Merge branch 'linus' into locking/core, to pick up fixes

Signed-off-by: Ingo Molnar <mingo@kernel.org>

32 files changed, 1400 insertions, 723 deletions

diff --git a/arch/sparc/include/asm/cpudata_64.h b/arch/sparc/include/asm/cpudata_64.h
index a6cfdabb6054..5b0ed48e5b0c 100644
--- a/arch/sparc/include/asm/cpudata_64.h
+++ b/arch/sparc/include/asm/cpudata_64.h
@@ -24,9 +24,10 @@ typedef struct {
 	unsigned int	icache_line_size;
 	unsigned int	ecache_size;
 	unsigned int	ecache_line_size;
-	unsigned short	sock_id;
+	unsigned short	sock_id;	/* physical package */
 	unsigned short	core_id;
-	int		proc_id;
+	unsigned short  max_cache_id;	/* groupings of highest shared cache */
+	unsigned short	proc_id;	/* strand (aka HW thread) id */
 } cpuinfo_sparc;
 
 DECLARE_PER_CPU(cpuinfo_sparc, __cpu_data);
diff --git a/arch/sparc/include/asm/spinlock_32.h b/arch/sparc/include/asm/spinlock_32.h
index d9c5876c6121..8011e79f59c9 100644
--- a/arch/sparc/include/asm/spinlock_32.h
+++ b/arch/sparc/include/asm/spinlock_32.h
@@ -134,7 +134,7 @@ static inline void arch_write_lock(arch_rwlock_t *rw)
 	*(volatile __u32 *)&lp->lock = ~0U;
 }
 
-static void inline arch_write_unlock(arch_rwlock_t *lock)
+static inline void arch_write_unlock(arch_rwlock_t *lock)
 {
 	__asm__ __volatile__(
 "	st		%%g0, [%0]"
diff --git a/arch/sparc/include/asm/spinlock_64.h b/arch/sparc/include/asm/spinlock_64.h
index 87990b7c6b0d..07c9f2e9bf57 100644
--- a/arch/sparc/include/asm/spinlock_64.h
+++ b/arch/sparc/include/asm/spinlock_64.h
@@ -96,7 +96,7 @@ static inline void arch_spin_lock_flags(arch_spinlock_t *lock, unsigned long fla
 
 /* Multi-reader locks, these are much saner than the 32-bit Sparc ones... */
 
-static void inline arch_read_lock(arch_rwlock_t *lock)
+static inline void arch_read_lock(arch_rwlock_t *lock)
 {
 	unsigned long tmp1, tmp2;
 
@@ -119,7 +119,7 @@ static void inline arch_read_lock(arch_rwlock_t *lock)
 	: "memory");
 }
 
-static int inline arch_read_trylock(arch_rwlock_t *lock)
+static inline int arch_read_trylock(arch_rwlock_t *lock)
 {
 	int tmp1, tmp2;
 
@@ -140,7 +140,7 @@ static int inline arch_read_trylock(arch_rwlock_t *lock)
 	return tmp1;
 }
 
-static void inline arch_read_unlock(arch_rwlock_t *lock)
+static inline void arch_read_unlock(arch_rwlock_t *lock)
 {
 	unsigned long tmp1, tmp2;
 
@@ -156,7 +156,7 @@ static void inline arch_read_unlock(arch_rwlock_t *lock)
 	: "memory");
 }
 
-static void inline arch_write_lock(arch_rwlock_t *lock)
+static inline void arch_write_lock(arch_rwlock_t *lock)
 {
 	unsigned long mask, tmp1, tmp2;
 
@@ -181,7 +181,7 @@ static void inline arch_write_lock(arch_rwlock_t *lock)
 	: "memory");
 }
 
-static void inline arch_write_unlock(arch_rwlock_t *lock)
+static inline void arch_write_unlock(arch_rwlock_t *lock)
 {
 	__asm__ __volatile__(
 "	stw		%%g0, [%0]"
@@ -190,7 +190,7 @@ static void inline arch_write_unlock(arch_rwlock_t *lock)
 	: "memory");
 }
 
-static int inline arch_write_trylock(arch_rwlock_t *lock)
+static inline int arch_write_trylock(arch_rwlock_t *lock)
 {
 	unsigned long mask, tmp1, tmp2, result;
 
diff --git a/arch/sparc/include/asm/topology_64.h b/arch/sparc/include/asm/topology_64.h
index bec481aaca16..7b4898a36eee 100644
--- a/arch/sparc/include/asm/topology_64.h
+++ b/arch/sparc/include/asm/topology_64.h
@@ -44,14 +44,20 @@ int __node_distance(int, int);
 #define topology_physical_package_id(cpu)	(cpu_data(cpu).proc_id)
 #define topology_core_id(cpu)			(cpu_data(cpu).core_id)
 #define topology_core_cpumask(cpu)		(&cpu_core_sib_map[cpu])
+#define topology_core_cache_cpumask(cpu)	(&cpu_core_sib_cache_map[cpu])
 #define topology_sibling_cpumask(cpu)		(&per_cpu(cpu_sibling_map, cpu))
 #endif /* CONFIG_SMP */
 
