diff --git a/arch/arm/include/asm/cacheflush.h b/arch/arm/include/asm/cacheflush.h
index 004c1bc95d2b9082de58e0626134a779705620b6..e4448e16046dd32ab69ca17ef65a6b21745a3e0b 100644
--- a/arch/arm/include/asm/cacheflush.h
+++ b/arch/arm/include/asm/cacheflush.h
@@ -215,7 +215,9 @@ static inline void vivt_flush_cache_mm(struct mm_struct *mm)
 static inline void
 vivt_flush_cache_range(struct vm_area_struct *vma, unsigned long start, unsigned long end)
 {
-	if (cpumask_test_cpu(smp_processor_id(), mm_cpumask(vma->vm_mm)))
+	struct mm_struct *mm = vma->vm_mm;
+
+	if (!mm || cpumask_test_cpu(smp_processor_id(), mm_cpumask(mm)))
 		__cpuc_flush_user_range(start & PAGE_MASK, PAGE_ALIGN(end),
 					vma->vm_flags);
 }
@@ -223,7 +225,9 @@ vivt_flush_cache_range(struct vm_area_struct *vma, unsigned long start, unsigned
 static inline void
 vivt_flush_cache_page(struct vm_area_struct *vma, unsigned long user_addr, unsigned long pfn)
 {
-	if (cpumask_test_cpu(smp_processor_id(), mm_cpumask(vma->vm_mm))) {
+	struct mm_struct *mm = vma->vm_mm;
+
+	if (!mm || cpumask_test_cpu(smp_processor_id(), mm_cpumask(mm))) {
 		unsigned long addr = user_addr & PAGE_MASK;
 		__cpuc_flush_user_range(addr, addr + PAGE_SIZE, vma->vm_flags);
 	}
diff --git a/arch/arm/include/asm/mutex.h b/arch/arm/include/asm/mutex.h
index 93226cf23ae0a838e192be1d675ee886be9dd52e..b1479fd04a951f1d82d55f77fd01cb9f0b072f38 100644
--- a/arch/arm/include/asm/mutex.h
+++ b/arch/arm/include/asm/mutex.h
@@ -7,121 +7,10 @@
  */
 #ifndef _ASM_MUTEX_H
 #define _ASM_MUTEX_H
-
-#if __LINUX_ARM_ARCH__ < 6
-/* On pre-ARMv6 hardware the swp based implementation is the most efficient. */
-# include <asm-generic/mutex-xchg.h>
-#else
-
 /*
- * Attempting to lock a mutex on ARMv6+ can be done with a bastardized
- * atomic decrement (it is not a reliable atomic decrement but it satisfies
- * the defined semantics for our purpose, while being smaller and faster
- * than a real atomic decrement or atomic swap.  The idea is to attempt
- * decrementing the lock value only once.  If once decremented it isn't zero,
- * or if its store-back fails due to a dispute on the exclusive store, we
- * simply bail out immediately through the slow path where the lock will be
- * reattempted until it succeeds.
+ * On pre-ARMv6 hardware this results in a swp-based implementation,
+ * which is the most efficient. For ARMv6+, we emit a pair of exclusive
+ * accesses instead.
  */
-static inline void
-__mutex_fastpath_lock(atomic_t *count, void (*fail_fn)(atomic_t *))
