Re: [PATCH 1/3] dt-bindings: arm/marvell: ABI unstability warning about Marvell 7K/8K

On Wed, Feb 24, 2016 at 2:10 PM, Thomas Petazzoni
[off-list ref] wrote:

Hello,

On Wed, 24 Feb 2016 17:25:46 +0000, Mark Rutland wrote:

quoted

On Wed, Feb 24, 2016 at 04:16:45PM +0100, Thomas Petazzoni wrote:

quoted

Since we are still very early in the support of Marvell Armada 7/8K
chips and those chips are significantly different from earlier 32 bits
Armada SOCs, it is difficult to commit to Device Tree ABI stability at
this point.

I don't see why difference to an existing SoC means that we cannot
guarantee support for this DT. Why does that contribute to instability?

If there's something we're unsure about, we should be seeding the DT
with sufficient information to do the right thing(TM) in future. What
areas in particular do you think are likely to be problematic? Is there
anything in particular that you think that approach doesn't work for?

At this point, we have incomplete information about the HW capabilities
and register layout. Our past experience has shown that when we start
supporting SoC that early,

that early ...?

You should fix the incomplete information problem. I pushed back on
this, you got more information and already the binding is closer to
reality.

quoted

From my PoV, if we're so uncertain about a DT binding that it cannot be
kept supported, it is a prototype not yet fit for mainline.

Are you really serious when you say that ?

The Linux kernel community has been encouraging hardware vendor for
*years* to get their code merged as early as possible. And now that
they are doing it, you are telling them that they shouldn't submit
their code to mainline ?

Mark didn't say don't submit. First, there is value in submitting even
if not accepted immediately and you have to carry some patches for
some time. It was also suggested that you err on the side of less
information in DT if things are uncertain and you have not done that.

If that's what you want, then sorry, but I don't buy this argument, and
I will continue to encourage HW vendors to submit their code early,
even if it means that indeed in the early stages, things remain a bit
in-flux and need to be refined progressively.

You seem to really be in your ivory tower and not seeing the reality of
an SoC development and how to push the support for it in mainline as
early as possible, so that the BSPs shipped to actual customers are as
close to mainline as possible. Together with the Marvell engineers, we
are doing a huge amount of work to ensure that this vendor BSP is as
close as mainline as possible. And you're basically telling us we
should this effort ?

Come on, be serious.

quoted

quoted

Therefore, this commit adds a warning to the Marvell 7K/8K DT binding
documentation to indicate that the bindings are for the moment
unstable.

As with the other thread, I'm strongly opposed to this mentality.

Many other platforms have made this choice: Atmel, Marvell Berlin, and
probably others.

Atmel has had a year now with that statement. Time to remove it IMO.

I think we need to only allow this at a binding level and not at SoC
family level and also require a timeline to marking stable.

I think it should be up to each platform vendor to decide whether they
care about DT stability or not, and at which point in time in the life
cycle of their platform they want to claim the DT bindings as stable.
You might believe that DT bindings are stable, but they are not. They
change constantly, in very subtle way, and nobody ever tests DT from
kernel version N-x with kernel version N. On any real-life platform, it
simply doesn't work, unless that platform is 5+ years old.

I made a talk at ELC about DT stability being a fairy tale, and I
remember very well Rob Herring coming to me during the social event
afterwards, telling me that it has never been said that all platforms
should conform strictly to this DT stability requirement.

True, but it is still an exception and I never said I wouldn't push back.

So please get to an agreement between DT binding maintainers. And stop
saying this ridiculous statement that HW vendors should stop submitting
their code to the mainline.

You want us to agree, then you won't like the answer. Bindings must be
stable. I'll allow exceptions to that, but not without push back.

In general, if there is disagreement about whether stability is
required, then it is required. See the recent sunxi discussion. That's
more between users and platform maintainers though.

Rob
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html