[Ksummit-discuss] [MAINTAINER SUMMIT] How can we treat staging drivers better?

[Laurent Pinchart]

Hi Tim,

On Monday, 10 September 2018 22:22:05 EEST Tim.Bird at sony.com wrote:
> > On Monday, 10 September 2018 21:52:30 EEST Takashi Iwai wrote:
> >> On Fri, 07 Sep 2018 21:44:54 +0200, Mauro Carvalho Chehab wrote:
> >>> Em Wed, 05 Sep 2018 15:35:53 +0200 Takashi Iwai escreveu:
> >>>> The staging driver is a wonderful process to promote the downstream
> >>>> code to the upstream, but I have doubt whether it's working really
> >>>> as expected for now.
> >>>> 
> >>>> - Often the drivers live forever in staging although they should
> >>>>   have been moved to the upper, properly maintained, subsystems.
> >>>> 
> >>>> - Code changes in staging are mostly only scratching surfaces, minor
> >>>>   code style cleanups, etc, what checkpatch suggests.
> >>>> 
> >>>> - There are little communications with the corresponding subsystem;
> >>>>   already a few times I was surprised by casually finding a staging
> >>>>   driver code by grepping for preparing API changes.
> 
> ...
> 
> >>>> - Then some drivers are pushed back after long time stay in staging
> >>>>   (lustre is the recent remarkable case);
> >>>>   it's understandable, but is definitely no happy end in both sides,
> >>>>   after all.
> >>> 
> >>> We had a recent case: the (really big) atomisp driver.
> >> 
> >> What was the reason of drawback, BTW?
> > 
> > Intel pushed a huge code drop that clearly required a staging period, and
> > then simply left it bitrot. No developer was committed to perform the work
> > needed to de-stage the driver. I don't know whether priorities had changed
> > internally or if there were no resources from day 1, but in the end it's
> > pretty clear that if we had known beforehand of the outcome, the driver
> > would likely not have been even a staging candidate.
> > 
> >> I think it'd be helpful if we can gather more data, e.g. good examples
> >> to show how it can succeed, as well as anti-patterns to learn what
> >> makes things failing.
> > 
> > Anti-pattern number one: push a large driver to staging knowing that you
> > won't work on it anymore, and expect the community to do your work for
> > free.
> > 
> > I would have expected a company like Intel to know better. Or, rather
> > sadly, I probably have given up on such expectations already :-S
> 
> I thought staging was where work was done on drivers that were submitted
> to Greg's Free Linux Driver project (https://lkml.org/lkml/2007/1/29/345)
> (among other things).
> 
> Is that project still active?  Do things still move through staging as
> originally intended?  Am I conflating two different things?

I'll let Greg answer.

> I'm not sure if Intel's driver qualifies as a Free Linux Driver project or
> not.

$ git show a49d25364dfb | diffstat -s
 920 files changed, 204645 insertions(+)

That may be part of the answer :-)

> I can think of other companies, less experienced with kernel work than
> Intel, that might still benefit from the Free Linux Driver project, and
> staging.
> 
> I thought the whole idea of staging was that something was better than
> nothing, and that cruddy code could be taken even if the hardware vendor
> didn't have the skills or inclination to do proper mainlining.

I'd say that in that case staging doesn't work for the company's benefit, but 
for the community's benefit. For smaller drivers that are deemed useful by the 
community, when the company manufacturing the hardware isn't helpful 
(regardless of the reason, I'm not trying to blame anyone here), and if there 
are community members willing and able to do the work, staging is certainly 
helpful, and better than nothing. For a 200k+ lines of code driver in a very 
bad state to start with, I can't really picture how anything could be done in 
someone's free time.

I don't know how the atomisp ended up in staging and whether it was an attempt 
from Intel to tick a "upstream driver" box or not, but one thing that is clear 
is that they haven't put the necessary resources behind it.

-- 
Regards,

Laurent Pinchart