POWER-FINAL-44 — dark recovery screen: DSI register reads lag one word

Status: FLASHED by the owner (2026-10-16; cold start lit, archive 20261016/148); tested over the air with 45. Package out/y2linux-platform-v1-power-final-candidate-44/ (fallback = 43, which the player runs), receipts out/deep-sleep-final/build44-receipts/. BOOTIMG f2c19f69980a6ea2491503437db29d0e32cd25a40cff6d4a28c9799242437385, Y2ROOT d53db06a9d54ebf89a6e49a6bd0c559b11dca95218c23c812dbe493bdf30d0a9, RECOVERY 56fdf20b07eb7f4fe61ec92f33aad8a958118578a870ddcc8ca2916500d6a77f. Kernel 6.18.56-y2linux-platform-v1-power-final-44, release …candidate.44, update sequence 4400. Container suite 718 OK, all seal checks PASS. Signed OTA (release key): …-44-ota-release/; development reinstall: …-44-ota-devtest/.

43 results

Install without a Power press PASSED (43 development reinstall, restart switch on): restart into recovery,
install 53 s (verify 1.7 s → backup 3.7 → system 34.9 → check 48.0 → boot 52.6 → installed 53.3), restart into the new system (boot reason 4), display lit (vblank 16 → 219), health Acknowledged.
The recovery's progress screen stayed dark (owner: "only the update display did not show with usb
connected"). The restart into the result was lit.

Why the recovery screen was dark

The recovery's kernel log (/data/logs/recovery-last.log) has the 43 SMI hold, an idle larb0 at the hand-over and a clean mutex reset, but no vblank after the first modeset (vblank wait timed out, flip_done timed out). Its DSI scan line reads the register file shifted by one word: each read returns the previous read's data (vactive=198048 is PSCTRL's 0x305a0, lcc=0x168 is VACT_NL's 360). The only earlier boot with this signature is archive 20261016/143 (42, cold start with USB plugged in, early hand-over at 0.49 s): its snapshot differs from the lit restart 147 in 77 DSI words and nothing else that matters (SMI, MMSYS gates, PHY, mutex identical). The real CON_CTRL reads 0: every read-modify-write of the DSI (mtk_dsi_mask()) stored a neighbour's value, DSI_EN never stuck, no frame started. 144 (the larb0 wedge fixed in 43) reads DSI correctly, so this is a second, intermittent failure mode. USB enumeration is not concurrent with the modeset (5 s in 143). Why the register port lags is still unknown.

Change	Detail
`mtk_dsi_poweron()` (patch 0017)	After the timing setup, reads PSCTRL then VACT_NL; if VACT_NL is not the mode's vactive (`Y2SCAN DSI read-back …`), power-cycles PHY and DSI clocks and repeats the setup, at most 3 times (`DSI register reads lag one register; power-cycling the DSI (attempt n)`). A correct read-back costs two register reads.
Hand-over	`handoff: clocks on; park video (ps=… vact=…)`: the same read-back of LK's timing shows whether the lag comes from the loader.
Recovery	Saves `/sys/kernel/debug/y2_display_snapshot` as `/data/logs/recovery-display.txt` with its kernel log.

Owner steps

With the owner's OK: install 44 over the air (restart switch on): staged …-44-ota-devtest or the release
package, Turn Off and Install in Reborn (with the switch on it restarts instead). Watch the recovery screen.
The agent checks the recovery log for the read-back lines, Acknowledged, and the new boot's display.
If the recovery screen is dark again: the agent saves recovery-display.txt and the log (archive), and a
recovery probe (y2-update-core probe, restart switch on) repeats the restart-entered recovery without installing anything.