updates — alarm-tegra-x1 (JG2HQZWB)

Joy-Con input over the rails, and two silent build traps found on the way. JOY-CON INPUT (untested on hardware) - Ported switchroot's patch 0016 as drivers/joycon.c, installed by a new install_switch_joycon() the same way the panel driver is, so it is reproducible rather than hand-applied. - Device tree: the rails are UARTs -- serial@70006040 (uartb, right Joy-Con) and serial@70006200 (uartc, left), with their pin groups. Those are ordinary pinmux pins, not padctl lanes, so they cannot disturb the USB/PCIe lanes the console depends on. - Deliberately did NOT ship switchroot's patch 0019 (hid-joycon.c): it binds HID USB/Bluetooth Nintendo devices, which mainline's hid-nintendo already covers. It does nothing for the rails. Two drivers for one job. - Port was small: power_supply_config lost of_node, serdev receive_buf returns size_t now, platform_driver.remove returns void now, six helpers became static. - NOT tested on hardware yet. It builds clean and is in the kernel; nothing has driven a Joy-Con. TRAP 1: a symbol can be set and the code still never built - CONFIG_JOYSTICK_JOYCON=y was present in .config AND include/config/auto.conf, while joycon.o was never compiled: drivers/input/Makefile gates the whole joystick/ directory on CONFIG_INPUT_JOYSTICK, which was off. kconfig never complained because joystick/Kconfig is sourced unconditionally, so the symbol was visible and settable -- the gate is only in the Makefile. - verify_required_options() could not catch this: reading .config proves a symbol survived kconfig, not that any code was built. Added verify_built_in_drivers(), which reads the built vmlinux for a string each installed driver defines and fails the build if one is missing. TRAP 2: mainline cannot invert the rails' TXD/RTS - switchroot's DT sets nvidia,invert-txd / nvidia,invert-rts. Mainline's serial-tegra.c defines TEGRA_UART_IRDA_CSR and never uses it. The DT here omits them and tests without inversion first; the bit layout is authoritative in the vendor tree if it turns out to be needed. ALSO - kconfig duplicate-line bug fixed: new options were appended next to their own "# CONFIG_X is not set" line, leaving two contradictory lines per symbol. - install_switch_patches() added so edits to upstream files are re-applied; it immediately found kernel-boot-markers.patch half-reverted, with its head.S markers silently gone. - Touchscreen: documented as UNRESOLVED. It answers on I2C and queues CONTROLLER_READY, but returns an error event whose payload reads as "bad code" and reports chip id 0x0000. Three real bugs were fixed on the way (unmuxed i2c3 pins, an interrupt storm, a poll-thread ordering bug); two hypotheses are recorded as wrong so nobody retries them.

first release