 extern cpumask_t cpu_core_map[NR_CPUS];
 extern cpumask_t cpu_core_sib_map[NR_CPUS];
+extern cpumask_t cpu_core_sib_cache_map[NR_CPUS];
+
+/**
+ * Return cores that shares the last level cache.
+ */
 static inline const struct cpumask *cpu_coregroup_mask(int cpu)
 {
-        return &cpu_core_map[cpu];
+	return &cpu_core_sib_cache_map[cpu];
 }
 
 #endif /* _ASM_SPARC64_TOPOLOGY_H */
diff --git a/arch/sparc/include/asm/uaccess_64.h b/arch/sparc/include/asm/uaccess_64.h
index b68acc563235..5373136c412b 100644
--- a/arch/sparc/include/asm/uaccess_64.h
+++ b/arch/sparc/include/asm/uaccess_64.h
@@ -82,7 +82,6 @@ static inline int access_ok(int type, const void __user * addr, unsigned long si
 	return 1;
 }
 
-void __ret_efault(void);
 void __retl_efault(void);
 
 /* Uh, these should become the main single-value transfer routines..
@@ -189,55 +188,34 @@ int __get_user_bad(void);
 unsigned long __must_check ___copy_from_user(void *to,
 					     const void __user *from,
 					     unsigned long size);
-unsigned long copy_from_user_fixup(void *to, const void __user *from,
-				   unsigned long size);
 static inline unsigned long __must_check
 copy_from_user(void *to, const void __user *from, unsigned long size)
 {
-	unsigned long ret;
-
 	check_object_size(to, size, false);
 
-	ret = ___copy_from_user(to, from, size);
-	if (unlikely(ret))
-		ret = copy_from_user_fixup(to, from, size);
-
-	return ret;
+	return ___copy_from_user(to, from, size);
 }
 #define __copy_from_user copy_from_user
 
 unsigned long __must_check ___copy_to_user(void __user *to,
 					   const void *from,
 					   unsigned long size);
-unsigned long copy_to_user_fixup(void __user *to, const void *from,
-				 unsigned long size);
 static inline unsigned long __must_check
 copy_to_user(void __user *to, const void *from, unsigned long size)
 {
-	unsigned long ret;
-
 	check_object_size(from, size, true);
 
-	ret = ___copy_to_user(to, from, size);
-	if (unlikely(ret))
-		ret = copy_to_user_fixup(to, from, size);
-	return ret;
+	return ___copy_to_user(to, from, size);
 }
 #define __copy_to_user copy_to_user
 
 unsigned long __must_check ___copy_in_user(void __user *to,
 					   const void __user *from,
 					   unsigned long size);
-unsigned long copy_in_user_fixup(void __user *to, void __user *from,
-				 unsigned long size);
 static inline unsigned long __must_check
 copy_in_user(void __user *to, void __user *from, unsigned long size)
 {
-	unsigned long ret = ___copy_in_user(to, from, size);
-
-	if (unlikely(ret))
-		ret = copy_in_user_fixup(to, from, size);
-	return ret;
+	return ___copy_in_user(to, from, size);
 }
 #define __copy_in_user copy_in_user
 
diff --git a/arch/sparc/kernel/head_64.S b/arch/sparc/kernel/head_64.S
index beba6c11554c..6aa3da152c20 100644
--- a/arch/sparc/kernel/head_64.S
+++ b/arch/sparc/kernel/head_64.S
@@ -926,48 +926,11 @@ tlb_type:	.word	0	/* Must NOT end up in BSS */
 EXPORT_SYMBOL(tlb_type)
 	.section	".fixup",#alloc,#execinstr
 
-	.globl	__ret_efault, __retl_efault, __ret_one, __retl_one
-ENTRY(__ret_efault)
-	ret
-	 restore %g0, -EFAULT, %o0
-ENDPROC(__ret_efault)
-EXPORT_SYMBOL(__ret_efault)
-
 ENTRY(__retl_efault)
 	retl
 	 mov	-EFAULT, %o0
 ENDPROC(__retl_efault)
 
