Hector Martin

Removing the short would instantly make it come back to life. But power cycling it (USB disconnect/connect) with the short intact didn't. However, if you shorted it with Vio off, then power cycled it, it was fine.

If you remove the short while it's dead and powered, it comes back. You can then restore the short and it's fine. Power cycle it, fine. But if you power it down, remove the short, put it back, then power it on again... still dead. It remembers whether it was dead at power-down.

I know what you're thinking, stray voltage on capacitors. Nope. I could power it down while shorted/dead, short/discharge everything, every pin on the board. Power it back on, still dead. 1-bit nonvolatile memory. ⸘‽

Eventually I figured out that our Vio LDOs had memory characteristics. Charge would remain on the pass transistor gate. Turns out, LDOs need power to *turn themselves off*. So if an LDO crowbars its own input rail, then it doesn't matter if you lower the enable signal.

The whole thing dying meant EN would go low. But that didn't matter. Without Vin, the LDO can't turn off. Powering down the whole Glasgow didn't matter. Power it on again, and current flows through the LDO before Vin can rise enough for it to function, and turn itself off.

The multi-pronged fix for this ended up being adding a large bulk capacitor after the USB current limit IC (whoops, missed that in the datasheet), and adding current limit resistors in series with Vio LDO outputs.

Glasgow revC1/2 can now withstand Vio shorts without not just not locking up, but not glitching at all. The 5V rail briefly dips slightly when you do that, but not enough to disturb the FPGA/FX2 supply rails or trigger the reset monitor.

I can't remember if the final component choice was such that shorting *both* Vio rails *at once* was safe, but if it isn't (which probably also depends on your USB cable/supply), the worst that will happen is the whole thing will reset, and instantly recover with Vio off.