What would it take to make a PS/2 keyboard interface hot-swap?

Question

A long time ago, I had to service some of the early Compaq Desktop PCs. They used to have fuses on the motherboard that blew when a customer hot-plugged a keyboard. I found this when googling on the topic:

The keyboard or mouse should not draw more than 275 mA from the host and care must be taken to avoid transient surges. Such surges can be caused by "hot-plugging" a keyboard/mouse (ie, connect/disconnect the device while the computer's power is on.)

Somehow, un-plugging or plugging in a keyboard can cause a surge. I don't recall having these issues with serial ports (RS-232, Appletalk, etc), parallel ports, VGA, etc.

I've always wondered. Why is there a surge risk here? What would have to change to make a PS/2 keyboard interface capable of hot-swap?

Answer 1

Somehow, un-plugging or plugging in a keyboard can cause a surge. I don't recall having these issues with serial ports (RS-232, Appletalk, etc), parallel ports, VGA, etc.

These don't power the equipment on the other end.

What would have to change to make a PS/2 keyboard interface capable of hot-swap?

There has to be a way to limit current to prevent a faulty device from shorting the power supply and frying the motherboard or simply crashing the PC due to a shorted +5V rail. Even if the device is not faulty, if it has a big enough capacitor, connecting it could cause inrush current and a dip on the +5V rail which could cause a crash.

A fuse is a crude way to do this.

USB was designed as hotplug, so it uses active current limiting. There is a whole family of chips to handle this. Here is an example.

Texas Instruments TPS2001D USB Power Distribution Switch is intended for applications where heavy capacitive loads and short circuits are likely to be encountered, such as USB. The TPS2001D limits the output current to a safe level by operating in a constant-current mode when the output load exceeds the current limit threshold. This feature provides a predictable fault current under all conditions. The fast overload response time eases the burden on the main 5V supply to provide regulated power when the output is shorted.

It also provides a FAULT signal to the computer so it knows what's happening.

To make PS/2 hotplug, that wouldn't be enough though.

You'd need a connector that makes contacts in a specific order: ground, then power, then signals. USB does this by making the pins that should make contact first longer. Without this type of connector, power and/or signal can make contact before ground, which means the device will receive IO signals before it has power, ESD protection diodes in the device and/or host will conduct current and it will power itself through its digital IOs, which can lead to fried ESD diodes, or worse latch-up. This requires special mitigation measures.

Also, you'd need the software in the PC to handle hotplug and register that a device has been plugged. With USB this is easy, the chipset does it automatically and raises an interrupt when a device is detected.

EDIT: Also an important part of the USB connector is the shield makes contact first, so any ESD event happens between shields and signal pins are not involved. This is probably the case with PS2 too, since the shield protrudes quite a bit.

The absolute worst case hotplug connectors are RCA audio (signal connects before ground), and of course the audio Jack (signal connects first AND all pins short together during insertion).

Answer 2

What it takes apparently is a desire to make the PS/2 ports capable of hot plugging. I recall a time when hot plugging a PS/2 keyboard or mouse would certainly mean the computer did not recognize the device. I don't recall ever seeing this causing damage that a reboot could not repair. Then came a point when hot plugging PS/2 devices just worked. This feature was seen on laptops first then desktop systems would allow hot plugging.

I have a couple USB to PS/2 adapters and they work fine with hot plugged PS/2 devices. One adapter has color coded ports where one port is mouse only and the other keyboard only. My other adapter will support a keyboard or mouse on either or both ports. This proves that nothing prevents device agnostic PS/2 ports, something I also recall seeing appearing on laptops first.

I do recall coming across an interesting article that discussed how early PS/2 controllers worked. They explained how a bad mouse could affect the keyboard and vice versa, why early ports might wig out with hot plugged devices, why they were device specific, and why new ports don't have these same failings. I wish I could recall where I saw this article so I could link to it and refresh my memory on how this all worked. I recall part of the issue was that it was typical for one chip to handle both ports, hot plugging would upset a delicate balance on what was a single circuit for both devices.

As other answers point out part of the problem is that the port supplies power to the device. Ports like VGA and parallel didn't have this same problem as they are output only and/or relied on the device to be self powered. Connecting power before the signal pins could mean the data got scrambled which might confuse the device as much as the computer. Perhaps a lucky or carefully handled hot plug would allow a device to work. Putting in the effort to build hardware that supports hot plugging avoids the need for this.

One style of Apple serial ports, the GeoPort, had a pin to power low power devices, I don't recall how well they handled hot plugging. There's Power-Over-Ethernet which I gather must support hot plugging. I recall some early Apple computers didn't like hot plugging on ADB. SCSI came in hot plug capable implementations, I have an old server in my basement that allows for this, but in most cases hot plugging SCSI risked expensive damage. Just more examples on ports that did or did not put in the effort to support hot plugging.

I know that my saying that lacking hot plug support in the spec is why early PS/2 ports don't handle hot plugging isn't much of an answer. I can imagine that there's several ways that newer PS/2 ports allow for hot plugging. USB and other ports make this easier to handle with varied lengths of the pins. PS/2 can do hot plugging without the differing lengths of pins. Making it work with same length pins means having filtering/de-bounce, and this can be done in hardware or software. It's possible they need a way to reset the device as a device confused by hot plugging is possible. Because of the various means hot plug can be handled I expect that some computers handle it better than others.

Answer 3

I'd made a circuit which the +5 V was connected through a resistor, sensing if anything was connected. If so, I'd slowly ramp up the voltage. The data lines would only be connected when the supply voltage had been stable for a second or so.

I'm not sure if the controller in the keyboard would start if the supply voltage rises too slowly, so that would be a compromise. Probably limiting the current to the 275 mA would be a sensible solution.