Skip to content

device/bluejay: reproduce the reviewed Wi-Fi stage DTs - #777

Merged
derekross merged 1 commit into
derekross:mainfrom
latentoperator:followup/bluejay-tested-wifi-overlays-20261114
Nov 15, 2026
Merged

derekross merged 1 commit into
derekross:mainfrom
latentoperator:followup/bluejay-tested-wifi-overlays-20261114

Conversation

@latentoperator

@latentoperator latentoperator commented Nov 14, 2026 •

Copy link
Copy Markdown
Contributor

What and why

The reviewed GS101 PCI power-control path needs regulator-wlan, but the historical wifi-controller overlay enables only the PHY and PCIe controller. Add wifi-reviewed-controller with the regulator, PHY and controller status switches, and document the w0/w1/w2 combinations with the existing pool and endpoint overlays. This follows the request on kernel #230.

Keep the separately pinned historical path reproducible: kernel c934e316b has no regulator-wlan node, so the historical controller overlay's behavior remains unchanged. No descriptor, package or automatic startup enables the new overlay.

The November 13 trials edited DT status properties directly. These source overlays reproduce those exact DT changes; the new overlay file itself has not been booted.

Device and build

  • Pixel 6a / Bluejay; no phone operation for this PR.
  • Prior attended headless w0/w1/w2 trials: 7.3.0-rc6-omamobile-bluejay-review-20261111, signed integration 720a495233734e4328a09c33c095d72a8291e1c4.
  • No package or version changes. Kernel merge order remains #227 and #228 before #230's two own commits.

Checks (no phone needed)

  • Policy and git diff --check pass.
  • All 69 overlays compile with dtc -@; descriptor overlay checks pass.
  • Applied all three combinations to the frozen integration base DTB. Every property equals both silicon DT variants extracted from each saved trial's actual vendor_boot image, except the explicitly excluded bootargs and silicon ID. The image SHA-256 matches its saved trial manifest. The only base-to-stage changes are the intended status properties; no DMA, interrupt or reservation change is hidden by the comparison.
  • Negative control using the historical controller overlay fails the reviewed w1 comparison because the regulator remains disabled.
  • The integration's PCIe/PHY/Wi-Fi binding source matches the schema checkpoint; all three applied stage DTs validate with no new diagnostics.
  • Complete tests/all and tests/packages pass locally. Independent review found no remaining blocker. Hosted CI will run on the published head.

On the phone: prior evidence and limits

The recorded w0 pool check passed; w1 linked on try 1 to 14e4:4441 with the endpoint unbound and no firmware loaded. The separate w2 trial verified restricted DMA below 4 GiB, DHCP and interface-bound HTTPS (200, 16,932 bytes), with IRQ 725 to 744 and zero observed oversize drops. These existing reports are on kernel #227 and #228.

No fresh phone-smoke or hardware run was performed for this reproducibility-only change. Future runs require their own exact module/root/recovery/watchdog/logging gates; the historical startup helpers still refuse this revised kernel. No suspend, roaming, hotspot, endurance, native-panel or release-installer qualification is inferred. No images, firmware, captures or identifying data are included.

Upstream baseline refresh

The first hosted run hit the stale upstream named-only test introduced by #775. Upstream #776 fixed it. Rebased onto that fixed base (259be39), with the overlay patch unchanged by range-diff. Policy, the corrected stage suite, all overlay compilation and exact packed-DT comparisons pass again locally. Fresh PR-event CI passed on 7341c494beeaedaed8b644d763d04f7b2c19776d against the fixed base (Policy and complete Tests: tests/all plus every package). The failed old run was not reused.

The revised PCI power-control path requires regulator-wlan, which the historical controller overlay cannot enable. Add a separately named reviewed controller overlay so the older pinned bring-up path stays reproducible.

Document the w0/w1/w2 overlay combinations and their runtime limits. Offline property comparisons against both silicon DTs in each saved trial image confirm exactly the intended status changes, with the historical overlay as a negative control. No descriptor or automatic startup is changed.

Assisted-by: OpenAI Codex
@latentoperator
latentoperator force-pushed the followup/bluejay-tested-wifi-overlays-20261114 branch from e34a930 to 7341c49 Compare November 14, 2026 17:49
@derekross

Copy link
Copy Markdown
Owner

Thank you, Chris, this is what I asked for on kernel #230. The reviewed controller overlay enables regulator-wlan with the PHY and the controller, the historical one stays as it was for the c934e316b path, and WIFI-PLAN is honest that the trials edited the DT directly and that the new file hasn't booted itself. I checked that its three targets are in #230's gs101-bluejay-wifi.dtsi, all disabled there, and that no descriptor applies the overlay. The run's numbers match your reports on #227 and #228, and 720a4952 is pushed. Thanks for that too. CI is green and it merges cleanly. Merging now.

For the next hardware run, the runbook will need the usual Expect/Pass, stop condition, rollback and risk for each of the three stages. Your new section already lists the gates it should carry.

@derekross
derekross merged commit d5e95ae into derekross:main Nov 15, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants