diff --git a/arch/sparc/include/asm/head_64.h b/arch/sparc/include/asm/head_64.h
index 10e9dabc4c41337c9f4a1dfc970aa22e9e24436f..f0700cfeedd7b270fa874aae03c8df6bf862152d 100644
--- a/arch/sparc/include/asm/head_64.h
+++ b/arch/sparc/include/asm/head_64.h
@@ -15,6 +15,10 @@
 
 #define	PTREGS_OFF	(STACK_BIAS + STACKFRAME_SZ)
 
+#define	RTRAP_PSTATE		(PSTATE_TSO|PSTATE_PEF|PSTATE_PRIV|PSTATE_IE)
+#define	RTRAP_PSTATE_IRQOFF	(PSTATE_TSO|PSTATE_PEF|PSTATE_PRIV)
+#define RTRAP_PSTATE_AG_IRQOFF	(PSTATE_TSO|PSTATE_PEF|PSTATE_PRIV|PSTATE_AG)
+
 #define __CHEETAH_ID	0x003e0014
 #define __JALAPENO_ID	0x003e0016
 #define __SERRANO_ID	0x003e0022
diff --git a/arch/sparc/include/asm/ttable.h b/arch/sparc/include/asm/ttable.h
index 71b5a67522abb2bb0d8457819612aa26e948fdc8..781b9f1dbdc2d24b54fe533c0961d8eab8efa44b 100644
--- a/arch/sparc/include/asm/ttable.h
+++ b/arch/sparc/include/asm/ttable.h
@@ -589,8 +589,8 @@ user_rtt_fill_64bit:					\
 	 restored;					\
 	nop; nop; nop; nop; nop; nop;			\
 	nop; nop; nop; nop; nop;			\
-	ba,a,pt	%xcc, user_rtt_fill_fixup;		\
-	ba,a,pt	%xcc, user_rtt_fill_fixup;		\
+	ba,a,pt	%xcc, user_rtt_fill_fixup_dax;		\
+	ba,a,pt	%xcc, user_rtt_fill_fixup_mna;		\
 	ba,a,pt	%xcc, user_rtt_fill_fixup;
 
 
@@ -652,8 +652,8 @@ user_rtt_fill_32bit:					\
 	 restored;					\
 	nop; nop; nop; nop; nop;			\
 	nop; nop; nop;					\
-	ba,a,pt	%xcc, user_rtt_fill_fixup;		\
-	ba,a,pt	%xcc, user_rtt_fill_fixup;		\
+	ba,a,pt	%xcc, user_rtt_fill_fixup_dax;		\
+	ba,a,pt	%xcc, user_rtt_fill_fixup_mna;		\
 	ba,a,pt	%xcc, user_rtt_fill_fixup;
 
 
diff --git a/arch/sparc/kernel/Makefile b/arch/sparc/kernel/Makefile
index 7cf9c6ea3f1f210c0856351e47d1a4252913667b..fdb13327fded36a313b054783adf1654725d237f 100644
--- a/arch/sparc/kernel/Makefile
+++ b/arch/sparc/kernel/Makefile
@@ -21,6 +21,7 @@ CFLAGS_REMOVE_perf_event.o := -pg
 CFLAGS_REMOVE_pcr.o := -pg
 endif
 
+obj-$(CONFIG_SPARC64)   += urtt_fill.o
 obj-$(CONFIG_SPARC32)   += entry.o wof.o wuf.o
 obj-$(CONFIG_SPARC32)   += etrap_32.o
 obj-$(CONFIG_SPARC32)   += rtrap_32.o
diff --git a/arch/sparc/kernel/rtrap_64.S b/arch/sparc/kernel/rtrap_64.S
index d08bdaffdbfccbafc9b3d67c30267fe397eb917f..216948ca43829d0c0a5be837ff183f86d03dfb49 100644
--- a/arch/sparc/kernel/rtrap_64.S
+++ b/arch/sparc/kernel/rtrap_64.S
@@ -14,10 +14,6 @@
 #include <asm/visasm.h>
 #include <asm/processor.h>
 
-#define		RTRAP_PSTATE		(PSTATE_TSO|PSTATE_PEF|PSTATE_PRIV|PSTATE_IE)
-#define		RTRAP_PSTATE_IRQOFF	(PSTATE_TSO|PSTATE_PEF|PSTATE_PRIV)
-#define		RTRAP_PSTATE_AG_IRQOFF	(PSTATE_TSO|PSTATE_PEF|PSTATE_PRIV|PSTATE_AG)
-
 #ifdef CONFIG_CONTEXT_TRACKING
 # define SCHEDULE_USER schedule_user
 #else
@@ -242,52 +238,17 @@ rt_continue:	ldx			[%sp + PTREGS_OFF + PT_V9_G1], %g1
 		 wrpr			%g1, %cwp
 		ba,a,pt			%xcc, user_rtt_fill_64bit
 
