Rockchip RK818 (PMIC)

Identity

Status — ● working

Regulators come up at boot via U-Boot’s PMIC defaults; the RTC, charger, and battery driver all attach. Charging now works after 8da75ba rk818: clear OTG/SWITCH2 in DCDC_EN — battery now charges from PD cleared DCDC_EN_REG[7] (OTG_EN) at init — the chip’s own 5 V boost was back-driving VBUS while the PD source was sourcing it, parking the charger FSM at “discharging” even with CHRG_EN set. The input-current limit follows the FUSB302 PD contract ( ef6035e fusb302+rk818: apply PD-negotiated input current to the charger ). SWITCH2_EN is deliberately separate from Type-C role handling after the Linux parity cleanup in 4d190f3 rk818: decouple SWITCH2 from Type-C OTG . The latest cleanup puts that handoff behind the local power_supply coordinator instead of letting fusb302(4) include RK818 headers directly ( 6469f2e power: add charger coordination layer ), and the prepared-sink contract from 06348c6 fusb302: prepare sink path after TCPC attach + 9041600 rk818: hold SWITCH2 through TCPC probe + 544bd86 rk818: preserve prepared Type-C sink path keeps DCDC_EN at 0x3f across the IRQ-driven enforce path so PD Request TX succeeds on both CC orientations.

Verified live on 2026-05-03 (kernel #138+, head 544bd86 at the relevant boot): with a 65 W PD wall on CC1, dev.fusb302.0.negotiated_voltage_mv: 9000, dev.rk818_pmu.0.rk818_hw_input_current_limit_ma: 3000, USB_CTRL 0xcb, DCDC_EN held at 0x3f across rk818[irq]: enforced sink mode (mask=0xc0, DCDC_EN 0xff->0x3f). Battery rate goes positive (charging) once the input-limit is raised. PMIC IRQ migration ( d1b3a0b rk818: drive charger updates from PMIC IRQ ) means charger state updates are event-driven rather than polled.

What’s missing: no suspend/resume integration; TSADC readout and soft caps are now sane, but hardware thermal trips remain separate work and not RK818-side. The noisy decoded rk818[poll] register dump is gone from the normal path; decoded state is now on-demand via hw.acpi.battery.charger_dump=1 or dev.rk818_pmu.0.rk818_debug=1. The fuel-gauge sysctls now expose chip-reported state (see Parity verification below), and the ACPI compatibility status bit is fixed (2 means charging).

Driver

• Our tree: src/sys/dev/iicbus/pmic/rockchip/rk818_battery.c + rk818_battery.h — combined battery + charger driver. Registers rk818-charger with src/sys/dev/power_supply/power_supply.c, so Type-C drivers ask for input-current / source-role changes through a small class interface instead of calling RK818 symbols directly.
• Linux mainline (split): drivers/mfd/rk808.c (MFD core), drivers/regulator/rk808-regulator.c (regulators), drivers/power/supply/rk818_battery.c, drivers/power/supply/rk818_charger.c.

Linux splits this chip across MFD core + regulator + battery + charger. We collapse it into a single battery + charger driver and let U-Boot own the regulators, which is enough for boot but means we don’t have runtime regulator control. The charger / fuel-gauge logic is loosely modelled on the Linux drivers but the FSM interpretation in rk818-charger-not-actually-charging documents how much had to be relearned by hand because the chip’s actual behaviour diverged from both Megi’s downstream driver and the public datasheet.

The decoupling of OTG and SWITCH2 (cross-driver audit §fusb302) is a prerequisite for any future source-role / DP-alt-mode work.

Open work

• Expose battery state to userspace through hw.battery.* / acpiio(4) so acpiconf -i and OS battery indicators work. Driver-specific sysctls (dev.rk8xx.0.{voltage_mv,current_ma,soc_pct,remaining_mah,full_mah,ts1_raw,status}) landed; bench verification below. Wiring into FreeBSD’s acpi_battery provider API would still be useful for stock tools (acpiconf -i 0, apm) but is not a clean fit for non-ACPI ARM and is deferred.
• Convert BAT_TS1_ADC raw to milli-degrees Celsius (NTC table lookup — see Linux’s rk818_bat_get_ntc_res plus ntc_table[]). Raw ADC is exposed for now.
• Wire the PMIC IRQ for charger / fuel-gauge state changes — currently we refresh cached state from a quiet 1 Hz gauge thread and emit POWER_SUPPLY devctl notifications when the cached charger state changes.
• Suspend/resume integration: PMIC sleep mode + low-power regulator transitions.
• Run-time regulator control (rk808-regulator equivalent) so consumers can mark rails always-on / boot-on at the kernel level rather than relying on U-Boot defaults.

Parity verification

The fuel-gauge readout refreshes once per second from rk818_battery_thread and caches values for sysctl readers. Compare against a Linux PinePhone Pro image’s /sys/class/power_supply/rk818-battery/* files (note: Linux reports voltage/current in microvolts/microamps, FreeBSD here in mV/mA).

Falsifiers

• Voltage way wrong (0 or > 5000 mV): VBAT register order (H vs L) or the calibration-K/B math is broken. Cross-check BAT_VOL_REGH = 0xC4, BAT_VOL_REGL = 0xC5 against include/linux/mfd/rk808.h in mainline, and that VCALIB0/VCALIB1 reads return non-zero values around 3000/4200.
• Current sign wrong (positive while discharging): the 12-bit sign-extension at bit 11 (if (val & 0x800) val -= 4096) is missing or inverted. Linux uses the same bit-11 test.
• soc_pct stuck at the same value for >10 minutes of bench charge: the chip’s coulomb counter wasn’t primed at boot. The driver only primes (GASCNT_CAL_REG3..0) when the DT monitored-battery node has a non-zero charge-full-design-microamp-hours — confirm with bootverbose logs that the design capacity was read. If still stuck, Linux also writes RELAX_VOL1/2 thresholds; we do not yet, because the gauge appears to learn well enough without that on the PPP. Add those writes if soc_pct consistently disagrees with voltage_mv extrapolation.
• remaining_mah is exactly 0 or wildly off (e.g. > 10000): the res_div is wrong. PPP uses a 10 mΩ shunt → res_div = 2 (Linux convention inverted from naïve reading; SAMPLE_RES_DIV2 in Linux corresponds to the 10 mΩ case). If readings are 2× too small, change battery_info.res_div to 1; if 2× too large, leave at 2 and check raw register values.
• full_mah is 0: either NEW_FCC has never been written (chip resets it to 0 on a hard battery yank) or the addresses are wrong. Linux falls back to pdata->design_capacity in that case; we currently do not.

Related

• Why the rk818 says “discharging” with PD plugged in — the war story behind 8da75ba rk818: clear OTG/SWITCH2 in DCDC_EN — battery now charges from PD .
• fusb302: a USB-PD sink that works — PD contract feeds the RK818 input limit.
• Cross-driver audit — fusb302 + rk818 section.
• USB-C / PD verification — recipe to confirm PD + charger state.
• Component: FUSB302.