[PATCH 0/4] Introducing Exynos ChipId driver

[Arnd Bergmann]

On Sunday 11 May 2014 18:47:28 Olof Johansson wrote:
> > > Also for platsmp.c and pm.c I can think of following approaches
> > > 1: Keep these macros till we get generic solution?
> > > 2: Allow chipid driver to expose APIs to check SoC id and SoC revisions 
> > > till we get
> > > generic solution. So that at least we can remove #ifdef  based macros
> > > as soc_is_exynosXYZ.
> > > 3: Use of "of_flat_dt_is_compatible" or similar APIs in these machine files 
> > > till we get
> > > generic solution. For some cases where we want to know SoC revision let us
> > > map chipid register and get revision there itself.
> > > 
> > > Please let me know what approach you think will be good?
> > 
> > I think 1 or 2 would be better than 3. Between those two, I'm undecided,
> > but I think either way the SoC specific values would be better kept in the
> > mach-samsung directory than in plat/cpu.h or linux/exynos-chipid.h.
> 
> The generic solution is already there: of_machine_is_compatible() is perfectly
> sensible to use for _some_ of these inits. Cpufreq is one of the few that comes
> to mind, and maybe some of the platsmp and pm stuff.
> 
> Note that none of them should be used in runtime, i.e. you should only use them
> at probe/setup time and maybe have a local state in the driver if needed.
> 
> I'd rather get people used to that format than everybody needing to implement
> a chipid driver now too, especially on platforms that might not even have a
> suitable chipid block to base a driver around -- not to mention having to
> fill the namespace with is_soc_*() stuff.

I was coming from the other angle: exynos is already using soc_is_*() in too
many places. I'd like to first see the ones cleaned up that really should be
doing something else because they have a device-local ID to look at.

If we end up with a couple of instances that don't have a good alternative,
we can use of_machine_is_compatible() for those, but I'd like to avoid doing
a blind conversion because that would likely lead to more instances in the
future, not fewer.

I agree that we should have to introduce new chip ID drivers on other
platforms for the purpose of finding out the SoC version, especially not
with private APIs.

	Arnd