[PATCH v2 06/12] ARM: mcpm: change max clusters to 4

[Dave Martin]

On Tue, Apr 15, 2014 at 02:45:05PM +0800, Haojian Zhuang wrote:
> On 11 April 2014 10:39, Nicolas Pitre <nicolas.pitre at linaro.org> wrote:
> > On Thu, 10 Apr 2014, Dave Martin wrote:
> >
> >> On Tue, Apr 08, 2014 at 04:00:46PM +0800, Haojian Zhuang wrote:
> >> > In order to support 4 clusters with 4 Cortex A15 Cores in each cluster,
> >> > enlarge maximum clusters from 2 to 4 in MCPM.
> >>
> >> CC Nico on mcpm patches please.  I'm happy to be CC'd as well.
> >
> > Indeed.
> >
> >> > Signed-off-by: Haojian Zhuang <haojian.zhuang at linaro.org>
> >> > ---
> >> >  arch/arm/include/asm/mcpm.h | 2 +-
> >> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >> >
> >> > diff --git a/arch/arm/include/asm/mcpm.h b/arch/arm/include/asm/mcpm.h
> >> > index 608516e..68f82cf 100644
> >> > --- a/arch/arm/include/asm/mcpm.h
> >> > +++ b/arch/arm/include/asm/mcpm.h
> >> > @@ -20,7 +20,7 @@
> >> >   * to consider dynamic allocation.
> >> >   */
> >> >  #define MAX_CPUS_PER_CLUSTER       4
> >> > -#define MAX_NR_CLUSTERS            2
> >> > +#define MAX_NR_CLUSTERS            4
> >>
> >> Because of the need for alignment to the biggest cacheline size in the
> >> system, the MCPM low-level locking structures consume a non-trivial
> >> amount of memory.
> >>
> >> Therefore, I'm not keen on the idea of simply increasing this #define
> >> every time a platform appears with a larger number of clusters.
> >> If hip04 is not built into the kernel, this just wastes memory.
> >>
> >> I'll leave it to Nico to decide whether we can increase the #define
> >> to 4 or whether this needs a proper fix now.  Ideally, we would have
> >> a way of choosing the maximum value required by the set of boards built
> >> into the kernel, or switch to dynamic allocation.
> >
> > I think we should go with the ability to select a maximum based on the
> > configured platforms. That could be as simple as having a
> > CONFIG_MCPM_QUAD_CLUSTER symbol to be selected by those platforms that
> > need it.
> >
> 
> I'll set MAX_NR_CLUSTERS from Kconfig.
> 
> If BIG_LITTLE isn't enabled & HIP04 is enabled, MAX_NR_CLUSTERS could
> be set to 4.

Perhaps the bL switcher should simply refuse to enable itself if there
are more than 2 clusters, or 2 clusters that are identical.

There's no reason why HIP04 and b.L switcher support shouldn't be built
into the same kernel image, since the decision about whether to enable
the switcher is made at runtime.

Cheers
---Dave