busybox `reboot bootloader` silently discards the word "bootloader

scope: generic · severity: trap · confidence: proven · subsystem: boot

sudo reboot bootloader does not reach the bootloader on a pmOS rootfs, and never did.

/usr/sbin/reboot is busybox. Its applet is reboot [-d DELAY] [-nf] — it takes no mode argument at all, so the word bootloader is silently discarded and you get an ordinary reboot. No error. No warning.

Every apparent success is something else. On taimen it was the A/B retry counter running out on its own and dropping the device into the bootloader anyway, which is why the command looked merely flaky for months rather than broken.

The kernel side is usually already fine: a PMIC pon node with mode-bootloader and a driver bound to it means the reboot-mode framework will write the mode to the register the bootloader reads. What is missing is userspace passing the mode string — reboot(2) with LINUX_REBOOT_CMD_RESTART2 rather than the plain RB_AUTOBOOT busybox issues.

porthole makes the syscall itself via python3 on the device:

142        = __NR_reboot on arm64 (generic syscall table)
0xfee1dead = LINUX_REBOOT_MAGIC1
0x28121969 = LINUX_REBOOT_MAGIC2
0xa1b2c3d4 = LINUX_REBOOT_CMD_RESTART2   (the one that carries a string)

~9 s to the bootloader, first try, and it does not burn a boot retry.

Set PORTHOLE_REBOOT_MODE_VIA_SYSCALL=1 once you have confirmed the pon node and its driver. Run it in the foreground and let the ssh connection die mid-call — that death is the success signal. Backgrounding it races: ssh tears the session down before the child reaches the syscall often enough that the request is simply lost.