On Tuesday 15 May 2007 13:07, Andrea Conti wrote:
> It looks like we're getting there :)
>
> This patch (against the stock openwrt-svn) combines the ideas from David
> and Claudio (with some fixes in the byte ordering) into something that
> can write a bootable kernel image to the flash.
>
> However, the resulting kernel cannot read the pre-2.6.17 yaffs format
> because of the different blockStatus encoding.
>
> While I can get two-way compatibility by also including the block status
> byte in an oob_free range and handling it at the yaffs level, I don't
> know if it doing that is a good idea.
>
> What do you think?
This whole area is unfortunately rather messy. The oob handling in mtd is
often broken and messy and there are many details to get right. Since jffs2
does not use oob, changes don't get fixed there unless yaffs-inclided people
do things.
This leads to people trying to work around problems on the yaffs side which
might not be the "correct and pure" place to fix these issues, but is often
the simplest. As a result there is no standard mechanism that works for all.
I do have some inband-tags code which is still experimental (though looks
pretty robust). This does not use oob for the tags. This makes for simpler
integration (no oob headaches) - which is a bit of a cop-out. Power-of-2
alignment is lost which causes some performance loss, but likely not much.
inband-tags uses yaffs2 mode even on 512-byte page devices which makes
checkpointing available on 512-byte page devices.
When you guys have a working solution could you reduce the problem, causes and
solution to a single doc. That would really help people in the future, as
well as fix what is broken.
Thanx
-- Charles
(text/plain)