RK3566: keep the rtw88 driver loaded across sleep

[Jacob-Matthew-Cook]

Summary

• What is the goal of this PR? Cut the RK3566 deep-sleep drain. The global modules.bad unloads rtw88_8821cs before suspend, but on these boards the SDIO card keeps power through sleep (keep-power-in-suspend, vcc_wifi always on), so the RTL8821CS sits fully awake all night with nothing managing it. sleep.sh now honours a platform/device modules.keep list of modules it must not unload, and RK3566 gets one with rtw88_8821cs, so the driver's own suspend path leaves the chip in its power-save state.

Testing

• How was this tested? Built and tested on an RG353M (kernel 7.0.2). 60 s RTC sleep through the logind path with Wi-Fi on: driver stays loaded, wlan0 re-associates 4 s after resume, suspend_stats clean, no rtw/mmc errors. Overnight drain on this build pending.
• Test results: Baseline before the change: 88 % → 63 % over a 10 h 9 min deep sleep, unplugged, ≈ 87 mA average. The RK817 sleep-enable registers (0xb5/0xb6) match the DT regulator-state-mem exactly (CPU/GPU/logic/boost/1v8 off; DDR, 3v3 and PMU rails on), so the excess is peripherals on the rails that stay up, dominated by the undriven Wi-Fi/BT module.
• Artifacts: none published. The RG353M test above ran the first form of this change (a full per-platform modules.bad override yielding the same unload list); the modules.keep form is built and is being re-verified on the same device.

Additional Context

• Affects every RK3566 device; only the RG353M was tested. The unload dates from Allwinner/sleep: implement sleep to freeze #1004 (Dec 2024), the H700 s2idle bring-up, and applies everywhere only because modules.bad is global; nothing on RK3566 ever needed it, and resume with the driver loaded is clean on 7.0.2.
• Why a keep-list rather than a per-platform copy of modules.bad: the global list stays the single source of truth. A platform override would silently shadow any module added to the global list later, and the diff should show only what differs — one module on one platform.
• Still on the sleeping rails and not addressed here: the BT half of the module (hci_h5 keeps it powered for BT wake) and the status LED.
• @porschemad911 — could you take a look?

---

AI Usage

Did you use AI tools to help write this code? YES

🤖 Generated with Claude Code