[GIT,PULL] MSM fix

Commit Message

Arnd Bergmann July 21, 2011, 2:49 p.m. UTC

On Thursday 21 July 2011, Russell King - ARM Linux wrote:
> > It looks like it would apply to 2.6.38 and 39 as well.  I'm not sure
> > it qualifies under the "It must fix a real bug that bothers people".
> > It would only apply if someone had set NR_CPUS to more than 2.
> 
> Err, this makes no sense.  MSM doesn't have a get_core_count function
> and as far as I can see never has done in mainline.

I've committed the version below now, with a clarified title.
The patch looks reasonable to me, but there may be some issue
I'm missing of course. The other platforms that used to have
a get_core_count function no longer have that because you changed
them to use scu_get_core_count, but that doesn't seem to apply
to mach-msm.

I did not add a stable tag though.

	Arnd

8<---
Subject: [PATCH] ARM: msm: platsmp: determine number of CPU cores at boot time

Previously we just assumed there were CONFIG_NR_CPUS cpus present in
the system. Instead, figure out the number of cpus from the MIDR
register.

Signed-off-by: Jeff Ohlstein <[e-mail]>
Signed-off-by: David Brown <[e-mail]>
[arnd: clarified patch title]
Signed-off-by: Arnd Bergmann <[e-mail]>

Comments

Russell King - ARM Linux July 21, 2011, 3:41 p.m. UTC | #1

On Thu, Jul 21, 2011 at 04:49:39PM +0200, Arnd Bergmann wrote:
> On Thursday 21 July 2011, Russell King - ARM Linux wrote:
> > > It looks like it would apply to 2.6.38 and 39 as well.  I'm not sure
> > > it qualifies under the "It must fix a real bug that bothers people".
> > > It would only apply if someone had set NR_CPUS to more than 2.
> > 
> > Err, this makes no sense.  MSM doesn't have a get_core_count function
> > and as far as I can see never has done in mainline.
> 
> I've committed the version below now, with a clarified title.
> The patch looks reasonable to me, but there may be some issue
> I'm missing of course. The other platforms that used to have
> a get_core_count function no longer have that because you changed
> them to use scu_get_core_count, but that doesn't seem to apply
> to mach-msm.

That patch description helps a lot to clarify what's going on.

Patch

diff --git a/arch/arm/mach-msm/platsmp.c b/arch/arm/mach-msm/platsmp.c
index 2034098..5ba77d0 100644
--- a/arch/arm/mach-msm/platsmp.c
+++ b/arch/arm/mach-msm/platsmp.c
@@ -18,6 +18,7 @@ 
 
 #include <asm/hardware/gic.h>
 #include <asm/cacheflush.h>
+#include <asm/cputype.h>
 #include <asm/mach-types.h>
 
 #include <mach/msm_iomap.h>
@@ -40,6 +41,12 @@  volatile int pen_release = -1;
 
 static DEFINE_SPINLOCK(boot_lock);
 
+static inline int get_core_count(void)
+{
+	/* 1 + the PART[1:0] field of MIDR */
+	return ((read_cpuid_id() >> 4) & 3) + 1;
+}
+
 void __cpuinit platform_secondary_init(unsigned int cpu)
 {
 	/* Configure edge-triggered PPIs */
@@ -147,9 +154,9 @@  int __cpuinit boot_secondary(unsigned int cpu, struct task_struct *idle)
  */
 void __init smp_init_cpus(void)
 {
-	unsigned int i;
+	unsigned int i, ncores = get_core_count();
 
-	for (i = 0; i < NR_CPUS; i++)
+	for (i = 0; i < ncores; i++)
 		set_cpu_possible(i, true);
 
         set_smp_cross_call(gic_raise_softirq);