-{
-	int __ex_flag, __res;
-
-	__asm__ (
-
-		"ldrex	%0, [%2]	\n\t"
-		"sub	%0, %0, #1	\n\t"
-		"strex	%1, %0, [%2]	"
-
-		: "=&r" (__res), "=&r" (__ex_flag)
-		: "r" (&(count)->counter)
-		: "cc","memory" );
-
-	__res |= __ex_flag;
-	if (unlikely(__res != 0))
-		fail_fn(count);
-}
-
-static inline int
-__mutex_fastpath_lock_retval(atomic_t *count, int (*fail_fn)(atomic_t *))
-{
-	int __ex_flag, __res;
-
-	__asm__ (
-
-		"ldrex	%0, [%2]	\n\t"
-		"sub	%0, %0, #1	\n\t"
-		"strex	%1, %0, [%2]	"
-
-		: "=&r" (__res), "=&r" (__ex_flag)
-		: "r" (&(count)->counter)
-		: "cc","memory" );
-
-	__res |= __ex_flag;
-	if (unlikely(__res != 0))
-		__res = fail_fn(count);
-	return __res;
-}
-
-/*
- * Same trick is used for the unlock fast path. However the original value,
- * rather than the result, is used to test for success in order to have
- * better generated assembly.
- */
-static inline void
-__mutex_fastpath_unlock(atomic_t *count, void (*fail_fn)(atomic_t *))
-{
-	int __ex_flag, __res, __orig;
-
-	__asm__ (
-
-		"ldrex	%0, [%3]	\n\t"
-		"add	%1, %0, #1	\n\t"
-		"strex	%2, %1, [%3]	"
-
-		: "=&r" (__orig), "=&r" (__res), "=&r" (__ex_flag)
-		: "r" (&(count)->counter)
-		: "cc","memory" );
-
-	__orig |= __ex_flag;
-	if (unlikely(__orig != 0))
-		fail_fn(count);
-}
-
-/*
- * If the unlock was done on a contended lock, or if the unlock simply fails
- * then the mutex remains locked.
- */
-#define __mutex_slowpath_needs_to_unlock()	1
-
-/*
- * For __mutex_fastpath_trylock we use another construct which could be
- * described as a "single value cmpxchg".
- *
- * This provides the needed trylock semantics like cmpxchg would, but it is
- * lighter and less generic than a true cmpxchg implementation.
- */
-static inline int
-__mutex_fastpath_trylock(atomic_t *count, int (*fail_fn)(atomic_t *))
-{
-	int __ex_flag, __res, __orig;
-
-	__asm__ (
-
-		"1: ldrex	%0, [%3]	\n\t"
-		"subs		%1, %0, #1	\n\t"
-		"strexeq	%2, %1, [%3]	\n\t"
-		"movlt		%0, #0		\n\t"
-		"cmpeq		%2, #0		\n\t"
-		"bgt		1b		"
-
-		: "=&r" (__orig), "=&r" (__res), "=&r" (__ex_flag)
-		: "r" (&count->counter)
-		: "cc", "memory" );
-
-	return __orig;
-}
-
-#endif
+#include <asm-generic/mutex-xchg.h>
 #endif
diff --git a/arch/arm/include/asm/setup.h b/arch/arm/include/asm/setup.h
index 23ebc0c82a3975ae5c455dd39598e93ab33922e7..24d284a1bfc75faa0ed90fb7882a67f00128f336 100644
--- a/arch/arm/include/asm/setup.h
+++ b/arch/arm/include/asm/setup.h
@@ -196,7 +196,7 @@ static const struct tagtable __tagtable_##fn __tag = { tag, fn }
 
