[PATCH] ppc64: dynamically allocate segment tables

PPC64 machines before Power4 need a segment table page allocated for each
CPU.  Currently these are allocated statically in a big array in head.S for
all CPUs.  The segment tables need to be in the first segment (so
do_stab_bolted doesn't take a recursive fault on the stab itself), but
other than that there are no constraints which require the stabs for the
secondary CPUs to be statically allocated.

This patch allocates segment tables dynamically during boot, using
lmb_alloc() to ensure they are within the first 256M segment.  This reduces
the kernel image size by 192k...

Tested on RS64 iSeries, POWER3 pSeries, and POWER5.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

4 files changed, 37 insertions, 22 deletions

diff --git a/arch/ppc64/kernel/head.S b/arch/ppc64/kernel/head.S
index 93ebcac0d5a2..3f447712e3ff 100644
--- a/arch/ppc64/kernel/head.S
+++ b/arch/ppc64/kernel/head.S
@@ -2131,13 +2131,6 @@ empty_zero_page:
 swapper_pg_dir:
 	.space	4096
 
-#ifdef CONFIG_SMP
-/* 1 page segment table per cpu (max 48, cpu0 allocated at STAB0_PHYS_ADDR) */
-	.globl	stab_array
-stab_array:
-	.space	4096 * 48
-#endif
-	
 /*
  * This space gets a copy of optional info passed to us by the bootstrap
  * Used to pass parameters into the kernel like root=/dev/sda1, etc.
diff --git a/arch/ppc64/kernel/setup.c b/arch/ppc64/kernel/setup.c
index e80f10c89824..687e85595208 100644
--- a/arch/ppc64/kernel/setup.c
+++ b/arch/ppc64/kernel/setup.c
@@ -1068,6 +1068,8 @@ void __init setup_arch(char **cmdline_p)
 	irqstack_early_init();
 	emergency_stack_init();
 
+	stabs_alloc();
+
 	/* set up the bootmem stuff with available memory */
 	do_init_bootmem();
 	sparse_init();
diff --git a/arch/ppc64/kernel/smp.c b/arch/ppc64/kernel/smp.c
index 2fcddfcb594d..793b562da653 100644
--- a/arch/ppc64/kernel/smp.c
+++ b/arch/ppc64/kernel/smp.c
@@ -65,8 +65,6 @@ struct smp_ops_t *smp_ops;
 
 static volatile unsigned int cpu_callin_map[NR_CPUS];
 
-extern unsigned char stab_array[];
-
 void smp_call_function_interrupt(void);
 
 int smt_enabled_at_boot = 1;
@@ -492,19 +490,6 @@ int __devinit __cpu_up(unsigned int cpu)
 
 	paca[cpu].default_decr = tb_ticks_per_jiffy;
 
-	if (!cpu_has_feature(CPU_FTR_SLB)) {
-		void *tmp;
-
-		/* maximum of 48 CPUs on machines with a segment table */
-		if (cpu >= 48)
-			BUG();
-
-		tmp = &stab_array[PAGE_SIZE * cpu];
-		memset(tmp, 0, PAGE_SIZE); 
-		paca[cpu].stab_addr = (unsigned long)tmp;
-		paca[cpu].stab_real = virt_to_abs(tmp);
-	}
-
 	/* Make sure callin-map entry is 0 (can be leftover a CPU
 	 * hotplug
 	 */
diff --git a/arch/ppc64/mm/stab.c b/arch/ppc64/mm/stab.c
index df4bbe14153c..1b83f002bf27 100644
--- a/arch/ppc64/mm/stab.c
+++ b/arch/ppc64/mm/stab.c
@@ -18,6 +18,8 @@
 #include <asm/mmu_context.h>
 #include <asm/paca.h>
 #include <asm/cputable.h>
+#include <asm/lmb.h>
+#include <asm/abs_addr.h>
 
 struct stab_entry {
 	unsigned long esid_data;
@@ -224,6 +226,39 @@ void switch_stab(struct task_struct *tsk, struct mm_struct *mm)
 extern void slb_initialize(void);
 
 /*
+ * Allocate segment tables for secondary CPUs.  These must all go in
+ * the first (bolted) segment, so that do_stab_bolted won't get a
+ * recursive segment miss on the segment table itself.
+ */
+void stabs_alloc(void)
+{
+	int cpu;
+
+	if (cpu_has_feature(CPU_FTR_SLB))
+		return;
+
+	for_each_cpu(cpu) {
+		unsigned long newstab;
+
+		if (cpu == 0)
+			continue; /* stab for CPU 0 is statically allocated */
+
+		newstab = lmb_alloc_base(PAGE_SIZE, PAGE_SIZE, 1<<SID_SHIFT);
+		if (! newstab)
+			panic("Unable to allocate segment table for CPU %d.\n",
+			      cpu);
+
+		newstab += KERNELBASE;
+
+		memset((void *)newstab, 0, PAGE_SIZE);
+
+		paca[cpu].stab_addr = newstab;
+		paca[cpu].stab_real = virt_to_abs(newstab);
+		printk(KERN_DEBUG "Segment table for CPU %d at 0x%lx virtual, 0x%lx absolute\n", cpu, paca[cpu].stab_addr, paca[cpu].stab_real);
+	}
+}
+
+/*
  * Build an entry for the base kernel segment and put it into
  * the segment table or SLB.  All other segment table or SLB
  * entries are faulted in.