Is it wrong to write HIGH to a pin used as an open collector?

Question

I'm a software engineer playing with PICs and have an old PS/2 mouse that I am using (successfully), which has CLK and DATA connections as open collectors.

So I have configured the internal pullups on the relevant PIC pins and am setting TRIS to 1 to let the lines go high and setting TRIS to 0 and writing 0 to the relevant PORT bit to force it low.

However, I've seen several similar PIC-to-PS/2 mouse implementations that, when writing to the mouse, set TRIS to 0 and write 1 to the PORT bit instead of setting TRIS to 1.

As I understand it, either of these will result in the line going high (is that correct?), but I'm not clear if one is somehow "better/safer" than the other.

Should I just follow the examples of the (presumably) more experienced implementations and write 1's to the PORT when the PIC is in control of the bus, or should I always make the line high by tri-stating to high impedance and letting the pullup do it?

I hope that makes sense and I haven't mixed any terminology - any general advice about best practices for open collector connections would be welcome.

Answer 1

It's not really wrong to set the pin HIGH, but doing so may cause output contention. That is, you may get into a situation where your code is driving the pin HIGH, but the device at the other end is trying to drive the same pin LOW. The PS/2 mouse (and keyboard) protocol uses open-collector signalling to avoid contention. In an open-collector system, signals are assumed to float HIGH when they are not driven by a pin, i.e. when the pin(s) are high impedance. A pin can safely be pulled LOW by the device at either end of the wire.

So, it's probably best to use TRIS in this case.

Answer 2

There is one case where it is useful to set the pin HIGH.

Where the open-collector logic is relatively slow, either because the pull-up resistor must be high (to save power) or because it's connected to a long wire or PCB trace, an output can be driven HIGH for one clock cycle and tri-stated in the next. (These must happen as an atomic operation, e.g. with interrupts disabled)

This gives a relatively fast rise time, and limits the duration of any contention to be short enough not to cause damage. (and of course that must be verified via the datasheets)