[ARM] Fix SA110/SA1100 cache flushing

We had two implementations for flushing the cache, which meant StrongARM
caches weren't being correctly flushed.  Fix this by always using the
v4wb_flush_kern_cache_all method, rather than duplicating it.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>

arch/arm/mm/cache-v4wb.S

@@ -10,7 +10,7 @@
 #include <linux/config.h>
 #include <linux/linkage.h>
 #include <linux/init.h>
-#include <asm/hardware.h>
+#include <asm/memory.h>
 #include <asm/page.h>
 #include "proc-macros.S"
 
@@ -46,6 +46,11 @@
  */
 #define CACHE_DLIMIT	(CACHE_DSIZE * 4)
 
+	.data
+flush_base:
+	.long	FLUSH_BASE
+	.text
+
 /*
  *	flush_user_cache_all()
  *
@@ -63,11 +68,21 @@ ENTRY(v4wb_flush_kern_cache_all)
 	mov	ip, #0
 	mcr	p15, 0, ip, c7, c5, 0		@ invalidate I cache
 __flush_whole_cache:
-	mov	r0, #FLUSH_BASE
-	add	r1, r0, #CACHE_DSIZE
-1:	ldr	r2, [r0], #32
-	cmp	r0, r1
+	ldr	r3, =flush_base
+	ldr	r1, [r3, #0]
+	eor	r1, r1, #CACHE_DSIZE
+	str	r1, [r3, #0]
+	add	r2, r1, #CACHE_DSIZE
+1:	ldr	r3, [r1], #32
+	cmp	r1, r2
 	blo	1b
+#ifdef FLUSH_BASE_MINICACHE
+	add	r2, r2, #FLUSH_BASE_MINICACHE - FLUSH_BASE
+	sub	r1, r2, #512			@ only 512 bytes
+1:	ldr	r3, [r1], #32
+	cmp	r1, r2
+	blo	1b
+#endif
 	mcr	p15, 0, ip, c7, c10, 4		@ drain write buffer
 	mov	pc, lr
 
@@ -82,6 +97,7 @@ __flush_whole_cache:
  *	- flags	- vma_area_struct flags describing address space
  */
 ENTRY(v4wb_flush_user_cache_range)
+	mov	ip, #0
 	sub	r3, r1, r0			@ calculate total size
 	tst	r2, #VM_EXEC			@ executable region?
 	mcrne	p15, 0, ip, c7, c5, 0		@ invalidate I cache

arch/arm/mm/proc-sa110.S

@@ -26,22 +26,7 @@
  * the cache line size of the I and D cache
  */
 #define DCACHELINESIZE	32
-#define FLUSH_OFFSET	32768
 
-	.macro flush_110_dcache	rd, ra, re
-	ldr	\rd, =flush_base
-	ldr	\ra, [\rd]
-	eor	\ra, \ra, #FLUSH_OFFSET
-	str	\ra, [\rd]
-	add	\re, \ra, #16384		@ only necessary for 16k
-1001:	ldr	\rd, [\ra], #DCACHELINESIZE
-	teq	\re, \ra
-	bne	1001b
-	.endm
-
-	.data
-flush_base:
-	.long	FLUSH_BASE
 	.text
 
 /*
@@ -145,13 +130,11 @@ ENTRY(cpu_sa110_dcache_clean_area)
  */
 	.align	5
 ENTRY(cpu_sa110_switch_mm)
-	flush_110_dcache	r3, ip, r1
-	mov	r1, #0
-	mcr	p15, 0, r1, c7, c5, 0		@ invalidate I cache
-	mcr	p15, 0, r1, c7, c10, 4		@ drain WB
+	str	lr, [sp, #-4]!
+	bl	v4wb_flush_kern_cache_all	@ clears IP
 	mcr	p15, 0, r0, c2, c0, 0		@ load page table pointer
-	mcr	p15, 0, r1, c8, c7, 0		@ invalidate I & D TLBs
-	mov	pc, lr
+	mcr	p15, 0, ip, c8, c7, 0		@ invalidate I & D TLBs
+	ldr	pc, [sp], #4
 
 /*
  * cpu_sa110_set_pte(ptep, pte)

arch/arm/mm/proc-sa1100.S

@@ -30,30 +30,6 @@
  * the cache line size of the I and D cache
  */
 #define DCACHELINESIZE	32
-#define FLUSH_OFFSET	32768
-
-	.macro flush_1100_dcache rd, ra, re
-	ldr	\rd, =flush_base
-	ldr	\ra, [\rd]
-	eor	\ra, \ra, #FLUSH_OFFSET
-	str	\ra, [\rd]
-	add	\re, \ra, #8192			@ only necessary for 8k
-1001:	ldr	\rd, [\ra], #DCACHELINESIZE
-	teq	\re, \ra
-	bne	1001b
-#ifdef FLUSH_BASE_MINICACHE
-	add	\ra, \ra, #FLUSH_BASE_MINICACHE - FLUSH_BASE
-	add	\re, \ra, #512			@ only 512 bytes
-1002:	ldr	\rd, [\ra], #DCACHELINESIZE
-	teq	\re, \ra
-	bne	1002b
-#endif
-	.endm
-
-	.data
-flush_base:
-	.long	FLUSH_BASE
-	.text
 
 	__INIT
 
@@ -79,9 +55,8 @@ ENTRY(cpu_sa1100_proc_fin)
 	stmfd	sp!, {lr}
 	mov	ip, #PSR_F_BIT | PSR_I_BIT | SVC_MODE
 	msr	cpsr_c, ip
-	flush_1100_dcache r0, r1, r2		@ clean caches
-	mov	r0, #0
-	mcr	p15, 0, r0, c15, c2, 2		@ Disable clock switching
+	bl	v4wb_flush_kern_cache_all
+	mcr	p15, 0, ip, c15, c2, 2		@ Disable clock switching
 	mrc	p15, 0, r0, c1, c0, 0		@ ctrl register
 	bic	r0, r0, #0x1000			@ ...i............
 	bic	r0, r0, #0x000e			@ ............wca.
@@ -167,14 +142,12 @@ ENTRY(cpu_sa1100_dcache_clean_area)
  */
 	.align	5
 ENTRY(cpu_sa1100_switch_mm)
-	flush_1100_dcache r3, ip, r1
-	mov	ip, #0
-	mcr	p15, 0, ip, c7, c5, 0		@ invalidate I cache
+	str	lr, [sp, #-4]!
+	bl	v4wb_flush_kern_cache_all	@ clears IP
 	mcr	p15, 0, ip, c9, c0, 0		@ invalidate RB
-	mcr	p15, 0, ip, c7, c10, 4		@ drain WB
 	mcr	p15, 0, r0, c2, c0, 0		@ load page table pointer
 	mcr	p15, 0, ip, c8, c7, 0		@ invalidate I & D TLBs
-	mov	pc, lr
+	ldr	pc, [sp], #4
 
 /*
  * cpu_sa1100_set_pte(ptep, pte)