Linux 2.6.23-rc8 for OpenMoko

Harald Welte laforge at
Mon Oct 8 15:52:49 CEST 2007

On Mon, Oct 01, 2007 at 12:31:27AM +0200, Christoph Hellwig wrote:
> On Sun, Sep 30, 2007 at 08:05:27AM -0300, Werner Almesberger wrote:
> > Christoph Hellwig wrote:
> > > Are the openmoko developers to time contrained to send it upstead
> > 
> > You hit the nail right on its head :-( Well, I haven't even had time
> > to help with OpenMoko mainline for some months now, so it's really
> > just Harald who's in charge of that, and he's in charge of quite a
> > lot of things ...
> So do you guys need some help dealing with rmk's patch queue?  This
> kind of reshuffling patches and responding to comments is what I do
> for relaxing when I'm too tired of real work :)

yes, please. It would be really great to have you pushing this.

If we at some point find somebody who can do somilar for u-boot, we
really have one big thing less to worry about.

> I can't comment too much on the actual ondisk format because I'm not
> an expert on flash memory, but many revisions it went through for
> different technologies are at least a little alarming.  Also the
> ondisk layout for hardlinks doesn't look to very encouraging.  As
> far as the actual code is concerned it's a complete nightmare.  The
> read/write path is buggy in more ways than it has lines of code and
> needs to be ripped out and replaced by new code entirely, with much
> more use of existing standard linux functionality.  It's not that
> dramatic on the namespace side, but that code has quite a lot bugs
> in the software design aswell as layering problems.  Add to that
> "features" like variant symlinks that have been NACKed multiple
> times because they're better implemented using bind mounts and you
> get a really nice cocktail ;-)

well, I'm not arguing that yaffs2 is the greatest invention since sliced
bread, but with LogFS in a too early stage, and JFFS2 being suboptimal
for large-page-nand, combined with the performance issues of JFFS2,
there is little other choice we have for now.

Also, I've never argued that yaffs2 should go mainline. 

> > We use it on HXD8, which has tons of NAND.
> How large is 'tons'?  If it's really going into the Gigabyte range
> I'd suggest looking at logfs, otherwise a recent jffs2 with the
> scalability improvements from olpc should be fine.

yes, tons: think about 3GB or more raw nand :)

and with the statements from Joern about Logfs' early status, it's
really not an option for us yet.  With the next generation of devices
I'm more than happy to re-visit.

- Harald Welte <laforge at>         
Software for the world's first truly open Free Software mobile phone

More information about the openmoko-kernel mailing list