New issues with the MMU

I had thought the MMU was pretty solid, since I gave it a good test after its v1.1 rework, and didn't find any issues. In testing the control unit, though, I've noticed some weirdness. Mapping addresses didn't work consistently. More often than not, about half the bits were incorrect. Initially I suspected I'd broken something in giving it some surgery last week to make it ignore control signals during resets, but a bit more probing with the multimeter suggested that the connections to both the A and B buses were just broken. A continuity test showed it was the connecter between the lower and upper backplanes that was the problem, and about half the pins just weren't connected at all. I suspect the join between the connectors and the board were very weak when I first soldered them, and repeated connecting and disconnecting has caused some to disconnect. Mercifully, all the bad joins seemed to be on the outside set of pins, so I could reach them with a soldering iron, which fixed that problem.

With that done, mapping addresses seemed to work, but then outputting the mapped address was still inconsistent. This turned out to be because the soldering on one of the physaddr bus drivers was bad. The solder paste hadn't melted properly, and the several pins of chip were only intermittently connected to the board.

You can see the dull grey colour around the highlighted pins, compared to the usual shiny solder joint on the others in the photo. Presumably the connections were better then the solder paste was wet, and that's why it passed the initial tests. A quick pass over with a soldering iron and some flux fixed these, and the MMU seems to be working again.