No subject

So I'd actually recommend that Artem push this patch (patch 03) out of
'master' and into the 'brian' branch, then we can review the final
product there. In the end, I intended nandwrite to only have a single
buffer for OOB, instead of having a read buffer and an oob buffer. We
would simply pass our unmodified OOB data to mtd_write() and then to
the new MEMWRITE ioctl, where the kernel would do AUTO or PLACE
layouts for us depending on the mode we chose.

Of course, this is just the ideal approach; many kernels will not
support MEMWRITE yet, so they would fall back to the old code. That's
what patch 07 did; it cut out one buffer and moved the old layout code
into its own libmtd function, legacy_auto_oob_layout().
legacy_auto_oob_layout allocates its own buffer when needed; this part
is very inefficient (and actually has a memory leak now that I look at
it...) Perhaps the "legacy" handling should be reviewed a little more
to prevent too many trivial buffers and memcpy's.

If nothing else, I will at least need to send a v2 for patch 07 so we
free `tmp_buf' in `legacy_auto_oob_layout()'. But it'd be better just
to completely rework that function's buffers...

Brian