-ENTRY(__retl_one)
-	retl
-	 mov	1, %o0
-ENDPROC(__retl_one)
-
-ENTRY(__retl_one_fp)
-	VISExitHalf
-	retl
-	 mov	1, %o0
-ENDPROC(__retl_one_fp)
-
-ENTRY(__ret_one_asi)
-	wr	%g0, ASI_AIUS, %asi
-	ret
-	 restore %g0, 1, %o0
-ENDPROC(__ret_one_asi)
-
-ENTRY(__retl_one_asi)
-	wr	%g0, ASI_AIUS, %asi
-	retl
-	 mov	1, %o0
-ENDPROC(__retl_one_asi)
-
-ENTRY(__retl_one_asi_fp)
-	wr	%g0, ASI_AIUS, %asi
-	VISExitHalf
-	retl
-	 mov	1, %o0
-ENDPROC(__retl_one_asi_fp)
-
 ENTRY(__retl_o1)
 	retl
 	 mov	%o1, %o0
diff --git a/arch/sparc/kernel/jump_label.c b/arch/sparc/kernel/jump_label.c
index 59bbeff55024..07933b9e9ce0 100644
--- a/arch/sparc/kernel/jump_label.c
+++ b/arch/sparc/kernel/jump_label.c
@@ -13,19 +13,30 @@
 void arch_jump_label_transform(struct jump_entry *entry,
 			       enum jump_label_type type)
 {
-	u32 val;
 	u32 *insn = (u32 *) (unsigned long) entry->code;
+	u32 val;
 
 	if (type == JUMP_LABEL_JMP) {
 		s32 off = (s32)entry->target - (s32)entry->code;
+		bool use_v9_branch = false;
+
+		BUG_ON(off & 3);
 
 #ifdef CONFIG_SPARC64
-		/* ba,pt %xcc, . + (off << 2) */
-		val = 0x10680000 | ((u32) off >> 2);
-#else
-		/* ba . + (off << 2) */
-		val = 0x10800000 | ((u32) off >> 2);
+		if (off <= 0xfffff && off >= -0x100000)
+			use_v9_branch = true;
 #endif
+		if (use_v9_branch) {
+			/* WDISP19 - target is . + immed << 2 */
+			/* ba,pt %xcc, . + off */
+			val = 0x10680000 | (((u32) off >> 2) & 0x7ffff);
+		} else {
+			/* WDISP22 - target is . + immed << 2 */
+			BUG_ON(off > 0x7fffff);
+			BUG_ON(off < -0x800000);
+			/* ba . + off */
+			val = 0x10800000 | (((u32) off >> 2) & 0x3fffff);
+		}
 	} else {
 		val = 0x01000000;
 	}
diff --git a/arch/sparc/kernel/mdesc.c b/arch/sparc/kernel/mdesc.c
index 11228861d9b4..8a6982dfd733 100644
--- a/arch/sparc/kernel/mdesc.c
+++ b/arch/sparc/kernel/mdesc.c
@@ -645,13 +645,20 @@ static void __mark_core_id(struct mdesc_handle *hp, u64 node,
 		cpu_data(*id).core_id = core_id;
 }
 
-static void __mark_sock_id(struct mdesc_handle *hp, u64 node,
-			   int sock_id)
+static void __mark_max_cache_id(struct mdesc_handle *hp, u64 node,
+				int max_cache_id)
 {
 	const u64 *id = mdesc_get_property(hp, node, "id", NULL);
 
-	if (*id < num_possible_cpus())
-		cpu_data(*id).sock_id = sock_id;
+	if (*id < num_possible_cpus()) {
+		cpu_data(*id).max_cache_id = max_cache_id;
+
+		/**
+		 * On systems without explicit socket descriptions socket
+		 * is max_cache_id
+		 */
+		cpu_data(*id).sock_id = max_cache_id;
+	}
 }
 
 static void mark_core_ids(struct mdesc_handle *hp, u64 mp,
@@ -660,10 +667,11 @@ static void mark_core_ids(struct mdesc_handle *hp, u64 mp,
 	find_back_node_value(hp, mp, "cpu", __mark_core_id, core_id, 10);
 }
 
-static void mark_sock_ids(struct mdesc_handle *hp, u64 mp,
-			  int sock_id)
+static void mark_max_cache_ids(struct mdesc_handle *hp, u64 mp,
+			       int max_cache_id)
 {
-	find_back_node_value(hp, mp, "cpu", __mark_sock_id, sock_id, 10);
+	find_back_node_value(hp, mp, "cpu", __mark_max_cache_id,
+			     max_cache_id, 10);
 }
 
 static void set_core_ids(struct mdesc_handle *hp)
@@ -694,14 +702,15 @@ static void set_core_ids(struct mdesc_handle *hp)
 	}
 }
 
-static int set_sock_ids_by_cache(struct mdesc_handle *hp, int level)
+static int set_max_cache_ids_by_cache(struct mdesc_handle *hp, int level)
 {
 	u64 mp;
 	int idx = 1;
 	int fnd = 0;
 
