S03usbdev: derive USB serial number from SoC hardware UID

[JakeHillion]

The USB gadget serial number was hardcoded to '0123456789ABCDEF', meaning every NanoKVM device presented the same serial to the host.

Read the SoC unique identifier from /sys/class/cvi-base/base_uid on each boot and concatenate its two halves into a single hex string for use as the USB serial number. Falls back to the old hardcoded value if the UID file is not present.

This gives each device a stable, hardware-derived serial number without requiring any persistent state, and satisfies the USB Mass Storage requirement of at least 12 characters. This means you can query the USB interface on the machine the KVM is plugged into to find out which NanoKVM is plugged in, instead of that there is a NanoKVM plugged in.

Test plan:

Installed script on two NanoKVMs with:
scp kvmapp/system/init.d/S03usbdev root@kvm04:/etc/init.d/S03usbdev

Plugged into a Linux machine, before I got:

[jake@li-bl-eq14-1:~]$ nix shell nixpkgs#usbutils --command lsusb -v | grep NanoKVM -A 50
Bus 003 Device 006: ID 3346:1009 sipeed NanoKVM
Negotiated speed: High Speed (480Mbps)
Device Descriptor:
...
  idVendor           0x3346 sipeed
  idProduct          0x1009 NanoKVM
  bcdDevice            5.10
  iManufacturer           1 sipeed
  iProduct                2 NanoKVM
  iSerial                 3 0123456789ABCDEF

After the update, I get the same, but the serial number is now unique per device. I unplugged the devices and plugged them back in, this is now a consistent and unique identifier between my NanoKVMs from the host device.

[yuzi-co]

I had written the same change independently before finding this PR, so I have dropped mine. One difference is worth folding in here.

The -f test confirms the file exists, but it does not confirm that awk '{print $2}' returns anything. If the field is missing or the format differs on a variant, _serial ends up empty or short, and the empty string is then written as the serial number.

That matters more than it looks. As you note, mass storage requires a serial of at least 12 characters. A serial that fails that requirement does not degrade to the old behaviour — the gadget fails to enumerate, so the operator loses keyboard, mouse and disk rather than just losing the ability to tell two devices apart. The fallback would be the safer outcome in that case, but the current shape cannot reach it once the file exists.

A length check picks the fallback instead:

serial='0123456789ABCDEF'
if [ -f /sys/class/cvi-base/base_uid ]
then
    uid_serial=$(awk '{print $2}' /sys/class/cvi-base/base_uid | tr -d '_')
    if [ ${#uid_serial} -ge 12 ]
    then
        serial="$uid_serial"
    fi
fi
echo "$serial" > strings/0x409/serialnumber

${#var} is POSIX and busybox ash supports it, so it needs nothing extra on the device.

I have not seen a board where base_uid is malformed — this is defensive rather than a reported failure. But the cost of being wrong is asymmetric enough that I think it earns the four lines.

Otherwise this reads correctly to me, and the per-device serial is genuinely useful once you have more than one on a bench.