Oh, thanks. I missed the "valid also when the pin is used as an input to an enabled analog peripheral" part.
Solved then.
Don't beat yourself up. Although I knew that I was looking for that specific information, I overlooked it last night. I expected and wanted to give exactly that answer.
[rant]
I had one major project on the AVR32AP, back in 2008. Critical omissions in the datasheet of that then new product led to several redesigns. We started out with 6 layers and ended with 12 layers. One crucial problem was the power supply during peak performance. MCU was fine with basic hardware tests, but tended to crash when the Linux application went full power.
Atmel pulled the same stunt in 2004 with the AT91RM9200.
I hate eventually seeing errata that burned thousands of Euros of my money.
Looking back at those two old Atmel components, I can only say that STM32 is gold by comparison.
[/rant]