kbd-disp-switch

Two PCs share one monitor and one USB keyboard/mouse dongle. Plug the dongle into a PC and the monitor automatically switches to that PC's HDMI input. Unplug it and the display stays put ("display follows arrivals only").

Built around udev + ddcutil, with a small ssh side-channel between the two hosts so that unplug/replug on the same host never causes a pointless monitor re-handshake (a 2–3 s black flicker on many monitors).

How it works

        +------------------+          +------------------+
        |  host A          |          |  host B          |
        |  HDMI-1          |          |  HDMI-2          |
        |                  |          |                  |
        |  udev rule       |          |  udev rule       |
        |  kbd_disp_switch |          |  kbd_disp_switch |
        |  (active)        |          |  (passive)       |
        +--------+---------+          +--------+---------+
                 |                              |
                 |  ssh (query + push)          |  reads pushed state
                 +------------------------------+

udev fires on dongle add (attach) and remove (detach).
On attach, a decision is scheduled (a systemd-run transient unit that sleeps SETTLE seconds, then decides). A quick unplug/replug cancels the pending decision.
On detach, the departure time (epoch) is recorded locally and pushed to the peer over ssh.
The decision is event-ordering, not a time window:
```
no-op iff my_D > 0 AND peer_D > 0 AND my_D > peer_D
```
where my_D = this host's last departure, peer_D = the peer's last departure. If my departure is the newest, the dongle just came straight back (round trip / flap) — the display never moved, so no write. Otherwise it's a real move — switch to MY_INPUT. Unknown peer state resolves to "switch" (safe default).

Why the ssh side-channel?

"Unplug → replug" is locally identical for (a) a round trip on the same host (no-op needed) and (b) a round trip through the peer (switch needed — the peer's attach already moved the display). Only the peer knows which happened, so the two hosts exchange departure timestamps. The hosts are asymmetric: one is active (queries the peer over ssh), the other passive (reads a state file the active host pushes). This avoids needing sshd on both hosts.

Why systemd-run?

udev kills leftover processes of RUN programs (cgroup kill — even setsid doesn't help), so a backgrounded decision job simply dies. A transient systemd-run unit survives and is cancellable via systemctl stop. The decision unit sleeps SETTLE internally because systemd-run --on-active timers proved unreliable (fired 6–60 s late in practice).

Why a custom DDC command?

Standard input switching (ddcutil setvcp 60 ...) is broken on recent LG monitors (returns "Invalid value"). The working method is the manufacturer side-channel: ddcutil setvcp xF4 x0090 --i2c-source-addr=x50 --noverify (HDMI 1) / x0091 (HDMI 2). See INSTALL.md for how to find the right command for your monitor.

Requirements

Linux with systemd (udev) on both hosts
ddcutil on both hosts
ssh client on the active host, sshd on the passive host
A monitor that supports DDC/CI input switching

Install

See INSTALL.md — step by step, including finding your device IDs, your monitor's DDC bus, and the right input-switching command.

Files

File	Purpose
`kbd_disp_switch`	main script (install to `/usr/local/sbin/`)
`kbd_disp_peer`	whitelisted ssh wrapper for the passive host
`kbd_disp.conf.example`	per-host config template (`/etc/kbd_disp.conf`)
`99-kbd-disp.rules`	udev rule (replace the USB IDs)

Configuration reference

Variable	Meaning
`MY_INPUT`	`1` → monitor HDMI 1, `2` → HDMI 2
`SETTLE`	debounce seconds before the decision (default 2)
`PEER_QUERY`	active host: `user@peer` for querying the peer's departure
`PEER_PUSH`	active host: `user@peer` for pushing our departure
`SSH_KEY`	automation key path (default `/root/.ssh/kbd_disp`)
`PEER_STATE_FILE`	passive host: local file the active host's pushes land in
`DDC_CMD`	DDC write command; `%s` = input value (default: LG side-channel)

Troubleshooting

The screen never switches on attach. Check the log (/var/lib/kbd_disp/switch.log) and that the udev rule matches: udevadm test --action=add /sys/bus/usb/devices/<dev>.
Detaches never fire. Remove events lack ID_VENDOR_ID/ATTR{idVendor} — the rule must match ENV{PRODUCT}=="<vid>/<pid>/*" (see the rule file).
The decision job dies. udev kills backgrounded jobs; the decision must run as a systemd-run unit.
getvcp 60 fails on an LG. Expected — use the side-channel command.
The monitor cannot report its current input. Many monitors (including the LG this was built against) return garbage for getvcp 60; the script therefore never reads the current input — it only writes on state change.

License

MIT — see LICENSE.