[PATCH v7 17/47] dt-bindings: memory: tegra20: Add memory client IDs

[Krzysztof Kozlowski]

On Thu, Nov 26, 2020 at 07:02:55PM +0100, Thierry Reding wrote:
> On Thu, Nov 26, 2020 at 06:39:22PM +0100, Krzysztof Kozlowski wrote:
> > On Thu, Nov 26, 2020 at 06:26:05PM +0100, Thierry Reding wrote:
> > > On Wed, Nov 04, 2020 at 07:48:53PM +0300, Dmitry Osipenko wrote:
> > > > Each memory client has unique hardware ID, add these IDs.
> > > > 
> > > > Acked-by: Rob Herring <robh at kernel.org>
> > > > Signed-off-by: Dmitry Osipenko <digetx at gmail.com>
> > > > ---
> > > >  include/dt-bindings/memory/tegra20-mc.h | 53 +++++++++++++++++++++++++
> > > >  1 file changed, 53 insertions(+)
> > > 
> > > Is there any chance you could drop these dt-bindings include patches
> > > (17, 18 and 19) so that I can pick them up into the Tegra tree? The
> > > device tree changes that I was going to pick up depend on this and
> > > fail to build if applied as-is.
> > > 
> > > I was looking at your linux-mem-ctrl tree and had initially thought I
> > > could just pull in one of the branches to get these dependencies, but it
> > > looks like the dt-bindings patches are on the for-v5.11/tegra-mc branch,
> > > which the ARM SoC maintainers wouldn't like to see me pull in for a
> > > dependency on device tree changes.
> > 
> > Partially you answered here. :) Since you should not pull my branch into
> > a DT branch, you also should not put these include/dt-bindings patches
> > there.  SoC guys will complain about this as well.
> > 
> > These patches are also needed for the driver, so if you take them, I
> > would need them back in a pull request. SoC folks could spot it as well
> > and point that such merge should not happen.
> > 
> > > If this is all fixed at this point, I'll just have to push back the
> > > device tree changes to v5.12, or perhaps see if the ARM SoC maintainers
> > > are willing to take a late pull request that's based on v5.11-rc1.
> > 
> > Yeah, that's a known problem. I asked about this Arnd and Olof in the
> > past and got reply with two solutions:
> > 1. Apply current version of patch without defines, just hard-coded
> >    numbers. After merging to Linus, replace the numbers with defines.
> > 
> > 2. Wait with DTS till dependencies reach Linus.
> 
> What I've done occasionally in the past was to put these kinds of
> patches into a separate "dt-bindings" branch that I could use to resolve
> dependencies from device tree files. The ARM SoC maintainers never had
> any issues with that approach.
> 
> I guess this is a bit of a special case, because the DT includes are
> ultimately really a part of the device tree, so mixing them both isn't
> problematic.

Indeed, that way could work... and no one would spot it. :) Many times
these headers were for clock symbols so if they go via SoC/DT tree,
merge back to clock tree could be accepted.

Best regards,
Krzysztof