-	/* Identify unique sockets by looking for cpus backpointed to by
-	 * shared level n caches.
+	/**
+	 * Identify unique highest level of shared cache by looking for cpus
+	 * backpointed to by shared level N caches.
 	 */
 	mdesc_for_each_node_by_name(hp, mp, "cache") {
 		const u64 *cur_lvl;
@@ -709,8 +718,7 @@ static int set_sock_ids_by_cache(struct mdesc_handle *hp, int level)
 		cur_lvl = mdesc_get_property(hp, mp, "level", NULL);
 		if (*cur_lvl != level)
 			continue;
-
-		mark_sock_ids(hp, mp, idx);
+		mark_max_cache_ids(hp, mp, idx);
 		idx++;
 		fnd = 1;
 	}
@@ -745,15 +753,17 @@ static void set_sock_ids(struct mdesc_handle *hp)
 {
 	u64 mp;
 
-	/* If machine description exposes sockets data use it.
-	 * Otherwise fallback to use shared L3 or L2 caches.
+	/**
+	 * Find the highest level of shared cache which pre-T7 is also
+	 * the socket.
 	 */
+	if (!set_max_cache_ids_by_cache(hp, 3))
+		set_max_cache_ids_by_cache(hp, 2);
+
+	/* If machine description exposes sockets data use it.*/
 	mp = mdesc_node_by_name(hp, MDESC_NODE_NULL, "sockets");
 	if (mp != MDESC_NODE_NULL)
-		return set_sock_ids_by_socket(hp, mp);
-
-	if (!set_sock_ids_by_cache(hp, 3))
-		set_sock_ids_by_cache(hp, 2);
+		set_sock_ids_by_socket(hp, mp);
 }
 
 static void mark_proc_ids(struct mdesc_handle *hp, u64 mp, int proc_id)
diff --git a/arch/sparc/kernel/smp_64.c b/arch/sparc/kernel/smp_64.c
index d3035ba6cd31..8182f7caf5b1 100644
--- a/arch/sparc/kernel/smp_64.c
+++ b/arch/sparc/kernel/smp_64.c
@@ -63,9 +63,13 @@ cpumask_t cpu_core_map[NR_CPUS] __read_mostly =
 cpumask_t cpu_core_sib_map[NR_CPUS] __read_mostly = {
 	[0 ... NR_CPUS-1] = CPU_MASK_NONE };
 
+cpumask_t cpu_core_sib_cache_map[NR_CPUS] __read_mostly = {
+	[0 ... NR_CPUS - 1] = CPU_MASK_NONE };
+
 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
 EXPORT_SYMBOL(cpu_core_map);
 EXPORT_SYMBOL(cpu_core_sib_map);
+EXPORT_SYMBOL(cpu_core_sib_cache_map);
 
 static cpumask_t smp_commenced_mask;
 
@@ -1265,6 +1269,10 @@ void smp_fill_in_sib_core_maps(void)
 		unsigned int j;
 
 		for_each_present_cpu(j)  {
+			if (cpu_data(i).max_cache_id ==
+			    cpu_data(j).max_cache_id)
+				cpumask_set_cpu(j, &cpu_core_sib_cache_map[i]);
+
 			if (cpu_data(i).sock_id == cpu_data(j).sock_id)
 				cpumask_set_cpu(j, &cpu_core_sib_map[i]);
 		}
diff --git a/arch/sparc/lib/GENcopy_from_user.S b/arch/sparc/lib/GENcopy_from_user.S
index b7d0bd6b1406..69a439fa2fc1 100644
--- a/arch/sparc/lib/GENcopy_from_user.S
+++ b/arch/sparc/lib/GENcopy_from_user.S
@@ -3,11 +3,11 @@
  * Copyright (C) 2007 David S. Miller (davem@davemloft.net)
  */
 
-#define EX_LD(x)		\
+#define EX_LD(x,y)		\
 98:	x;			\
 	.section __ex_table,"a";\
 	.align 4;		\
-	.word 98b, __retl_one;	\
+	.word 98b, y;		\
 	.text;			\
 	.align 4;
 
diff --git a/arch/sparc/lib/GENcopy_to_user.S b/arch/sparc/lib/GENcopy_to_user.S
index 780550e1afc7..9947427ce354 100644
--- a/arch/sparc/lib/GENcopy_to_user.S
+++ b/arch/sparc/lib/GENcopy_to_user.S
@@ -3,11 +3,11 @@
  * Copyright (C) 2007 David S. Miller (davem@davemloft.net)
  */
 
-#define EX_ST(x)		\
+#define EX_ST(x,y)		\
 98:	x;			\
 	.section __ex_table,"a";\
 	.align 4;		\
-	.word 98b, __retl_one;	\
+	.word 98b, y;		\
 	.text;			\
 	.align 4;
 
diff --git a/arch/sparc/lib/GENmemcpy.S b/arch/sparc/lib/GENmemcpy.S
index 89358ee94851..059ea24ad73d 100644
--- a/arch/sparc/lib/GENmemcpy.S
+++ b/arch/sparc/lib/GENmemcpy.S
@@ -4,21 +4,18 @@
  */
 
