On Tuesday 16 December 2008, Kevin Hilman wrote:
> > I suspect the dm355evm issue with reported bad blocks is
> > related to the 4-bit ECC mode used by the firmware loaded
> > on it.  It needs 10 bytes ECC data per 512 bytes data,
> > but evidently the resulting 40 bytes per page exceeds
> > what that version of Linux (and current ones??) allow
> > in the OOB area.
> >
> > So ... the ECC data is instead stored inline, which
> > means the first few bytes of OOB data on *good* blocks
> > will be used for data, which will probably look just
> > like a factory marker for a bad block.
> >
> > I think that'll be a mess to fix, if that's right.
> > Ideally the existing bad block tables can be extracted
> > and restored to a standard format.

Or the quick way to convert would probably be to just
use the "old" kernel to erase the whole device, which
will put 0xff values everywhere -- including OOB areas
of good blocks -- and then convert to a less quirky
layout.

The first BBT scan of that flash after erasing won't
find the existing BBT, but it should at least be able
to recover the list of factory-marked bad blocks since
the erase should have ignored them.


> IIRC, there was some discussion in TI about updating ubl/u-boot to fix
> the bootloader-kernel mismatch.  Anyone in TI care to comment?

I'd also be interested in knowing about any discussions
that have been had with the mainline MTD folks.  Those
NAND_ECC_HWx_y constants seem unique to the DaVinci tree,
and unlikely to be accepted in mainline.

I'd assume it *should* be noncontroversial to remove any
remaining limitation on how much ECC data can be kept in
the OOB area.

Once Linux can handle that cleanly, I'd expect U-Boot
(and its UBL) could change easily enough.  Updating
existing flash chips ... a mess.  Updating RBL might
be nice.  I'd expect a minimal change to handle larger
pages (4K vs 2K), not necessarily with any other updates
(like using only the OOB area for ECC) included.

- Dave


_______________________________________________
Davinci-linux-open-source mailing list
[email protected]
http://linux.davincidsp.com/mailman/listinfo/davinci-linux-open-source

Reply via email to