I wrote: > Hmm, my wild guess would be that the introduction of read_subpage in > the NAND layer somehow caused this.
I've been poking around this a bit more. The problem seems to be that NAND_CMD_RNDOUT sometimes doesn't produce quite the right data. It looks as if this may be linked to the offset not being a multiple of four, but I need to gather a bit more data to confirm this pattern. A quick work-around is to always set "gaps" in nand_read_subpage. That way, it always reads the whole OOB data. - Werner