 struct membank {
 	phys_addr_t start;
-	unsigned long size;
+	phys_addr_t size;
 	unsigned int highmem;
 };
 
@@ -217,7 +217,7 @@ extern struct meminfo meminfo;
 #define bank_phys_end(bank)	((bank)->start + (bank)->size)
 #define bank_phys_size(bank)	(bank)->size
 
-extern int arm_add_memory(phys_addr_t start, unsigned long size);
+extern int arm_add_memory(phys_addr_t start, phys_addr_t size);
 extern void early_print(const char *str, ...);
 extern void dump_machine_table(void);
 
diff --git a/arch/arm/kernel/entry-armv.S b/arch/arm/kernel/entry-armv.S
index 0d1851ca6eb993a628f623c257487200fe529604..0f82098c9bfe3618115dcc4e8b2d74d95c7ddcf6 100644
--- a/arch/arm/kernel/entry-armv.S
+++ b/arch/arm/kernel/entry-armv.S
@@ -244,6 +244,19 @@ svc_preempt:
 	b	1b
 #endif
 
+__und_fault:
+	@ Correct the PC such that it is pointing at the instruction
+	@ which caused the fault.  If the faulting instruction was ARM
+	@ the PC will be pointing at the next instruction, and have to
+	@ subtract 4.  Otherwise, it is Thumb, and the PC will be
+	@ pointing at the second half of the Thumb instruction.  We
+	@ have to subtract 2.
+	ldr	r2, [r0, #S_PC]
+	sub	r2, r2, r1
+	str	r2, [r0, #S_PC]
+	b	do_undefinstr
+ENDPROC(__und_fault)
+
 	.align	5
 __und_svc:
 #ifdef CONFIG_KPROBES
@@ -261,25 +274,32 @@ __und_svc:
 	@
 	@  r0 - instruction
 	@
-#ifndef	CONFIG_THUMB2_KERNEL
+#ifndef CONFIG_THUMB2_KERNEL
 	ldr	r0, [r4, #-4]
 #else
+	mov	r1, #2
 	ldrh	r0, [r4, #-2]			@ Thumb instruction at LR - 2
 	cmp	r0, #0xe800			@ 32-bit instruction if xx >= 0
-	ldrhhs	r9, [r4]			@ bottom 16 bits
-	orrhs	r0, r9, r0, lsl #16
+	blo	__und_svc_fault
+	ldrh	r9, [r4]			@ bottom 16 bits
+	add	r4, r4, #2
+	str	r4, [sp, #S_PC]
+	orr	r0, r9, r0, lsl #16
 #endif
-	adr	r9, BSYM(1f)
+	adr	r9, BSYM(__und_svc_finish)
 	mov	r2, r4
 	bl	call_fpe
 
+	mov	r1, #4				@ PC correction to apply
+__und_svc_fault:
 	mov	r0, sp				@ struct pt_regs *regs
-	bl	do_undefinstr
+	bl	__und_fault
 
 	@
 	@ IRQs off again before pulling preserved data off the stack
 	@
-1:	disable_irq_notrace
+__und_svc_finish:
+	disable_irq_notrace
 
 	@
 	@ restore SPSR and restart the instruction
@@ -423,25 +443,33 @@ __und_usr:
 	mov	r2, r4
 	mov	r3, r5
 
