On Mar 22, 2006, at 4:17 PM, Garrett D'Amore wrote:

> Ah.  See this is where we get hung up.  The idea is that historically
> filesystems have needed to sit on top of a block device.  I'm not sure
> this is the best abstraction for flash devices.

Having a block device abstraction is nice for other reasons.  For  
example, maybe some region in flash has a raw binary executable image  
for the reset vector.  To update that, it might be nice to be able to  
simply do:

	dd if=newcode.bin of=/dev/flash0 obs=64k conv=osync

-- thorpej