On Tuesday 29 November 2005 09:28, Martin Egholm Nielsen wrote:
> Can this occur, say, with a bad NAND? Would YAFFS/MTD puke up
> with a lot of checksum errors?

The most you could hope for is an uncorrected ECC. ECC can 
correct a single bit error within each 256 byte sub-page block 
of data. If the data was corrupted on the way out to flash, 
before the ECC was calculated, the ECC will be consistent with 
the data, if the corruption happen 'below ECC', say on the chip 
or i/o to/from the chip, then the ECC should be inconsistent 
with the data. If you can figure out the physical block/page 
address of the offending data, you could instrument the ECC code 
(in MTD) to show how the stored data and stored ECC measures up. 
If the ECC is good, then the corruption very probably happened 
'above' the ECC when the data was written. None of the above 
really answers the question of how/what introduced the 
corruption, it just narrows the scope.

-imcd