PMU power-off on the mainline kernel (2026-10-14)

Second board in the keyboard case, landscape image 9m683dbp… (commit 21923ce8: the vendor’s system-off callback restored in nix/patches/mainline/k230-pmu-pwrkey.c). USB-C connected throughout.

Before

Keyboard-base 6.6 (../../keyboard-base/battery-policy/README.md): the policy’s systemctl poweroff reached systemd-shutdown, then the board halted with the panel lit and stayed powered until the reset button. No power-off handler existed (no SBI SRST extension; PMU driver ported without the callback).

After

• Boot (board-read, dmesg): K230 PMU power key input registered on IRQ 17, and no failed to register PMU poweroff handler warning.
• Restart (board-read): the deploy’s reboot into this system returned on its own (readlink /run/current-system = 9m683dbp…). Not observed by camera or serial.
• Power off: systemd-run --on-active=2 systemctl poweroff over SSH at 07:08:36 UTC. Journal of that boot ends Finished System Power Off, systemd-shutdown: Syncing …, Sending SIGTERM … at 02:08:49 CDT. Operator: “it shut off this time” (panel dark, unlike the earlier halt).
• Power on: operator held the power key about 3 s (“i held the power key for three seconds and it woke”). USB-C stayed connected; the cable alone did not start it. Next boot began 02:09:15 CDT; the charger read online 1 and the gauge Full afterwards.

Evidence classes: board-read (journal, dmesg) and operator report. No serial console was attached and no camera frame was taken; the kernel’s own PMU poweroff: message is printed after journald stops, so it is not in the journal.

Limits

• Triggered by systemctl poweroff, not by the shell’s confirmed Power off sheet (power-key task 3.4 asks for that path) and not by the battery policy. All three end in the same systemctl poweroff.
• Wake is the power key’s long press only; a USB-C plug-in is not a wake source in this configuration.