Re: [PATCH v8 00/18] Tegra124 CL-DVFS / DFLL clocksource + cpufreq

From: Thierry Reding
Date: Tue Apr 14 2015 - 10:43:55 EST

On Fri, Apr 10, 2015 at 02:11:57PM -0700, Michael Turquette wrote:
> Quoting Thierry Reding (2015-03-11 03:07:43)
> > Hi Mike,
> >
> > Have you had a chance to look at these changes to the Tegra clock
> > driver? If you're fine with it, I'd like to take these patches through
> > the Tegra tree because the rest of the series depends on them. I can
> > provide a stable branch in case we need to base other Tegra clock
> > changes on top of this.
> Hi Thierry,
> Clock patches (and corresponding DT binding descriptions and changes to
> DTS) look fine to me. Please add:
> Acked-by: Michael Turquette <mturquette@xxxxxxxxxx>
> I did have a question about the beahvior of clk_put in one of Mikko's
> patches but it should not gate this series. I'm just trying to find out
> if we have a bug in the framework or if the Tegra driver is a special
> case.
> Also I do not think a stable branch is necessary.

Given how this didn't work at all for v4.1, why don't we try to do this
the conventional way for v4.2? What I propose is that I collect all the
patches that I've had in these branches into a single branch that I can
send you shortly after v4.1-rc1. Then you can pull that into the clock
tree when you're happy and I can pull that into the Tegra tree again if
needed to resolve dependencies.

One problem that I foresee is that the EMC clock driver relies on some
symbols exported by the EMC driver, so it will fail to build on its own
if merged into the clock tree. Usually I'd say we can solve this by
merging a stable branch from the Tegra tree into the clock tree, but if
any of the ARM SoC maintainers then decides that, for whatever reason,
the Tegra pull request isn't good enough, you'd need to either redo the
clock tree or we slip in the changes via the clock tree anyway.

Perhaps a solution would be to implement stubs for the API exposed by
the EMC driver and merge that through the clock tree, which should give
us a tree that can be built but be a no-op at runtime. Then we can add
the actual code to enable EMC scaling in the Tegra tree by providing a
real implementation. That way we only have the Tegra tree depend on the
clock tree.

How does that sound?


Attachment: pgp1DwWPxHfSf.pgp
Description: PGP signature