The bootloader reboot can drop the phone off USB entirely

scope: device:google-taimen · severity: trap · confidence: proven · subsystem: usb

Evidence

2026-09-04 19:10, porthole build fast. tkflash-boot ran reboot(RESTART2, bootloader) through the PMIC reboot-mode register; the phone reached the bootloader screen and stayed there, and for ten minutes lsusb showed neither 18d1:4ee0 (fastboot) nor the d001 gadget. Unplugging and replugging the cable enumerated it immediately, in the same bootloader session. Seen once, on the run recorded in issue #59.

Symptom — you asked for the bootloader, the phone went there, and then it is not on the bus at all. fastboot devices is empty and stays empty; lsusb shows no 18d1:4ee0 and no d001. Nothing enumerates, for as long as you are willing to wait — ten minutes, in the one case recorded. The screen says the phone is fine, which is what makes this read as a dead device rather than a dead cable.

Cause — not established. What is established is where it is NOT: the phone is up, the bootloader is running and it is drawing its screen. Only the USB side is missing, and it comes back without the phone moving. So this is the bootloader’s gadget failing to come up on that boot, not a hang, not a bad image and not the reboot being refused.

Do not spend the session on the theory. The observation that matters is that it is recoverable in seconds and that waiting is not what recovers it.

What to do — replug the cable. It enumerated immediately, in the same bootloader session, with nothing else changed. Then carry on; the device is where you asked it to be.

Do not reboot out of the bootloader to “reset USB”, and do not start a fresh porthole build fast to get back to a known state. The rung resumes: a run interrupted here has already pushed its modules, and tkpush-modules skips the push and goes straight to the flash when it finds the device in FASTBOOT with that exact module set recorded (issue #59). Starting over costs the four minutes of build and the bootloader wait again, three times over in the run this note comes from.

Related: a-refusal-is-not-a-hang, busybox-reboot-eats-the-mode-string.