I checked: our firmware's USB_OTG routing for 2.4 is identical to Turing's original, and the logic says nodes 1–4 should all work in host mode. Node 1 working means the BMC side of the path is fine, so I'd like to find out what differs for 2–4. Two questions and a quick test:
Are all four nodes the same module (RK1?) on the same OS image?
With the dongle plugged in, on the BMC:
tpi usb host -n 2
tpi usb status
cat /sys/bus/platform/devices/usb-port-power/state
mount -t debugfs none /sys/kernel/debug 2>/dev/null; grep -E 'PG[0-5]|usbotg' /sys/kernel/debug/gpio
then on node 2:
lsusb, dmesg | tail -n 30, and cat /sys/class/usb_role/*/role 2>/dev/null
Same again with -n 3.
If the BMC's pin levels are right but the node shows nothing at all in dmesg, it's the module not taking the host role. If node 2 works and 3/4 don't, it points at the board's second USB switch chip.
For your init container, once it routes: curl -su user:pass 'https://
/api/bmc?opt=set&type=usb&mode=0&node=
' (that API counts nodes from 0), or tpi usb host -n N --host
. The setting survives BMC reboots.