 #ifdef __KERNEL__
+#include <linux/linkage.h>
 #define GLOBAL_SPARE	%g7
 #else
 #define GLOBAL_SPARE	%g5
 #endif
 
 #ifndef EX_LD
-#define EX_LD(x)	x
+#define EX_LD(x,y)	x
 #endif
 
 #ifndef EX_ST
-#define EX_ST(x)	x
-#endif
-
-#ifndef EX_RETVAL
-#define EX_RETVAL(x)	x
+#define EX_ST(x,y)	x
 #endif
 
 #ifndef LOAD
@@ -45,6 +42,29 @@
 	.register	%g3,#scratch
 
 	.text
+
+#ifndef EX_RETVAL
+#define EX_RETVAL(x)	x
+ENTRY(GEN_retl_o4_1)
+	add	%o4, %o2, %o4
+	retl
+	 add	%o4, 1, %o0
+ENDPROC(GEN_retl_o4_1)
+ENTRY(GEN_retl_g1_8)
+	add	%g1, %o2, %g1
+	retl
+	 add	%g1, 8, %o0
+ENDPROC(GEN_retl_g1_8)
+ENTRY(GEN_retl_o2_4)
+	retl
+	 add	%o2, 4, %o0
+ENDPROC(GEN_retl_o2_4)
+ENTRY(GEN_retl_o2_1)
+	retl
+	 add	%o2, 1, %o0
+ENDPROC(GEN_retl_o2_1)
+#endif
+
 	.align		64
 
 	.globl	FUNC_NAME
@@ -73,8 +93,8 @@ FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
 	sub		%g0, %o4, %o4
 	sub		%o2, %o4, %o2
 1:	subcc		%o4, 1, %o4
-	EX_LD(LOAD(ldub, %o1, %g1))
-	EX_ST(STORE(stb, %g1, %o0))
+	EX_LD(LOAD(ldub, %o1, %g1),GEN_retl_o4_1)
+	EX_ST(STORE(stb, %g1, %o0),GEN_retl_o4_1)
 	add		%o1, 1, %o1
 	bne,pt		%XCC, 1b
 	add		%o0, 1, %o0
@@ -82,8 +102,8 @@ FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
 	andn		%o2, 0x7, %g1
 	sub		%o2, %g1, %o2
 1:	subcc		%g1, 0x8, %g1
-	EX_LD(LOAD(ldx, %o1, %g2))
-	EX_ST(STORE(stx, %g2, %o0))
+	EX_LD(LOAD(ldx, %o1, %g2),GEN_retl_g1_8)
+	EX_ST(STORE(stx, %g2, %o0),GEN_retl_g1_8)
 	add		%o1, 0x8, %o1
 	bne,pt		%XCC, 1b
 	 add		%o0, 0x8, %o0
@@ -100,8 +120,8 @@ FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
 
 1:
 	subcc		%o2, 4, %o2
-	EX_LD(LOAD(lduw, %o1, %g1))
-	EX_ST(STORE(stw, %g1, %o1 + %o3))
+	EX_LD(LOAD(lduw, %o1, %g1),GEN_retl_o2_4)
+	EX_ST(STORE(stw, %g1, %o1 + %o3),GEN_retl_o2_4)
 	bgu,pt		%XCC, 1b
 	 add		%o1, 4, %o1
 
@@ -111,8 +131,8 @@ FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
 	.align		32
 90:
 	subcc		%o2, 1, %o2
-	EX_LD(LOAD(ldub, %o1, %g1))
-	EX_ST(STORE(stb, %g1, %o1 + %o3))
+	EX_LD(LOAD(ldub, %o1, %g1),GEN_retl_o2_1)
+	EX_ST(STORE(stb, %g1, %o1 + %o3),GEN_retl_o2_1)
 	bgu,pt		%XCC, 90b
 	 add		%o1, 1, %o1
 	retl
diff --git a/arch/sparc/lib/Makefile b/arch/sparc/lib/Makefile
index 885f00e81d1a..69912d2f8b54 100644
--- a/arch/sparc/lib/Makefile
+++ b/arch/sparc/lib/Makefile
@@ -38,7 +38,7 @@ lib-$(CONFIG_SPARC64) +=  NG4patch.o NG4copy_page.o NG4clear_page.o NG4memset.o
 lib-$(CONFIG_SPARC64) += GENmemcpy.o GENcopy_from_user.o GENcopy_to_user.o
 lib-$(CONFIG_SPARC64) += GENpatch.o GENpage.o GENbzero.o
 
