bbt and bitflip

[Matthew L. Creech]

On Fri, Jun 24, 2011 at 3:55 PM, Artem Bityutskiy <dedekind1 at gmail.com> wrote:
>
> If you confirm that everything is great and robust when the on-flash BBT
> gets corrupted - the NAND core for sure notices any corruptions and
> falls-back to the traditional scanning method and restores the on-flash
> BBT - then I apologize for saying that I do not trust it. Also, I do not
> really know the details of this, so I may be completely wrong.
>

I'm still trying to pinpoint the cause of the BBT corruptions that I
encountered, but _if_ they were caused by bit-flips, then it
definitely appears that your initial statement was accurate: the
devices were in many cases unbootable because the corrupted BBT was
treated as legit.

This also seems to be the case when looking at the code in nand_bbt.c:
if an ECC error is encountered in read_bbt(), it prints a warning but
then continues to process the BBT data.  Maybe there's an additional
assumption that I'm overlooking, but at a glance it seems like that
would allow bit errors to accumulate and wrongly flag good blocks as
bad.

-- 
Matthew L. Creech