[PATCH 2/4] mtd: nand: Account the blocks used by the BBT in the ecc_stats

Brian Norris computersforpeace at gmail.com
Mon May 12 19:36:51 PDT 2014

On Mon, May 12, 2014 at 11:27:53PM -0300, Ezequiel Garcia wrote:
> On 21 Mar 08:57 AM, Ezequiel Garcia wrote:
> > Strictly speaking we should be updating the ecc_stats in the master
> > MTD object, with the blocks used by the bad block table.
> > 
> > This is already being done for bad and reserved blocks detected doing
> > the BBT search, but not for the blocks used by the BBT itself. This commit
> > adds the latter.
> > 
> > It should be noted that the ecc_stats structure is kept only for userspace
> > information, accesible through an ioctl. However, since the master MTD object
> > is not tied to any /dev/mtd{N} device node in the filesystem, there's currently
> > no way to retrieve this information.
> > 
> > This ecc_stats is used for the MTD partitions typically allocated and
> > registered by mtd_device_parse_register(). These have a device node, but scan
> > for bad blocks and updates the ecc_stats in a different code path.
> > 
> > Signed-off-by: Ezequiel Garcia <ezequiel.garcia at free-electrons.com>
> > --
> > For the reasons exposed above, it's not clear we should remove the ecc_stats
> > update in the master MTD altogether or simply take account of the BBT blocks
> > for consistency. I've chosen the latter, for it seemed a safer changer.
> > 
> > I'm open to discussion, though.
> Brian,
> Can you comment a bit on this one? Should I keep this change in v2?

I'm not really sure. I'm honestly having a hard time tracking all the
potentially-configurable knobs of nand_bbt.c. It looks like only
diskonchip sets a reserved_block_code, so some of the existing code
isn't really even tested widely. And like you mention, the ecc_stats
from the master MTD are not propagated directly to the partition (nor
should they be), so the stat is really unused.

I'm not 100% confident that we won't double-count any 'bbtblocks' in
your current patch. Maybe we should rewrite some of this stuff...

I'll look at this again when my eyes are fresh.


More information about the linux-mtd mailing list