-lib-$(CONFIG_SPARC64) += copy_in_user.o user_fixup.o memmove.o
+lib-$(CONFIG_SPARC64) += copy_in_user.o memmove.o
 lib-$(CONFIG_SPARC64) += mcount.o ipcsum.o xor.o hweight.o ffs.o
 
 obj-$(CONFIG_SPARC64) += iomap.o
diff --git a/arch/sparc/lib/NG2copy_from_user.S b/arch/sparc/lib/NG2copy_from_user.S
index d5242b8c4f94..b79a6998d87c 100644
--- a/arch/sparc/lib/NG2copy_from_user.S
+++ b/arch/sparc/lib/NG2copy_from_user.S
@@ -3,19 +3,19 @@
  * Copyright (C) 2007 David S. Miller (davem@davemloft.net)
  */
 
-#define EX_LD(x)		\
+#define EX_LD(x,y)		\
 98:	x;			\
 	.section __ex_table,"a";\
 	.align 4;		\
-	.word 98b, __retl_one_asi;\
+	.word 98b, y;		\
 	.text;			\
 	.align 4;
 
-#define EX_LD_FP(x)		\
+#define EX_LD_FP(x,y)		\
 98:	x;			\
 	.section __ex_table,"a";\
 	.align 4;		\
-	.word 98b, __retl_one_asi_fp;\
+	.word 98b, y##_fp;	\
 	.text;			\
 	.align 4;
 
diff --git a/arch/sparc/lib/NG2copy_to_user.S b/arch/sparc/lib/NG2copy_to_user.S
index 4e962d993b10..dcec55f254ab 100644
--- a/arch/sparc/lib/NG2copy_to_user.S
+++ b/arch/sparc/lib/NG2copy_to_user.S
@@ -3,19 +3,19 @@
  * Copyright (C) 2007 David S. Miller (davem@davemloft.net)
  */
 
-#define EX_ST(x)		\
+#define EX_ST(x,y)		\
 98:	x;			\
 	.section __ex_table,"a";\
 	.align 4;		\
-	.word 98b, __retl_one_asi;\
+	.word 98b, y;		\
 	.text;			\
 	.align 4;
 
-#define EX_ST_FP(x)		\
+#define EX_ST_FP(x,y)		\
 98:	x;			\
 	.section __ex_table,"a";\
 	.align 4;		\
-	.word 98b, __retl_one_asi_fp;\
+	.word 98b, y##_fp;	\
 	.text;			\
 	.align 4;
 
diff --git a/arch/sparc/lib/NG2memcpy.S b/arch/sparc/lib/NG2memcpy.S
index d5f585df2f3f..c629dbd121b6 100644
--- a/arch/sparc/lib/NG2memcpy.S
+++ b/arch/sparc/lib/NG2memcpy.S
@@ -4,6 +4,7 @@
  */
 
 #ifdef __KERNEL__
+#include <linux/linkage.h>
 #include <asm/visasm.h>
 #include <asm/asi.h>
 #define GLOBAL_SPARE	%g7
@@ -32,21 +33,17 @@
 #endif
 
 #ifndef EX_LD
-#define EX_LD(x)	x
+#define EX_LD(x,y)	x
 #endif
 #ifndef EX_LD_FP
-#define EX_LD_FP(x)	x
+#define EX_LD_FP(x,y)	x
 #endif
 
 #ifndef EX_ST
-#define EX_ST(x)	x
+#define EX_ST(x,y)	x
 #endif
 #ifndef EX_ST_FP
-#define EX_ST_FP(x)	x
-#endif
-
-#ifndef EX_RETVAL
-#define EX_RETVAL(x)	x
+#define EX_ST_FP(x,y)	x
 #endif
 
