Re: [PATCH] mach-shmobile: Emma Mobile EV2 SMP prototype code

On Wed, May 9, 2012 at 10:45 PM, Arnd Bergmann [off-list ref] wrote:

On Wednesday 09 May 2012, Marc Zyngier wrote:

quoted

On 09/05/12 13:12, Arnd Bergmann wrote:

quoted

On Wednesday 09 May 2012, Magnus Damm wrote:

quoted

 static unsigned int __init shmobile_smp_get_core_count(void)
 {

@@ -31,6 +32,9 @@ static unsigned int __init shmobile_smp_

   if (is_r8a7779())
           return r8a7779_get_core_count();

+  if (is_emev2())
+          return emev2_get_core_count();
+
   return 1;
 }

@@ -41,6 +45,9 @@ static void __init shmobile_smp_prepare_

   if (is_r8a7779())
           r8a7779_smp_prepare_cpus();
+
+  if (is_emev2())
+          emev2_smp_prepare_cpus();
 }

 int shmobile_platform_cpu_kill(unsigned int cpu)
...

This shows that we really want an abstraction for soc-specific SMP ops
even within one platform, and we'll need the same thing for multiplatform.

Marc Zyngier already proposed a solution for this last year, but I
think we couldn't agree on the details back then before he lost interest.
I think we should pick that up again and get it into 3.6 so the code above
can be simplified and we can do the multiplatform solution. We'll probably
discuss the details in Hong Kong in a couple of weeks, so there is no
point in changing it now, but I'd hope that you can migrate this to
whatever we come up with in the following merge window.

I'm happy to revive the series if there is an interest.

Ok, good. I think we were almost there the last time, but I don't
know if Russell still had any objections. Magnus, can you comment on
the "[PATCH v6 09/15] ARM: SoC: convert shmobile SMP to SoC descriptor"
patch from February to see if it fits your needs?

I took the liberty to give some comments here instead. I've gone
through the patch and it looks good in general. I have not tested it
but I'd be happy to fix whatever fallout that may happen.

The patch is pretty close to be a perfect fit from my point of view.
This is definitely a step in the right direction. I am however a bit
hesitant with how the CPU hotplug code ended up, but at the same time
I don't really have any better suggestions. I think I simply need to
convert it to be a full per-soc implementation.

Thanks!

/ magnus