POWER-FINAL-20 — reclaim the never-initialised UART1 clock

Status: INSTALLED, FIRST BOOT CHECKED (2026-10-13). Package out/y2linux-platform-v1-power-final-candidate-20/ (fallback = 19), receipts out/deep-sleep-final/build20-receipts/. BOOTIMG 1ffdb4a2d024cf2d1c8554c518b859b2e7fe8a56186aefea4f8cbb35e4e93c55, Y2ROOT 98ffdca9c2595f245c0ad94b25c9e89c99796f2234324fbd22c994c046859548. Identity: build Y2LINUX-PLATFORM-V1-POWER-FINAL-20, kernel 6.18.56-y2linux-platform-v1-power-final-20, root 2026.08-platform-v1.50, release 1.0.0-platform-v1-power-final-candidate.20; Linux c853915 on main, Reborn unchanged (217e28d). Flash BOOTIMG + ANDROID only with MT6582_preserve_data_scatter.txt, Download Only. Scope audit: "POWER-FINAL-20 UART1 clock reclaim scope audit" in the roadmap gap audit.

Why

After 19 every INFRA clock gate matches stock at sleep entry and INFRA still stays powered (archive 116). The last peripheral gate that differs is PERI_PDN0 bit 17, UART1: stock gates it in every scenario checklist (idle, deep idle, suspend). UART1 has no Linux owner. 19's boot handoff showed `UART1 lcr=0 ier=0 lsr=0x0 dma=0 reason=tx_not_drained retained=1`: an all-zero, never-initialised UART was taken for an undrained transmitter, so its clock stayed on from the bootloader for the whole uptime (UART2/3 read the idle 0x60 and were gated).

What 20 changes

• y2_unused_uart_never_initialised() (idle-clock-policy.h): an all-zero LCR/IER/DMA/LSR set is not busy; the CCF

  unused-clock handoff reclaims UART1 at boot like UART2/3. Any non-zero field keeps the previous rule. clocks.unused_handoff reports reason=never_initialised retained=0.

• Every UART1 register read in the kernel was already conditional on its clock being on.
• Expected side effect: the deep-idle UART owner check no longer sees an unknown UART1.

Tests

Locked-container production suite: 689 cases, exit 0 (the new assertions extend test_cpu_idle_completion). Package validation, ELF, installed ARM, privacy and inventory PASS.

Hardware results

First boot after the flash (boot 53bad0d8), USB networking did not work. The gadget enumerated and configured (stage 6, usb0 up with 10.42.0.1), the Y2 received 36 packets but sent none (usb0 tx_packets=0, DMA completions 10/10, no unmask loss); the host saw bulk-OUT transmit timeouts (NETDEV WATCHDOG, 83 TX errors, ARP for 10.42.0.1 FAILED). Wi-Fi ping answered. The owner turned the player off and on (clean shutdown) and replugged; the next boot 6fd4d362 worked at once. A milder version occurred on 18's first check (first SSH timed out, 1 of 2 pings answered), so it is not attributed to 20's change; cause open (first connection after a flash). Evidence: archive out/deep-sleep-final/evidence/20261013/117-power20-first-check/ (rows of boot 53bad0d8 in sleep-diagnostics.jsonl).

Working boot 6fd4d362 (archive 117): kernel -power-final-20, taint 0, application_ready 40.4 s; clocks.unused_handoff: UART1 ... reason=never_initialised retained=0; uart1_gate_now=1; awake PERI_PDN0 0x2eed3fd (bit 17 set); uart_sleep_owner_ready=1 (the deep-idle UART owner check no longer refuses); all 16–19 features armed; Wi-Fi, Bluetooth, USB (archive taken over it), charging.

What to check after the flash

1. Boot, display, audio, radios, USB; cat /sys/module/clocks/parameters/unused_handoff shows UART1

   reason=never_initialised retained=0; PERI_PDN0 bit 17 set (golden row PERI_PDN0_STA now matching).

2. A ~35-min sleep: golden SoC differences down by one; y2_infra/state last_lost.
3. The overnight drain against 16's ≈8.6 mA (first night for 17–20 together).