 #ifndef LOAD
@@ -140,45 +137,110 @@
 	fsrc2		%x6, %f12; \
 	fsrc2		%x7, %f14;
 #define FREG_LOAD_1(base, x0) \
-	EX_LD_FP(LOAD(ldd, base + 0x00, %x0))
+	EX_LD_FP(LOAD(ldd, base + 0x00, %x0), NG2_retl_o2_plus_g1)
 #define FREG_LOAD_2(base, x0, x1) \
-	EX_LD_FP(LOAD(ldd, base + 0x00, %x0)); \
-	EX_LD_FP(LOAD(ldd, base + 0x08, %x1));
+	EX_LD_FP(LOAD(ldd, base + 0x00, %x0), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x08, %x1), NG2_retl_o2_plus_g1);
 #define FREG_LOAD_3(base, x0, x1, x2) \
-	EX_LD_FP(LOAD(ldd, base + 0x00, %x0)); \
-	EX_LD_FP(LOAD(ldd, base + 0x08, %x1)); \
-	EX_LD_FP(LOAD(ldd, base + 0x10, %x2));
+	EX_LD_FP(LOAD(ldd, base + 0x00, %x0), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x08, %x1), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x10, %x2), NG2_retl_o2_plus_g1);
 #define FREG_LOAD_4(base, x0, x1, x2, x3) \
-	EX_LD_FP(LOAD(ldd, base + 0x00, %x0)); \
-	EX_LD_FP(LOAD(ldd, base + 0x08, %x1)); \
-	EX_LD_FP(LOAD(ldd, base + 0x10, %x2)); \
-	EX_LD_FP(LOAD(ldd, base + 0x18, %x3));
+	EX_LD_FP(LOAD(ldd, base + 0x00, %x0), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x08, %x1), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x10, %x2), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x18, %x3), NG2_retl_o2_plus_g1);
 #define FREG_LOAD_5(base, x0, x1, x2, x3, x4) \
-	EX_LD_FP(LOAD(ldd, base + 0x00, %x0)); \
-	EX_LD_FP(LOAD(ldd, base + 0x08, %x1)); \
-	EX_LD_FP(LOAD(ldd, base + 0x10, %x2)); \
-	EX_LD_FP(LOAD(ldd, base + 0x18, %x3)); \
-	EX_LD_FP(LOAD(ldd, base + 0x20, %x4));
+	EX_LD_FP(LOAD(ldd, base + 0x00, %x0), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x08, %x1), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x10, %x2), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x18, %x3), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x20, %x4), NG2_retl_o2_plus_g1);
 #define FREG_LOAD_6(base, x0, x1, x2, x3, x4, x5) \
-	EX_LD_FP(LOAD(ldd, base + 0x00, %x0)); \
-	EX_LD_FP(LOAD(ldd, base + 0x08, %x1)); \
-	EX_LD_FP(LOAD(ldd, base + 0x10, %x2)); \
-	EX_LD_FP(LOAD(ldd, base + 0x18, %x3)); \
-	EX_LD_FP(LOAD(ldd, base + 0x20, %x4)); \
-	EX_LD_FP(LOAD(ldd, base + 0x28, %x5));
+	EX_LD_FP(LOAD(ldd, base + 0x00, %x0), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x08, %x1), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x10, %x2), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x18, %x3), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x20, %x4), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x28, %x5), NG2_retl_o2_plus_g1);
 #define FREG_LOAD_7(base, x0, x1, x2, x3, x4, x5, x6) \
-	EX_LD_FP(LOAD(ldd, base + 0x00, %x0)); \
-	EX_LD_FP(LOAD(ldd, base + 0x08, %x1)); \
-	EX_LD_FP(LOAD(ldd, base + 0x10, %x2)); \
-	EX_LD_FP(LOAD(ldd, base + 0x18, %x3)); \
-	EX_LD_FP(LOAD(ldd, base + 0x20, %x4)); \
-	EX_LD_FP(LOAD(ldd, base + 0x28, %x5)); \
-	EX_LD_FP(LOAD(ldd, base + 0x30, %x6));
+	EX_LD_FP(LOAD(ldd, base + 0x00, %x0), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x08, %x1), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x10, %x2), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x18, %x3), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x20, %x4), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x28, %x5), NG2_retl_o2_plus_g1); \
+	EX_LD_FP(LOAD(ldd, base + 0x30, %x6), NG2_retl_o2_plus_g1);
 