+	@ r2 = regs->ARM_pc, which is either 2 or 4 bytes ahead of the
+	@      faulting instruction depending on Thumb mode.
+	@ r3 = regs->ARM_cpsr
 	@
-	@ fall through to the emulation code, which returns using r9 if
-	@ it has emulated the instruction, or the more conventional lr
-	@ if we are to treat this as a real undefined instruction
-	@
-	@  r0 - instruction
+	@ The emulation code returns using r9 if it has emulated the
+	@ instruction, or the more conventional lr if we are to treat
+	@ this as a real undefined instruction
 	@
 	adr	r9, BSYM(ret_from_exception)
-	adr	lr, BSYM(__und_usr_unknown)
+
 	tst	r3, #PSR_T_BIT			@ Thumb mode?
-	itet	eq				@ explicit IT needed for the 1f label
-	subeq	r4, r2, #4			@ ARM instr at LR - 4
-	subne	r4, r2, #2			@ Thumb instr at LR - 2
-1:	ldreqt	r0, [r4]
+	bne	__und_usr_thumb
+	sub	r4, r2, #4			@ ARM instr at LR - 4
+1:	ldrt	r0, [r4]
 #ifdef CONFIG_CPU_ENDIAN_BE8
-	reveq	r0, r0				@ little endian instruction
+	rev	r0, r0				@ little endian instruction
 #endif
-	beq	call_fpe
+	@ r0 = 32-bit ARM instruction which caused the exception
+	@ r2 = PC value for the following instruction (:= regs->ARM_pc)
+	@ r4 = PC value for the faulting instruction
+	@ lr = 32-bit undefined instruction function
+	adr	lr, BSYM(__und_usr_fault_32)
+	b	call_fpe
+
+__und_usr_thumb:
 	@ Thumb instruction
+	sub	r4, r2, #2			@ First half of thumb instr at LR - 2
 #if CONFIG_ARM_THUMB && __LINUX_ARM_ARCH__ >= 6 && CONFIG_CPU_V7
 /*
  * Thumb-2 instruction handling.  Note that because pre-v6 and >= v6 platforms
@@ -455,7 +483,7 @@ __und_usr:
 	ldr	r5, .LCcpu_architecture
 	ldr	r5, [r5]
 	cmp	r5, #CPU_ARCH_ARMv7
-	blo	__und_usr_unknown
+	blo	__und_usr_fault_16		@ 16bit undefined instruction
 /*
  * The following code won't get run unless the running CPU really is v7, so
  * coding round the lack of ldrht on older arches is pointless.  Temporarily
@@ -463,15 +491,18 @@ __und_usr:
  */
 	.arch	armv6t2
 #endif
-2:
- ARM(	ldrht	r5, [r4], #2	)
- THUMB(	ldrht	r5, [r4]	)
- THUMB(	add	r4, r4, #2	)
+2:	ldrht	r5, [r4]
 	cmp	r5, #0xe800			@ 32bit instruction if xx != 0
-	blo	__und_usr_unknown
-3:	ldrht	r0, [r4]
+	blo	__und_usr_fault_16		@ 16bit undefined instruction
+3:	ldrht	r0, [r2]
 	add	r2, r2, #2			@ r2 is PC + 2, make it PC + 4
+	str	r2, [sp, #S_PC]			@ it's a 2x16bit instr, update
 	orr	r0, r0, r5, lsl #16
+	adr	lr, BSYM(__und_usr_fault_32)
+	@ r0 = the two 16-bit Thumb instructions which caused the exception
+	@ r2 = PC value for the following Thumb instruction (:= regs->ARM_pc)
+	@ r4 = PC value for the first 16-bit Thumb instruction
+	@ lr = 32bit undefined instruction function
 
 #if __LINUX_ARM_ARCH__ < 7
 /* If the target arch was overridden, change it back: */
@@ -482,17 +513,13 @@ __und_usr:
 #endif
 #endif /* __LINUX_ARM_ARCH__ < 7 */
 #else /* !(CONFIG_ARM_THUMB && __LINUX_ARM_ARCH__ >= 6 && CONFIG_CPU_V7) */
-	b	__und_usr_unknown
+	b	__und_usr_fault_16
 #endif
- UNWIND(.fnend		)
+ UNWIND(.fnend)
 ENDPROC(__und_usr)
 
-	@
-	@ fallthrough to call_fpe
-	@
-
 /*
- * The out of line fixup for the ldrt above.
+ * The out of line fixup for the ldrt instructions above.
  */
 	.pushsection .fixup, "ax"
 	.align	2
@@ -524,11 +551,12 @@ ENDPROC(__und_usr)
  * NEON handler code.
  *
  * Emulators may wish to make use of the following registers:
- *  r0  = instruction opcode.
- *  r2  = PC+4
+ *  r0  = instruction opcode (32-bit ARM or two 16-bit Thumb)
+ *  r2  = PC value to resume execution after successful emulation
  *  r9  = normal "successful" return address
- *  r10 = this threads thread_info structure.
+ *  r10 = this threads thread_info structure
  *  lr  = unrecognised instruction return address
+ * IRQs disabled, FIQs enabled.
  */
 	@
 	@ Fall-through from Thumb-2 __und_usr
@@ -659,12 +687,17 @@ ENTRY(no_fp)
 	mov	pc, lr
 ENDPROC(no_fp)
 
-__und_usr_unknown:
-	enable_irq
+__und_usr_fault_32:
+	mov	r1, #4
+	b	1f
+__und_usr_fault_16:
+	mov	r1, #2
+1:	enable_irq
 	mov	r0, sp
 	adr	lr, BSYM(ret_from_exception)
-	b	do_undefinstr
-ENDPROC(__und_usr_unknown)
+	b	__und_fault
+ENDPROC(__und_usr_fault_32)
+ENDPROC(__und_usr_fault_16)
 
