Skip to content

Oriole: describe the disabled rear torch and own external strobe - #242

Open
Gr-linkk wants to merge 3 commits into
derekross:omamobilefrom
Gr-linkk:omamobile-oriole-torch-candidate
Open

Gr-linkk wants to merge 3 commits into
derekross:omamobilefrom
Gr-linkk:omamobile-oriole-torch-candidate

Conversation

@Gr-linkk

@Gr-linkk Gr-linkk commented Nov 11, 2026 •

Copy link
Copy Markdown
Contributor

Describe the Pixel 6 rear torch in the native board tree while preserving the external strobe safeguard currently supplied by its separate bring-up controller. Keep USI7, HSI2C7 and the LED controller disabled; this does not enable torch at boot.

The three commits add the optional strobe-gpios binding, request that GPIO inactive before HWEN in the existing LM3644 driver, and describe Oriole's wiring. The board uses one sink, a conservative 6000 microamp cap, 100 kHz bus timing and default off. No private staging framework, firmware, calibration or package payload is included. Existing boards without the property retain their behavior.

Validation:

  • Strict checkpatch on each file change, W=1 torch compilation, binding check including YAML lint, and full schema validation of disabled and test-enabled DTBs pass.
  • Full public-tree W=1 Image/modules build and modpost pass on the original df3a12b9 base. The final commits are on fc7086a6; its four intervening upstream commits touch unrelated files, and the three-file patch is byte-identical, so those build/schema results are reused. Its 1171 compiler warnings exactly match the saved broad-build baseline by normalized file/line/message; none are new torch warnings.
  • The modified module also builds and links with the exact retained Oriole integration SDK/config/exports and original Android clang toolchain. All 22 imports resolve. This is build evidence, not a claim that the public Image matches the running phone.
  • On a Pixel 6 running 7.3.0-rc6-oriole-hsi1-late-dev0-g4bd35d34abb9-dirty, the modified LM3644 module was tested against its matching integration SDK on a fresh boot with the old torch provider suppressed. A guarded staging helper supplied the node and released both initially-low GPIOs before LED probe; the native driver then owned both STROBE and HWEN. Cold ID/reset checks, default-off state, maximum brightness 2, and both raw GPIO levels low passed.
  • A guarded three-second brightness-1 pulse (nominal 2550 microamps on one sink) passed with an independent five-second off fallback and verified final brightness zero, STROBE/HWEN low, and storage protection. The owner physically confirmed visible on/off during the new native-driver test; the observation is not attributed to a particular pulse.
  • The existing shell's toggleTile action was invoked through its documented IPC as the desktop user, exercising the same backend as the torch tile. It selected brightness 2 (nominal 5460 microamps), reported On, then returned to brightness zero/HWEN low and Off in about 1.04 seconds, before the five-second fallback. This was an IPC test, not a physical touchscreen tap. Current values are programmed nominal values, not measured LED current.
  • The unchanged NFC provider started successfully alongside the native torch on the shared 100 kHz I2C bus and reached its ready, initially-off state. This establishes startup coexistence, not NFC tag reading or payment support.
  • An ordinary reboot from the verified torch-off state passed ordered service shutdown: the previous-boot journal records NFC stopping successfully before the native torch service stopped successfully. On the next boot, the test interruption markers were cleared, temporary unit/payload were absent, and the durable test conditions correctly suppressed both original providers. USB access required a manual cable reconnect; this is not a claim of reliable unattended reconnection.
  • Test-only suppression was then removed and the original qualified torch and NFC services were restored. The final system was running, all six baseline services were active, the torch was off, protection checks passed, and test-only files were absent.

The new native GPIO ownership, bounded light output, shell control, NFC startup coexistence, and off-state ordered shutdown/clean restart have been tested on the existing integration kernel. The board nodes remain disabled. Shutdown while lit, interrupted-native-marker/power-loss recovery, persistent native-driver autostart, and a phone boot of the public-tree Image have not been qualified. No default-enable or full-phone support claim is made here. The private staging helper and test packaging are not part of this PR.

Related work: Raven torch support in #113 is already merged. This reuses that LM3644 driver and existing standard LED/shell control path; it does not replace them.

AI assistance: Codex helped compare the GPL board wiring and retained evidence, implement the three focused changes, and run the builds and guarded pulse. Commits credit Assisted-by: LLM.

Independent Astra review of the settled patch reported no findings. Final commit checks pass strict checkpatch (zero warnings/checks), whitespace and identifier/blob policy; all three commits have the human author's DCO sign-off.

