Re: [Ksummit-discuss] [MAINTAINERS SUMMIT] Challenges in Upstream vs. Embargoed Development in Intel Graphics.

[Sean Paul <seanpaul@chromium.org>]

On Thu, Sep 6, 2018 at 6:43 AM Linus Walleij <linus.walleij@linaro.org> wrote:
>
> On Thu, Sep 6, 2018 at 12:25 PM Arnd Bergmann <arnd@arndb.de> wrote:
>
> > I think in the generalized case, you also want the reverse, but
> > that may be harder: When targetting specific software products
> > that you want to integrate your code, there should be a deadline
> > for the latest point by which code needs to be posted in
> > public.
>
> This brings in the process of procurement, as in how companies
> making products source their misc hardware like sensors,
> touchscreens, displays, FPGAs or whatnot.
>
> Maybe this is obvious.
>
> It happened at one point that we were sourcing hardware from
> a third party, and it was pretty complex and I asked procurement
> to put a demand on the company to provide upstream support
> so we could just grab the latest kernel and use that driver.
>
> I heard other very FOSS-oriented companies ask the same
> and is pretty much what I've heard people like Jon Masters
> and the Chromebook people say in relation to upstream first
> (they can slam me if they disagree) - others also want an
> upstream first approach from their component suppliers and
> it is going to be part of the procurement process so having
> upstream first is going to be a competitive advantage or
> even strict requirement for the component vendor.

Speaking really only for display, but most of this applies to other areas:

Pretty much this. Chromebook display stack must use drm/kms, and the
patches must* go upstream before landing in our kernel. Looking at our
4.14 branch [1], most of the commits there are
UPSTREAM/BACKPORT/FROMGIT/FROMLIST prefixed which means they've landed
upstream or have been reviewed upstream.

We carry downstream patches in a separate working branch shared
between developers and rebase on our 4.14 kernel regularly. Everything
in the working branch should be already sent upstream or being
polished for upstream. Reviews are done on the list. Once a patch
lands upstream, it's backported to our kernel and removed from the
working branch.

Sean

[1]- https://chromium.googlesource.com/chromiumos/third_party/kernel/+log/chromeos-4.14/drivers/gpu/drm

* There are always exceptions

>
> As it happened in my case the vendor was very unhappy
> with this and unused to this kind of requirements. (They have
> since changed their attitude so no-one needs to be outed.)
>
> What I realized was that instead of being "hard" on vendors
> with this, I could gain more by being let's say "firm".
>
> I required that in order to procure their component, they
> should present an upstreaming strategy, and post an initial
> patch set for the specific product before we would agree
> to procurement. This was more of a gentlemen agreement
> than a hard contract clause, but it worked very well to
> transform that company, and I think it is a good way, because
> being too hard can be counter-productive, I guess it comes
> from simple diplomacy, people do not like threats.
>
> Yours,
> Linus Walleij
> _______________________________________________
> Ksummit-discuss mailing list
> Ksummit-discuss@lists.linuxfoundation.org
> https://lists.linuxfoundation.org/mailman/listinfo/ksummit-discuss