 	.align	5
 __pabt_usr:
diff --git a/arch/arm/kernel/ftrace.c b/arch/arm/kernel/ftrace.c
index df0bf0c8cb790a5ee501c55a204a45ddbcb48aab..34e56647dceeee88d99f65d5fd0a6e00fb46a0fd 100644
--- a/arch/arm/kernel/ftrace.c
+++ b/arch/arm/kernel/ftrace.c
@@ -179,19 +179,20 @@ void prepare_ftrace_return(unsigned long *parent, unsigned long self_addr,
 	old = *parent;
 	*parent = return_hooker;
 
-	err = ftrace_push_return_trace(old, self_addr, &trace.depth,
-				       frame_pointer);
-	if (err == -EBUSY) {
-		*parent = old;
-		return;
-	}
-
 	trace.func = self_addr;
+	trace.depth = current->curr_ret_stack + 1;
 
 	/* Only trace if the calling function expects to */
 	if (!ftrace_graph_entry(&trace)) {
-		current->curr_ret_stack--;
 		*parent = old;
+		return;
+	}
+
+	err = ftrace_push_return_trace(old, self_addr, &trace.depth,
+				       frame_pointer);
+	if (err == -EBUSY) {
+		*parent = old;
+		return;
 	}
 }
 
diff --git a/arch/arm/kernel/process.c b/arch/arm/kernel/process.c
index 19c95ea65b2f25c25371734a7a03f147e929978e..693b744fd572f163f053c9fef42d563e246e04d3 100644
--- a/arch/arm/kernel/process.c
+++ b/arch/arm/kernel/process.c
@@ -247,6 +247,7 @@ void machine_shutdown(void)
 void machine_halt(void)
 {
 	machine_shutdown();
+	local_irq_disable();
 	while (1);
 }
 
@@ -268,6 +269,7 @@ void machine_restart(char *cmd)
 
 	/* Whoops - the platform was unable to reboot. Tell the user! */
 	printk("Reboot failed -- System halted\n");
+	local_irq_disable();
 	while (1);
 }
 
diff --git a/arch/arm/kernel/setup.c b/arch/arm/kernel/setup.c
index e15d83bb4ea378c1316db6a7d705eab30248a98b..a81dcecc734388f7745e399e38504645f4e0e758 100644
--- a/arch/arm/kernel/setup.c
+++ b/arch/arm/kernel/setup.c
@@ -508,7 +508,7 @@ void __init dump_machine_table(void)
 		/* can't use cpu_relax() here as it may require MMU setup */;
 }
 
-int __init arm_add_memory(phys_addr_t start, unsigned long size)
+int __init arm_add_memory(phys_addr_t start, phys_addr_t size)
 {
 	struct membank *bank = &meminfo.bank[meminfo.nr_banks];
 
@@ -538,7 +538,7 @@ int __init arm_add_memory(phys_addr_t start, unsigned long size)
 	}
 #endif
 
-	bank->size = size & PAGE_MASK;
+	bank->size = size & ~(phys_addr_t)(PAGE_SIZE - 1);
 
 	/*
 	 * Check whether this memory region has non-zero size or
@@ -558,7 +558,7 @@ int __init arm_add_memory(phys_addr_t start, unsigned long size)
 static int __init early_mem(char *p)
 {
 	static int usermem __initdata = 0;
-	unsigned long size;
+	phys_addr_t size;
 	phys_addr_t start;
 	char *endp;
 
diff --git a/arch/arm/kernel/smp.c b/arch/arm/kernel/smp.c
index aea74f5bc34abdcefbd90570e3c1b7e534b29763..ebd8ad274d76bb82488240e9543d7a1d99b5c674 100644
--- a/arch/arm/kernel/smp.c
+++ b/arch/arm/kernel/smp.c
@@ -563,7 +563,8 @@ void smp_send_stop(void)
 
 	cpumask_copy(&mask, cpu_online_mask);
 	cpumask_clear_cpu(smp_processor_id(), &mask);
-	smp_cross_call(&mask, IPI_CPU_STOP);
+	if (!cpumask_empty(&mask))
+		smp_cross_call(&mask, IPI_CPU_STOP);
 
