On Tue, 2007-03-20 at 10:58 -0800, David Lang wrote:
I've seen no real proposals about how this could be done, so it's a
purely academic question. But I'm dubious. The block layer is optimised,
perhaps unsurprisingly, for block devices. Making it handle our special
case might be possible, but I don't really think it's likely to fly once
it becomes real code and not just mental self-abuse. Hell, we haven't
even got block _discard_ support merged yet, because it's too esoteric
for people to care about.
The MTD API does need to be re-thought. It's no longer quite so
unthinkable that we'll encounter flash sizes above 4GiB, and the way we
(theoretically) handle asynchronous erases while read and write are
synchronous is a bit icky. I'm not averse to using queues and making it
look a _bit_ like block devices in some respects, but in practice I
don't think it'll be very close at all.
The MTD API is intended to represent the raw capabilities of the
underlying flash devices, with all the bizarre restrictions and features
that the various types of flash chip have. If you want to use it as a
block device, that's what translation layers are for.
--
dwmw2
-
