To partially answer my own e-mail, it looks like the problem is some 
missing locking in some local extensions to the mtd nand driver that is 
causing the erase command to sometimes fail.

Ben Gamsa wrote:
> I'm seeing quite a few failed erases on my system using a Micro 
> 29F4G08AAC flash part.  This results in the bad block count growing 
> quite quickly.  However a scrub and rescan shows only the original 
> blocks marked bad out of manufacturing are actually bad.  Changing the 
> yaffs code to retry (just once) an erase that fails, always seems to 
> show the second erase succeeding (although I've only seen a handful of 
> erase failures since trying the change).  I've seen some discussion that 
> perhaps the bad-block detection algorithm is overly aggressive at 
> marking blocks bad, so I'm wondering if it's safe to retry the erase and 
> consider everything good if the second erase succeeds?
> 

-- 
Ben Gamsa       ben@somanetworks.com
SOMA Networks   312 Adelaide St. W. Suite 600 Toronto, Ontario, M5V1R2