[tozes]

What and why

On the 1.75C, PWR reaches only the AXP2101 (no TCA9554 mirrors it), so the firmware never saw it. The AXP2101 latches a short press in IRQ status 1 (0x49, bit 3) when IRQ enable 1 (0x41, bit 3) is set. The driver enables that bit, and the app polls it every 100 ms through a new Power::take_key_press() hook, clearing it by writing the bit back.

A short press turns the screen off (the same screen sleep as Screen timeout, with or without a timeout set), and the next press, a touch or BOOT turns it back on; Wi-Fi and the Hermes connection stay up. Holding PWR keeps the PMIC's own power-off. Other boards are unchanged.

Part of a series for the AMOLED-1.75C; the physical report is #103.

How you know it works

Tests on this branch (based on 8069101):

• hermes-gadget build-sim --test: core 127/127. Five app tests (off and on without a timeout, other input only wakes, a prompt wakes, a lost connection wakes, a dimmed screen returns) and the driver's read-once-and-clear in test_power.cpp.
• pytest without the simulator-window tests (no virtual display here): 269 passed, 29 skipped.
• pio run builds: esp32s3-touch-amoled-175c, esp32s3-touch-amoled-175. CI builds every board.

Board: Waveshare ESP32-S3-Touch-AMOLED-1.75C, SKU 33691 (the unit in #103), with Hermes v0.21.6+199.g1744a19 on Fedora 44. The checks ran on builds of a working branch that combined this change with the others in this series (30f8bc0); this branch's own image was not installed on its own.

Physical: short presses turned the screen off and on, a touch woke it, and holding PWR still forced the board off.

Documentation and sources

docs/hardware.md (the 1.75C controls and AXP2101 row), docs/using-gadget.md and a changelog line. Register bits from X-Powers' AXP2101 datasheet (SWcharge v1.0, section 6.13), already linked from hardware.md.

Checklist

• One change per PR, with a title that says what it does
• Relevant docs updated, or the reason no changes apply is explained
• Verification results and remaining limitations recorded
• Applicable vendor sources, licenses, and contributor credit retained
• For a new board: the checklist in CONTRIBUTING.md (not a new board)

[Adolanium]

Thanks @tozes, merged. The driver test is careful: it checks that enabling the key keeps the other interrupt enables, drops a press latched before boot, and doesn't clear the long-press bit. Polling through Power::take_key_press() also keeps every other board unchanged.

I pushed three commits on top:

• a merge with main after Firmware: AMOLED panels draw a true black background #106 and Firmware: battery savings on the AMOLED-1.75C #109 (changelog and validation-table conflicts)
• a follow-up that credits you in the changelog and adds the PWR key to the 1.75C rows in supported-hardware.md and hardware-validation.md
• a wording fix in using-gadget.md. It said a PWR press sleeps the screen "at any time", but power_tick wakes it again on the same tick while recording, playing, or in Wi-Fi setup, pairing or an update, so the guide now says that. I think that behavior is right; a press that does nothing during a reply is better than a screen that goes dark mid-answer.

Looking forward to the next batch.