 	/* Wait up to one second for other CPUs to stop */
 	timeout = USEC_PER_SEC;
diff --git a/arch/arm/kernel/traps.c b/arch/arm/kernel/traps.c
index 8b97d739b17b1040f4f3e4bb7fe069d84808fa72..7978d4f0f3aef59ee177a0525db44c6819110ed6 100644
--- a/arch/arm/kernel/traps.c
+++ b/arch/arm/kernel/traps.c
@@ -402,18 +402,10 @@ static int call_undef_hook(struct pt_regs *regs, unsigned int instr)
 
 asmlinkage void __exception do_undefinstr(struct pt_regs *regs)
 {
-	unsigned int correction = thumb_mode(regs) ? 2 : 4;
 	unsigned int instr;
 	siginfo_t info;
 	void __user *pc;
 
-	/*
-	 * According to the ARM ARM, PC is 2 or 4 bytes ahead,
-	 * depending whether we're in Thumb mode or not.
-	 * Correct this offset.
-	 */
-	regs->ARM_pc -= correction;
-
 	pc = (void __user *)instruction_pointer(regs);
 
 	if (processor_mode(regs) == SVC_MODE) {
diff --git a/arch/arm/mm/tlb-v7.S b/arch/arm/mm/tlb-v7.S
index 845f461f8ec16847e69414f5c966a9ca4f961db6..c2021139cb563fd14a74c327d453f3cf0cee183d 100644
--- a/arch/arm/mm/tlb-v7.S
+++ b/arch/arm/mm/tlb-v7.S
@@ -38,11 +38,19 @@ ENTRY(v7wbi_flush_user_tlb_range)
 	dsb
 	mov	r0, r0, lsr #PAGE_SHIFT		@ align address
 	mov	r1, r1, lsr #PAGE_SHIFT
+#ifdef CONFIG_ARM_ERRATA_720789
+	mov	r3, #0
+#else
 	asid	r3, r3				@ mask ASID
+#endif
 	orr	r0, r3, r0, lsl #PAGE_SHIFT	@ Create initial MVA
 	mov	r1, r1, lsl #PAGE_SHIFT
 1:
+#ifdef CONFIG_ARM_ERRATA_720789
+	ALT_SMP(mcr	p15, 0, r0, c8, c3, 3)	@ TLB invalidate U MVA all ASID (shareable)
+#else
 	ALT_SMP(mcr	p15, 0, r0, c8, c3, 1)	@ TLB invalidate U MVA (shareable)
+#endif
 	ALT_UP(mcr	p15, 0, r0, c8, c7, 1)	@ TLB invalidate U MVA
 
 	add	r0, r0, #PAGE_SZ
@@ -67,7 +75,11 @@ ENTRY(v7wbi_flush_kern_tlb_range)
 	mov	r0, r0, lsl #PAGE_SHIFT
 	mov	r1, r1, lsl #PAGE_SHIFT
 1:
+#ifdef CONFIG_ARM_ERRATA_720789
+	ALT_SMP(mcr	p15, 0, r0, c8, c3, 3)	@ TLB invalidate U MVA all ASID (shareable)
+#else
 	ALT_SMP(mcr	p15, 0, r0, c8, c3, 1)	@ TLB invalidate U MVA (shareable)
+#endif
 	ALT_UP(mcr	p15, 0, r0, c8, c7, 1)	@ TLB invalidate U MVA
 	add	r0, r0, #PAGE_SZ
 	cmp	r0, r1
diff --git a/arch/arm/vfp/entry.S b/arch/arm/vfp/entry.S
index 4fa9903b83cf5dbb54a15623ec45fd00552d8521..cc926c98598141a2fbcdb762148ec17782982d06 100644
--- a/arch/arm/vfp/entry.S
+++ b/arch/arm/vfp/entry.S
@@ -7,18 +7,20 @@
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License version 2 as
  * published by the Free Software Foundation.
- *
- * Basic entry code, called from the kernel's undefined instruction trap.
- *  r0  = faulted instruction
- *  r5  = faulted PC+4
- *  r9  = successful return
- *  r10 = thread_info structure
- *  lr  = failure return
  */
 #include <asm/thread_info.h>
 #include <asm/vfpmacros.h>
 #include "../kernel/entry-header.S"
 
