Suspend state probe — test plan (proposed, not yet run)

Goal: obtain the real PMIC/regulator/clock/GPIO state just before and just after sleep on DEEPSLEEP-FINAL-04 and compare it with the awake baseline (archive 78) and the vendor suspend golden table (25 of 44 entries differ awake). This is evidence for ranking candidates; it is not a measurement of current and not a reading taken while the SoC sleeps.

What the probe does (read-only)

tools/development/power-session/suspend-probe.py runs on the Y2 (stdlib Python, copied to tmpfs /run/y2). It reads the PMIC register dump (/sys/kernel/debug/regmap/1000d000.pwrap/registers, ~26 ms), regulator and clock summaries, GPIO, wakeup sources, SPM state, runtime-PM status. Captures: pre-entry (journal suspend_syncing, 0.1–0.7 s before freeze), maintenance (0.8 s after each maintenance wake), post-resume (first userspace instant after a BOOTTIME jump). A capture during which a suspend began is flagged suspend_during_capture. It never writes PM/PMIC/RTC/guard/wakeup state and holds no wakeup source. Captures go to /data/logs/suspend-probe-<boot>-<time>/NNN-label.json (small; ~60 KB each, max 24).

Device changes if approved

Only: the probe file and log on tmpfs /run/y2, and its capture directory under /data/logs. No product, kernel, helper or policy change; the 04 image is untouched.

Procedure

Owner: connect USB, device on the player, paused. Claude: archive first (capture-return.py), then arm-suspend-probe.py under the host lease.
Owner: screen dark, unplug, tell Claude the time, wait ~25 minutes untouched (>= 2 maintenance wakes; sleep starts ~40–60 s after darkness).
Owner: re-plug and tell Claude. Claude: archive first, then analyse: tools/development/analyze-suspend-probe.py <captures> --baseline <archive 78 regs.txt>.
Compare: rails on at pre-entry/maintenance/post-resume; LP_SEL bits; golden differences remaining; enabled clocks; SD/eMMC/connectivity/USB runtime-PM state.

Stop conditions / safety

Probe self-limits (--max-seconds 3600, 24 captures) and stops on boot change. If a stall/reset occurs, do not restart; connect USB and archive (04 stall records apply). Do not read /sys/power/wakeup_count or /proc/kmsg over SSH.

Interpretation limits

Registers read at pre-entry are the state before the freezer/SPM acts; the SPM/PCM may change PMIC state during sleep and restore it before post-resume. A difference between pre-entry and the golden table is therefore a candidate, not proof of a leak. A true during-sleep current needs an external measurement.