both UKIs pass. Both DSPs start before root mount; the early modprobe and battery uevent errors are gone. I confirmed audio playback, keyboard, touchscreen, charger transitions, and Bluetooth calls and audio with my phone.
Publish the package prerequisites in each target channel before merging the runtime/ISO changes. Camera and sleep remain disabled; camera work is progressing separately.