[GIT PULL 3/5] ARM: tegra: Core SoC changes for v4.5-rc1

[Olof Johansson]

On Wed, Dec 30, 2015 at 04:46:47PM +0100, Thierry Reding wrote:
> On Tue, Dec 22, 2015 at 11:54:46AM -0800, Olof Johansson wrote:
> > On Thu, Dec 17, 2015 at 01:50:37PM +0100, Thierry Reding wrote:
> > > Hi ARM SoC maintainers,
> > > 
> > > The following changes since commit 8005c49d9aea74d382f474ce11afbbc7d7130bec:
> > > 
> > >   Linux 4.4-rc1 (2015-11-15 17:00:27 -0800)
> > > 
> > > are available in the git repository at:
> > > 
> > >   git://git.kernel.org/pub/scm/linux/kernel/git/tegra/linux.git tags/tegra-for-4.5-soc
> > > 
> > > for you to fetch changes up to a262e87ff354f12447bb6268bd63edf7ba1c20e0:
> > > 
> > >   ARM: tegra: select USB_ULPI from EHCI rather than platform (2015-11-24 16:47:26 +0100)
> > > 
> > > Thanks,
> > > Thierry
> > > 
> > > ----------------------------------------------------------------
> > > ARM: tegra: Core SoC changes for v4.5-rc1
> > > 
> > > The big thing here is Tegra210 support, which is really only the Kconfig
> > > symbol. Other than that there's a few miscellaneous fixes.
> > 
> > Hi,
> > 
> > We merge arch/arm and arch/arm64 and drivers/bus contents through separate
> > branches, so having them come in from you in one combined branch isn't working
> > very well.
> > 
> > If you don't have any downstream users of this I can just cherry-pick them into
> > the appropriate branches, but if you do, please send separate requests for
> > those three categories.
> 
> I don't think that'll work. Or at least, the outcome will be almost the
> same because of interdependencies between the branches.
> 
> For the patches in this branch:
> 
>       soc/tegra: Provide per-SoC Kconfig symbols
>       soc/tegra: Add Tegra210 support
> 
> These are dependencies for the following:
> 
>       ARM: tegra: select USB_ULPI from EHCI rather than platform
> 
> Whereas these two:
> 
>       ARM: tegra: Ensure entire dcache is flushed on entering LP0/1
>       amba: Hide TEGRA_AHB symbol
> 
> can be applied in any order. So I could send you a pull request with
> only the drivers/bus patch, and another one with the dcache flush fix.
> 
> But that's about it. The Kconfig changes are dependent on each other,
> hence will need to be brought in via a common ancestor, unless you want
> to resolve a rather annoying conflict in arm-soc.

Doing the refactoring/cleanup in a separate new branch that both of them build
on wouldn't be a bad way to do it in the future. That way dependences would
still flow in the direction we're used to (cleanups/refactor -> soc -> driver
code).

> As for the for-4.5/dt branch that has a dependency on both the
> for-4.5/clk branch for a DT include file and on the for-4.5/soc branch
> for the new ARCH_TEGRA_210_SOC Kconfig symbol. I suppose it wouldn't be
> strictly necessary to pull in the branch because it is a ("soft") make
> dependency and the Tegra210 DTBs will simply be ignored unless the
> for-4.5/soc branch is also merged. That's slightly confusing because
> then there's a branch with DTS files that can't be built.

The sorting of 64-bit patches into separate branches is sort of new, in the
past we grouped all 64-bit contributions in one branch. So yeah there would be
some dependencies here. I'm personally OK with things not being buildable at
first (or, having nothing to build in the case where SoC is merged first).

> Anyway, let me know what you prefer and I'm sure I can work something
> out that will suite you.

I think the above could be worth considering for now.

Given that the code has been in -next through your own tree, and we're
at a very late phase in the release cycle (and I've owed you this answer
for a while, delayed due to travel and being sick), I'll merge in all
the tegra contents into one late/tegra branch that we'll aim to send in
during this merge window. We've occasionally done that on other platforms
as well when dependencies have been messy.


-Olof