On Wed, Sep 02, 2015 at 06:02:25PM +0800, Yakir Yang wrote:
> 在 2015/9/2 16:34, Thierry Reding 写道:
[...]
> >At the very least your code must compile when applied against a recent
> >upstream tree. I would also expect you to make sure the code works at
> >runtime, though, contrary to build testing, not everybody will be able
> >to verify that you've actually done so. It is ultimately your platform
> >maintainer's (i.e. Heiko's) responsibility to ensure that because they
> >will get to deal with user complaints if people can't run an upstream
> >kernel on the devices.
> 
> Oh, first time to know this rule. So I should work on Heiko's github
> kernel branch at the first time to start send upstream.

It's usually not necessary to rebase on a specific platform tree. Most
platform trees should feed into linux-next anyway, so linux-next would
be the appropriate base in almost all cases.

Note, though, that that's only true if you expect somebody else to merge
your code. The reason is that whoever will end up applying your patches
will likely apply to a tree that feeds into linux-next, and that way you
both end up having roughly the same base.

On the other hand if you are a maintainer yourself you should be keeping
a branch based on the latest -rc1. That's especially important if your
tree feeds into linux-next, because basing on linux-next will break very
horribly that way.

So for this particular case I would expect either Mark or Inki to apply
these patches when they're ready. Their trees should be based on the
latest -rc1. At least the Exynos DRM tree feeds into linux-next, so you
should be fine if you use linux-next as a base.

Mark, have you ever considered having your tree added to linux-next?

I'm beginning to think that we need to make that a requirement for all
DRM drivers so that we can resolve integration issues early on rather
than Dave having to deal with them when he pulls code in.

Thierry