On Thu, 2008-08-28 at 18:43 -0400, Jeremy Katz wrote:
Compelling patches that are destined for the next Linus rev (... with
someone who's willing to own helping to fix them up if they break the
build) go in with some frequency. The big thing is that it's a
mentality shift. But the advantages you gain (both from thinking "get
kernel changes upstream Linus kernel" and "get packages changes in
upstream Fedora") end up making it so that going from Fedora 9->10->11
becomes a far less difficult process. And then maybe we can get to
where the actual build being shipped on the XOs has few to no forked
packages
Yup.
Also note that a properly configured kernel from Linus will now boot
(just not a generic one); but it doesn't have everything we would like
to have, much less patches to current drivers; for example, a dcon fix
got pushed last night.
On the device tree front, the issue is there are several versions out
there, PPC, apple and Sun and us; so you have the usual reluctance to
change. Andres will have some idea how this "discussion" is going.
Some of what we have to do, however, will need some baking before they
go upstream. Let me give you a concrete example.
Right now, resume time is dominated by USB initialization (accounting
for ~1 second out of our current 1.3 seconds). Some of it is slop, but
other parts should be done by lazy evaluation, and will require some
significant changes to the USB system. These will need some serious
baking (but eventually speed resume for everyone, even if not as fast as
we can go). And as the USB spec is poor, at best, and arbitrary busses
can be hung off of it, when this work occurs, its baking will need to be
careful. We believe our hardware is good down to 50ms, btw.
- Jim
--
Jim Gettys <jg(a)laptop.org>
One Laptop Per Child