Our latest batch of boards, about half failed UART communications tests. Investigation found that every hundred to maybe every thousand characters (8-bit no parity, 38.4Kbaud), FFs are transmitted instead of the desired characters, usually 2 of them, but as many as four.
Plugging in a ColdFire Multilink to debug, the problem disappears! Indicates that the debugger somehow cures the problem.
Further digging shows that CPU pin 43, TCLK, was the pin which the debugger affected. This is a very strange pin. It sits at 1.7v DC, and will push/pull 45 mA to get it to ground/3.3v! Including good boards. Most uPs won't drive this much.
A 10K pullup does almost nothing, of course. A 180-ohm pullup gets the pin to 2v and fixes 80% of boards. The rest need a capacitor (a 220 pic seems to work) to ground to get stable communication.
Of course, the documentation shows no connection whatsoever between this pin and serial output. Any ideas? Our boards are working, but only by "magic."