Some boards connect the LM3644's external STROBE input to a GPIO.
Torch-only operation must keep this active-high input low rather than
leave it available to another consumer while HWEN is asserted.

Describe an optional strobe-gpios property so the torch driver can own
and hold that line inactive. Existing boards without the property keep
their current behavior.

Assisted-by: LLM
Signed-off-by: Garrett lindquist <gr_link@hotmail.com>
Oriole's camera wiring includes a separate external strobe GPIO. The
existing torch bring-up controller holds it low, but the native driver
only owns HWEN. Moving the board description into the native device tree
must preserve that ownership without the separate controller.

Request the optional strobe GPIO as an inactive output before requesting
HWEN or identifying the chip. Abort probe on a GPIO acquisition error.
Device-managed ownership keeps strobe inactive throughout the device's
lifetime and releases HWEN before strobe during probe unwind or removal.
Boards without strobe-gpios retain the existing torch behavior.

Build-tested with W=1, full Image/modules linking and modpost. The changed
driver also links against the retained Oriole integration SDK. The new
strobe ownership has not been exercised on hardware.

Assisted-by: LLM
Signed-off-by: Garrett lindquist <gr_link@hotmail.com>
Describe Oriole's LM3644 on HSI2C7 at 0x63, with HWEN on gpp23-6 and the
external strobe on gpp27-0, following Google's GPL camera device tree:
https://android.googlesource.com/kernel/devices/google/raviole/+/81506b991a5385be6c59ee27df4dda2b1379c507/dts/gs101-oriole-camera.dtsi

Keep USI7, HSI2C7 and the LED controller disabled pending qualification
of the native GPIO ownership on the current kernel. Use 100 kHz bus
timing, one current sink, a conservative 6000 microamp limit and default
off. The existing I2C pin group uses only gpp14-0/1.

The existing qualified controller and native LED driver produced visible
rear light at brightness 1 during a three-second test and returned to
verified HWEN-low. That path uses separate strobe ownership and does not
qualify the new strobe-gpios property or enable these nodes by default.

The disabled DTB and an enable-only test overlay pass schema validation.

Assisted-by: LLM
Signed-off-by: Garrett lindquist <gr_link@hotmail.com>
@derekross

Copy link
Copy Markdown
Owner

Thank you, @Gr-linkk. This is a tidy series: the binding first, then the driver, then the board, with USI7, the bus and the LED node all left disabled. I checked the wiring against Google's gs101-oriole-camera.dtsi at the commit you cite (HW_ENABLE GPP23[6], FLASH_STROBE GPP27[0], the strobe held low while the flash is enabled), and it matches. The gate passes on today's omamobile and CI's dtbs_check and dt_binding_check are clean. The driver change is safe for the other phones that use this driver: none of their nodes has strobe-gpios, so the optional lookup returns nothing and the rest of the probe is as it was.

Three small things before I merge:

  1. The commit messages and the description disagree, and the commit messages are what lands. The driver commit ends "The new strobe ownership has not been exercised on hardware", and the board commit says the tested path "does not qualify the new strobe-gpios property". The description says the native driver then owned both lines on your Pixel 6. That run was on 7.3.0-rc6-oriole-hsi1-late-dev0-…-dirty with your own helper supplying the node, so it isn't a test of this tree, and that's fine for nodes that stay disabled. Please make the two agree: either a sentence or two in each commit saying what ran, on which kernel release, and that it was not this tree, or keep "not exercised" and drop the claim from the description.

  2. The binding text. "When supplied, the torch-only driver holds it low to prevent external flash triggering" describes what Linux does. A binding describes the hardware, so something like: "GPIO connected to the STROBE pin, an active-high input that triggers flash mode."

  3. samsung,clkreq-on on &usi7. In this tree only the UART has it ("needed for UART mode"), and the Pixel 6 Pro's torch uses the same USI7 in I2C mode without it. It keeps the block's clock on all the time. Did the bus fail without it on your phone? If so, please say so in the commit message with what you saw; if not, please drop it.

For later, not for this PR: turning the nodes on is its own commit after a boot of this tree, with the enable overlay under device/oriole/overlays/ in omarchy-mobile and a runbook step for the attended pulse. Before that boot it would be worth reading the pulls the bootloader leaves on gpp23-6 and gpp27-0. The Pixel 6 Pro has the same strobe line and its node doesn't own it, so your property may be wanted there too.

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