Fuel gauge under Linux (task 5.3, partial; board-read)

gauge-2026-10-13.log is from system l002j7bg… (the first build with nix/patches/mainline/bq27xxx-bq27220.patch) at 22:48 UTC. It records:

the bq27xxx driver binding the ti,bq27220 node at /soc/i2c@91409000/fuel-gauge@55;
bq27220-0 reporting Full, 100 %, 4.207 V, 0 A, 36.6 °C and 3000 mAh design capacity.

USB-C was connected and the cell was full, so the current was 0. That matches the probe’s raw registers.

Not yet covered

Nothing for task 5.3; the cable-out online reading is at the end.

Unplug and replug (2026-10-13 22:17–22:32 CDT)

Evidence class: board-read (operator unplugged and replugged USB-C on request). unplug-replug-2026-10-14.log samples the gauge’s power_supply and the charger’s REG0B (status) and REG11 (VBUS) once a second, read-only (i2cget -y 1 0x6b, no driver bound to 0x6B).

Time	Event	Gauge	Charger
22:17–22:26	on USB, charge done	100 %, 4.209 V, 0 mA, `Full`	REG0B 0x5e (adapter, charge termination done, power good), REG11 0x97 (VBUS 4.9 V)
22:26:1x	USB-C unplugged	-400 to -411 mA steady, `Full`	REG0B 0x02 (no input), REG11 0x17
22:30:51	after ~4.6 min	99 %, ~4.04 V
22:31:3x	USB-C replugged	+719 mA, later +672 mA, `Full`	REG0B 0x56 (fast charging, power good), REG11 0x98

Discharge reading: about 405 mA for 5.3 min, roughly 36 mAh, which moved the gauge one percent of its 3000 mAh. The gauge said Full throughout, including while discharging and charging: bq27xxx reports Full whenever the BQ27220’s FC flag is set, and that flag stays set after termination until the charge falls below its clear threshold. nix/patches/mainline/ bq27xxx-bq27220-status.patch makes a nonzero current decide.

Where the charger’s INT goes (task 5.2a)

LILYGO’s pin map (Xinyuan-LilyGO/T-Display-K230, k230_bsp/docs/HARDWARE_PINMAP.md) lists the BQ25896’s I2C lines only. During the same unplug and replug, gpiomon -e both watched every K230 GPIO line not already claimed (gpiochip0 all but 2, 22, 23, 24, 25; gpiochip1 all but 10, 11, 20; every one already an input, so requesting it changed nothing). Neither event produced an edge on any line (mon0.txt and mon1.txt were empty).

Limit: a pad muxed to a non-GPIO function would not show an edge either, and this kernel’s pinctrl debugfs prints the same placeholder for every pad, so the mux of unclaimed pads is not shown here. The XL9555 has no interrupt output wired either, so a charger INT routed to an expander input could not be caught by polling a 256 µs pulse. Result: INT reaches no GPIO-muxed K230 line. The charger driver therefore polls (bq25890-poll-without-irq.patch).

The charger node (task 5.2)

nix/dts/k230-tdisplay-mainline-drm-common.dtsi adds charger@6b, compatible = "ti,bq25896", with linux,skip-reset and linux,read-back-settings. The driver therefore reads the charge parameters from the chip; the ti,* values are still required by the binding and are set to what the registers already hold, so nothing changes if read-back is ever dropped. Host proof: nix build .#deviceTreeMainlineDrm (2026-10-14, built k230-tdisplay-mainline-drm.dtb; dtc shows the node).

Decoded from the probe’s registers (../probe/probe-2026-10-13T193545Z.log) against the BQ25896 datasheet:

Register	Value	Field	Decoded	Node property / driver
REG00	0x7f	EN_HIZ, EN_ILIM, IINLIM	0, 1, 3250 mA	read back (`input_current_limit` 3250000)
REG02	0x11	CONV_RATE, ICO_EN, AUTO_DPDM	0, 1, 1	driver writes CONV_RATE=1 (continuous ADC)
REG03	0x1a	OTG_CONFIG, CHG_CONFIG, SYS_MIN	0, 1, 3500 mV	`ti,minimum-sys-voltage` 3500000; driver writes CHG_CONFIG=1 (already 1)
REG04	0x0b	ICHG	704 mA	`ti,charge-current` 704000
REG05	0x13	IPRECHG, ITERM	128 mA, 256 mA	`ti,precharge-current` 128000, `ti,termination-current` 256000
REG06	0x72	VREG, BATLOWV, VRECHG	4288 mV, 3.0 V, 100 mV	`ti,battery-regulation-voltage` 4288000
REG07	0x8d	EN_TERM, WATCHDOG, EN_TIMER, CHG_TIMER	1, disabled, 1, 12 h	driver writes WATCHDOG=0 (already 0)
REG08	0x03	BAT_COMP, VCLAMP, TREG	0, 0, 120 °C	`ti,thermal-regulation-threshold` 120
REG0A	0x73	BOOSTV, BOOST_LIM	4998 mV, 1.4 A	`ti,boost-voltage` 4998000, `ti,boost-max-current` 1400000

Bound on the board (tasks 5.2, 5.2a, 5.3; board-read)

System 9lp7f072… (kernel 7.3.0-rc5, bq25890-poll-without-irq.patch and bq27xxx-bq27220-status.patch). dmesg: bq25890-charger 1-006b: no interrupt; polling every 2000 ms. registers-bound-2026-10-14.log (05:02:48 UTC, i2cget -f, read-only) against the probe:

REG02 0x11 → 0x51: CONV_RATE, the one field named above. No other configuration register (0x00–0x0A, 0x0D, 0x14) changed.
REG0B and 0x0E–0x13 are status and ADC readings and move with the charge state (REG0B 0x56 here: fast charging, power good).

bq25890-charger-0 reports type USB, online 1, BQ25896, constant charge current 704000, voltage 4288000, input limit 3250000.

devmem is installed on this image (STRICT_DEVMEM, IO_STRICT_DEVMEM off, I2C_CHARDEV). Reading the IOMUX block at 0x91105000 + 4 * pin returned 0x800019d1 for all 64 pads, so it does not reveal the pad mux either; the 5.2a limit stands.

Charger Hi-Z: gauge status follows the current (2026-10-14; board-read)

hiz-2026-10-14.log, 05:01:07–05:02:25 UTC. Writing 0 to the charger’s online sets EN_HIZ (REG00 0x7f → 0xff): the input is disconnected and the battery carries the system, as with the cable out, but from software. Writing 1 clears it. Sampled every 5 s:

Phase	Charger	Gauge
before	online 1, Full	`Full`, 0 mA, 4.201 V, 100 %
Hi-Z, 45 s	online 0, Discharging	`Discharging`, -394 to -417 mA, 4.121 → 4.096 V
restored, 30 s	online 1, Charging	`Charging`, +574 mA tapering to +396 mA, 4.284 V

Before bq27xxx-bq27220-status.patch the gauge said Full in both directions (unplug log above). This run used Hi-Z, not a pulled cable, so it proves the status mapping and the charger’s online reporting, not the cable-out online reading.

Cable out with the charger bound (2026-10-14; board-read)

cable-out-2026-10-14.log, 05:11:14–05:11:19 UTC, system 9lp7f072…. The operator pulled USB-C and reported it; the samples were read over SSH afterwards.

Charger: online 0, Discharging; REG00 0x7f (EN_HIZ clear, so this is the cable, not Hi-Z), REG0B 0x02 (no input), REG11 0x00 (no VBUS).
Gauge: Discharging, -394 to -491 mA, 4.118 → 4.093 V, 100 %.