[PATCH] ppc32: remove secondary_ti

Instead of copying current_set[cpu] to secondary_ti, directly index into
current_set by the cpu number like head_64.S.  In addition to removing
the variable, a slow to respond cpu will not clobber another cpus
stack.

---
grep found no other reference in arch/powerpc.   compile tested but I have
no 32bit smp hardware.

Index: next.git/arch/powerpc/kernel/head_32.S
===================================================================

--- next.git.orig/arch/powerpc/kernel/head_32.S	2008-11-16 01:36:45.000000000 -0600
+++ next.git/arch/powerpc/kernel/head_32.S	2008-11-16 01:42:02.000000000 -0600

@@ -906,10 +906,11 @@ __secondary_start:
 	bl	init_idle_6xx
 #endif /* CONFIG_6xx */
 
-	/* get current_thread_info and current */
-	lis	r1,secondary_ti@ha
+	/* get current_thread_info (current_set[cpu]) and current =  */
+	slwi	r1,r24,2
+	addis	r1,r1,current_set@ha
 	tophys(r1,r1)
-	lwz	r1,secondary_ti@l(r1)
+	lwz	r1,current_set@l(r1)
 	tophys(r2,r1)
 	lwz	r2,TI_TASK(r2)
 

Index: next.git/arch/powerpc/kernel/smp.c
===================================================================

--- next.git.orig/arch/powerpc/kernel/smp.c	2008-11-16 01:36:05.000000000 -0600
+++ next.git/arch/powerpc/kernel/smp.c	2008-11-16 01:36:22.000000000 -0600

@@ -58,7 +58,6 @@
 #endif
 
 int smp_hw_index[NR_CPUS];
-struct thread_info *secondary_ti;
 
 cpumask_t cpu_possible_map = CPU_MASK_NONE;
 cpumask_t cpu_online_map = CPU_MASK_NONE;

@@ -320,7 +319,6 @@ int __cpuinit __cpu_up(unsigned int cpu)
 {
 	int c;
 
-	secondary_ti = current_set[cpu];
 	if (!cpu_enable(cpu))
 		return 0;
 

On Sun, 2008-11-16 at 15:43 -0600, Milton Miller wrote:

Instead of copying current_set[cpu] to secondary_ti, directly index into
current_set by the cpu number like head_64.S.  In addition to removing
the variable, a slow to respond cpu will not clobber another cpus
stack.

Except that we don't know our CPU number at this stage do we ? IE. r24
is our HW number, is it guaranteed to be what we want to index
current_set ?

Ben.

On Nov 16, 2008, at 2:26 PM, Benjamin Herrenschmidt wrote:

On Sun, 2008-11-16 at 15:43 -0600, Milton Miller wrote:

quoted

Instead of copying current_set[cpu] to secondary_ti, directly index into
current_set by the cpu number like head_64.S.  In addition to removing
the variable, a slow to respond cpu will not clobber another cpus
stack.

Except that we don't know our CPU number at this stage do we ? IE. r24
is our HW number, is it guaranteed to be what we want to index
current_set ?

No, you are right, current_set is ordered by logical cpu id.

We could search the hwcpuid array similar to our search of the paca on ppc64.  But for now go ahead and reject this patch.

milton