Re: APEX, NSLU2, skip stuff

[Marc Singer <elf@buici.com>]

On Mon, Aug 14, 2006 at 09:02:33AM +0930, Rod Whitby wrote:
> Marc Singer wrote:
> > On Mon, Aug 14, 2006 at 12:25:43AM +0200, Martin Michlmayr wrote:
> >> * Marc Singer <elf@buici.com> [2006-08-13 12:44]:
> >>> All good.  I flashed the image the way I flash any upgrade.  Since
> >>> it booted, I believe it's all good.
> >> So, I can confirm that it also works for me.  I've booted both into
> >> debian-installer and into a Debian system, and I also tested a kernel
> >> which is bigger than 1 MB. :-)
> >>
> >> I noticed that APEX loads everything from a FIS partition into memory.
> >> Maybe it would be possible to honour the size entry from the Sercom
> >> header, but I don't think this is terribly important.
> 
> APEX should honour the data size entry in the FIS directory, which
> should be correctly written by slugimage.  The SerComm header is only
> there for compatibility with the Linksys/SerComm hacked up RedBoot.

That's what I do, the FIS data and skip the SERCOMM headers.  I
suggested that we abolish the SERCOMM header on the initrd, but Martin
doesn't like that idea.  Considering that nothing can load our
modified firmware except for APEX, there doesn't seem to be much
reason to keep the SERCOMM header on the initrd.  It does mean that
slugimage would need a little more work.  Does't matter to me either
way.

> > I'd rather not deal with that, but I could if we determine that it is
> > important.  Doing so will mean that this work with the skip headers is
> > void.  The SERCOMM header at the start of the kernel will be incorrect
> > because it refers to the size of the first chunk of the kernel.  The
> > time to copy isn't significant.
> 
> The SerComm header at the start of the first chunk should only refer to
> the size of the first chunk.  The SerComm header at the start of the
> second chunk should only refer to the size of the second chunk.  These
> SerComm headers should only be read by the Linksys/SerComm hacked up
> RedBoot.

Agreed.

> > Also, I don't know if the SERCOMM header is a standard part of the FIS
> > spec, or if it was added by SERCOMM.  (Something which can be
> > determined.)  If it is 'required' then we could add another layer of
> > headers to make it work.  In other words, we would headerize the
> > kernel, split it into two pieces, each with headers and store them in
> > flash.  Then. we'd add a second skip to the kernel partition that
> > skips the kernel1 SERCOMM header.  A little messy, but technically
> > correct.
> 
> It was added by SerComm, and we should leave it to them (i.e. we should
> put it in there for Linksys/SerComm RedBoot compatibility, but APEX
> should ignore it).

:-)