Broken device trees for exynos in linux-next

[kgene@kernel.org (Kukjin Kim)]

Olof Johansson wrote:
> 
> On Thu, Aug 15, 2013 at 5:04 PM, Kukjin Kim <kgene@kernel.org> wrote:
> > I think, maybe we discussed about that? :) current exynos_defconfig
> cannot
> > support exynos5440 because of LPAE and I remember we decided LPAE and
> > non-LPAE should be separated. So as I commented before,
> exynos5440_defconfig
> > is needed. If you have any concerns, please let me know.
> 
> Having a SoC-specific defconfig makes no sense. You can run with LPAE
> enabled on A15 and A7-based systems even if they don't have enough
> memory to need it.
> 
Hmm, I'm not sure. If so, I'm wondering why LPAE is implemented with 'ifdef'
and why Red Hat and Canonical provide PAE enabled kernel separately in
x86...

> Really, what we want is to just turn on the LPAE functionality and
> keep everything else common. Forking into two defconfigs seems like
> the wrong idea, even if we did discuss it before. Having something
> like a config fragment to include would make more sense, since that
> could be shared across all platforms (and apply with
> multi_v7_defconfig for those who want to run that on LPAE as well).
> 
If we could make LPAE enabled defconfig for all ARM platforms, I'm fine. I
think your concern is creating SoC specific defconfig and I agree with you.
But I'm not sure how we can support LPAE enabled defconfig for ARM
platforms.

> Or, you know, just have your build script enable it without having an
> in-tree config fragment. That'd work too.
> 
I don't think so because the defconfig will be used by customer for their
product so I should provide some defconfig for exynos5440 in mainline...

> The main case where this isn't sufficient is on platforms where _all_
> memory sits above 4G, since you can't boot a non-LPAE kernel on those
> at all. It seems like 5440 has memory starting at 2GB so it's not one
> of those.
> 
5440 has 8GB memory by default.

So let me know which way is fine to us?

- Kukjin