 	.register	%g2,#scratch
 	.register	%g3,#scratch
 
 	.text
+#ifndef EX_RETVAL
+#define EX_RETVAL(x)	x
+__restore_fp:
+	VISExitHalf
+__restore_asi:
+	retl
+	 wr	%g0, ASI_AIUS, %asi
+ENTRY(NG2_retl_o2)
+	ba,pt	%xcc, __restore_asi
+	 mov	%o2, %o0
+ENDPROC(NG2_retl_o2)
+ENTRY(NG2_retl_o2_plus_1)
+	ba,pt	%xcc, __restore_asi
+	 add	%o2, 1, %o0
+ENDPROC(NG2_retl_o2_plus_1)
+ENTRY(NG2_retl_o2_plus_4)
+	ba,pt	%xcc, __restore_asi
+	 add	%o2, 4, %o0
+ENDPROC(NG2_retl_o2_plus_4)
+ENTRY(NG2_retl_o2_plus_8)
+	ba,pt	%xcc, __restore_asi
+	 add	%o2, 8, %o0
+ENDPROC(NG2_retl_o2_plus_8)
+ENTRY(NG2_retl_o2_plus_o4_plus_1)
+	add	%o4, 1, %o4
+	ba,pt	%xcc, __restore_asi
+	 add	%o2, %o4, %o0
+ENDPROC(NG2_retl_o2_plus_o4_plus_1)
+ENTRY(NG2_retl_o2_plus_o4_plus_8)
+	add	%o4, 8, %o4
+	ba,pt	%xcc, __restore_asi
+	 add	%o2, %o4, %o0
+ENDPROC(NG2_retl_o2_plus_o4_plus_8)
+ENTRY(NG2_retl_o2_plus_o4_plus_16)
+	add	%o4, 16, %o4
+	ba,pt	%xcc, __restore_asi
+	 add	%o2, %o4, %o0
+ENDPROC(NG2_retl_o2_plus_o4_plus_16)
+ENTRY(NG2_retl_o2_plus_g1_fp)
+	ba,pt	%xcc, __restore_fp
+	 add	%o2, %g1, %o0
+ENDPROC(NG2_retl_o2_plus_g1_fp)
+ENTRY(NG2_retl_o2_plus_g1_plus_64_fp)
+	add	%g1, 64, %g1
+	ba,pt	%xcc, __restore_fp
+	 add	%o2, %g1, %o0
+ENDPROC(NG2_retl_o2_plus_g1_plus_64_fp)
+ENTRY(NG2_retl_o2_plus_g1_plus_1)
+	add	%g1, 1, %g1
+	ba,pt	%xcc, __restore_asi
+	 add	%o2, %g1, %o0
+ENDPROC(NG2_retl_o2_plus_g1_plus_1)
+ENTRY(NG2_retl_o2_and_7_plus_o4)
+	and	%o2, 7, %o2
+	ba,pt	%xcc, __restore_asi
+	 add	%o2, %o4, %o0
+ENDPROC(NG2_retl_o2_and_7_plus_o4)
+ENTRY(NG2_retl_o2_and_7_plus_o4_plus_8)
+	and	%o2, 7, %o2
+	add	%o4, 8, %o4
+	ba,pt	%xcc, __restore_asi
+	 add	%o2, %o4, %o0
+ENDPROC(NG2_retl_o2_and_7_plus_o4_plus_8)
+#endif
+
 	.align		64
 
 	.globl	FUNC_NAME
@@ -230,8 +292,8 @@ FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
 	sub		%g0, %o4, %o4	! bytes to align dst
 	sub		%o2, %o4, %o2
 1:	subcc		%o4, 1, %o4
-	EX_LD(LOAD(ldub, %o1, %g1))
-	EX_ST(STORE(stb, %g1, %o0))
+	EX_LD(LOAD(ldub, %o1, %g1), NG2_retl_o2_plus_o4_plus_1)
+	EX_ST(STORE(stb, %g1, %o0), NG2_retl_o2_plus_o4_plus_1)
 	add		%o1, 1, %o1
 	bne,pt		%XCC, 1b
 	add		%o0, 1, %o0
@@ -281,11 +343,11 @@ FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
 	 nop
 	/* fall through for 0 < low bits < 8 */
 110:	sub		%o4, 64, %g2
-	EX_LD_FP(LOAD_BLK(%g2, %f0))
-1:	EX_ST_FP(STORE_INIT(%g0, %o4 + %g3))
-	EX_LD_FP(LOAD_BLK(%o4, %f16))
+	EX_LD_FP(LOAD_BLK(%g2, %f0), NG2_retl_o2_plus_g1)
+1:	EX_ST_FP(STORE_INIT(%g0, %o4 + %g3), NG2_retl_o2_plus_g1)
+	EX_LD_FP(LOAD_BLK(%o4, %f16), NG2_retl_o2_plus_g1)
 	FREG_FROB(f0, f2, f4, f6, f8, f10, f12, f14, f16)
-	EX_ST_FP(STORE_BLK(%f0, %o4 + %g3))
+	EX_ST_FP(STORE_BLK(%f0, %o4 + %g3), NG2_retl_o2_plus_g1)
 	FREG_MOVE_8(f16, f18, f20, f22, f24, f26, f28, f30)
 	subcc		%g1, 64, %g1
 	add		%o4, 64, %o4
@@ -296,10 +358,10 @@ FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
 
 120:	sub		%o4, 56, %g2
 	FREG_LOAD_7(%g2, f0, f2, f4, f6, f8, f10, f12)
-1:	EX_ST_FP(STORE_INIT(%g0, %o4 + %g3))
-	EX_LD_FP(LOAD_BLK(%o4, %f16))
+1:	EX_ST_FP(STORE_INIT(%g0, %o4 + %g3), NG2_retl_o2_plus_g1)
+	EX_LD_FP(LOAD_BLK(%o4, %f16), NG2_retl_o2_plus_g1)
 	FREG_FROB(f0, f2, f4, f6, f8, f10, f12, f16, f18)
-	EX_ST_FP(STORE_BLK(%f0, %o4 + %g3))
+	EX_ST_FP(STORE_BLK(%f0, %o4 + %g3), NG2_retl_o2_plus_g1)
 	FREG_MOVE_7(f18, f20, f22, f24, f26, f28, f30)
 	subcc		%g1, 64, %g1
 	add		%o4, 64, %o4
@@ -310,10 +372,10 @@ FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
 <