+@ VFP entry point.
+@
+@  r0  = instruction opcode (32-bit ARM or two 16-bit Thumb)
+@  r2  = PC value to resume execution after successful emulation
+@  r9  = normal "successful" return address
+@  r10 = this threads thread_info structure
+@  lr  = unrecognised instruction return address
+@  IRQs disabled.
+@
 ENTRY(do_vfp)
 #ifdef CONFIG_PREEMPT
 	ldr	r4, [r10, #TI_PREEMPT]	@ get preempt count
diff --git a/arch/arm/vfp/vfphw.S b/arch/arm/vfp/vfphw.S
index d50f0e486cf2b322b6e8eff1baa7c73272e66c70..ea0349f6358658065b52aa1473e877ee4fa8f5ba 100644
--- a/arch/arm/vfp/vfphw.S
+++ b/arch/arm/vfp/vfphw.S
@@ -62,13 +62,13 @@
 
 @ VFP hardware support entry point.
 @
-@  r0  = faulted instruction
-@  r2  = faulted PC+4
-@  r9  = successful return
+@  r0  = instruction opcode (32-bit ARM or two 16-bit Thumb)
+@  r2  = PC value to resume execution after successful emulation
+@  r9  = normal "successful" return address
 @  r10 = vfp_state union
 @  r11 = CPU number
-@  lr  = failure return
-
+@  lr  = unrecognised instruction return address
+@  IRQs enabled.
 ENTRY(vfp_support_entry)
 	DBGSTR3	"instr %08x pc %08x state %p", r0, r2, r10
 
@@ -162,9 +162,12 @@ vfp_hw_state_valid:
 					@ exception before retrying branch
 					@ out before setting an FPEXC that
 					@ stops us reading stuff
-	VFPFMXR	FPEXC, r1		@ restore FPEXC last
-	sub	r2, r2, #4
-	str	r2, [sp, #S_PC]		@ retry the instruction
+	VFPFMXR	FPEXC, r1		@ Restore FPEXC last
+	sub	r2, r2, #4		@ Retry current instruction - if Thumb
+	str	r2, [sp, #S_PC]		@ mode it's two 16-bit instructions,
+					@ else it's one 32-bit instruction, so
+					@ always subtract 4 from the following
+					@ instruction address.
 #ifdef CONFIG_PREEMPT
 	get_thread_info	r10
 	ldr	r4, [r10, #TI_PREEMPT]	@ get preempt count
diff --git a/arch/arm/vfp/vfpmodule.c b/arch/arm/vfp/vfpmodule.c
index 586961929e964ce63d70987c0c7c36184aff4db8..fb849d044bde9b231d481201c985f84cc37c0a3d 100644
--- a/arch/arm/vfp/vfpmodule.c
+++ b/arch/arm/vfp/vfpmodule.c
@@ -457,10 +457,16 @@ static int vfp_pm_suspend(void)
 
 		/* disable, just in case */
 		fmxr(FPEXC, fmrx(FPEXC) & ~FPEXC_EN);
+	} else if (vfp_current_hw_state[ti->cpu]) {
+#ifndef CONFIG_SMP
+		fmxr(FPEXC, fpexc | FPEXC_EN);
+		vfp_save_state(vfp_current_hw_state[ti->cpu], fpexc);
+		fmxr(FPEXC, fpexc);
+#endif
 	}
 
 	/* clear any information we had about last context state */
-	memset(vfp_current_hw_state, 0, sizeof(vfp_current_hw_state));
+	vfp_current_hw_state[ti->cpu] = NULL;
 
 	return 0;
 }