-user_rtt_fill_fixup:
-		rdpr	%cwp, %g1
-		add	%g1, 1, %g1
-		wrpr	%g1, 0x0, %cwp
-
-		rdpr	%wstate, %g2
-		sll	%g2, 3, %g2
-		wrpr	%g2, 0x0, %wstate
-
-		/* We know %canrestore and %otherwin are both zero.  */
-
-		sethi	%hi(sparc64_kern_pri_context), %g2
-		ldx	[%g2 + %lo(sparc64_kern_pri_context)], %g2
-		mov	PRIMARY_CONTEXT, %g1
-
-661:		stxa	%g2, [%g1] ASI_DMMU
-		.section .sun4v_1insn_patch, "ax"
-		.word	661b
-		stxa	%g2, [%g1] ASI_MMU
-		.previous
-
-		sethi	%hi(KERNBASE), %g1
-		flush	%g1
+user_rtt_fill_fixup_dax:
+		ba,pt	%xcc, user_rtt_fill_fixup_common
+		 mov	1, %g3
 
-		or	%g4, FAULT_CODE_WINFIXUP, %g4
-		stb	%g4, [%g6 + TI_FAULT_CODE]
-		stx	%g5, [%g6 + TI_FAULT_ADDR]
+user_rtt_fill_fixup_mna:
+		ba,pt	%xcc, user_rtt_fill_fixup_common
+		 mov	2, %g3
 
-		mov	%g6, %l1
-		wrpr	%g0, 0x0, %tl
-
-661:		nop
-		.section		.sun4v_1insn_patch, "ax"
-		.word			661b
-		SET_GL(0)
-		.previous
-
-		wrpr	%g0, RTRAP_PSTATE, %pstate
-
-		mov	%l1, %g6
-		ldx	[%g6 + TI_TASK], %g4
-		LOAD_PER_CPU_BASE(%g5, %g6, %g1, %g2, %g3)
-		call	do_sparc64_fault
-		 add	%sp, PTREGS_OFF, %o0
-		ba,pt	%xcc, rtrap
-		 nop
+user_rtt_fill_fixup:
+		ba,pt	%xcc, user_rtt_fill_fixup_common
+		 clr	%g3
 
 user_rtt_pre_restore:
 		add			%g1, 1, %g1
diff --git a/arch/sparc/kernel/signal32.c b/arch/sparc/kernel/signal32.c
index 3c25241fa5cbd416ca5f95bff873c72c609a6a10..91cc2f4ae4d98ba67ae74e5e31cf58ee06d57db4 100644
--- a/arch/sparc/kernel/signal32.c
+++ b/arch/sparc/kernel/signal32.c
@@ -138,12 +138,24 @@ int copy_siginfo_from_user32(siginfo_t *to, compat_siginfo_t __user *from)
 	return 0;
 }
 
+/* Checks if the fp is valid.  We always build signal frames which are
+ * 16-byte aligned, therefore we can always enforce that the restore
+ * frame has that property as well.
+ */
+static bool invalid_frame_pointer(void __user *fp, int fplen)
+{
+	if ((((unsigned long) fp) & 15) ||
+	    ((unsigned long)fp) > 0x100000000ULL - fplen)
+		return true;
+	return false;
+}
+
 void do_sigreturn32(struct pt_regs *regs)
 {
 	struct signal_frame32 __user *sf;
 	compat_uptr_t fpu_save;
 	compat_uptr_t rwin_save;
-	unsigned int psr;
+	unsigned int psr, ufp;
 	unsigned int pc, npc;
 	sigset_t set;
 	compat_sigset_t seta;
@@ -158,11 +170,16 @@ void do_sigreturn32(struct pt_regs *regs)
 	sf = (struct signal_frame32 __user *) regs->u_regs[UREG_FP];
 
 	/* 1. Make sure we are not getting garbage from the user */
-	if (!access_ok(VERIFY_READ, sf, sizeof(*sf)) ||
-	    (((unsigned long) sf) & 3))
+	if (invalid_frame_pointer(sf, sizeof(*sf)))
+		goto segv;
+
+	if (get_user(ufp, &sf->info.si_regs.u_regs[UREG_FP]))
+		goto segv;
+
+	if (ufp & 0x7)
 		goto segv;
 
-	if (get_user(pc, &sf->info.si_regs.pc) ||
+	if (__get_user(pc, &sf->info.si_regs.pc) ||
 	    __get_user(npc, &sf->info.si_regs.npc))
 		goto segv;
 
@@ -227,7 +244,7 @@ void do_sigreturn32(struct pt_regs *regs)
 asmlinkage void do_rt_sigreturn32(struct pt_regs *regs)
 {
 	struct rt_signal_frame32 __user *sf;
-	unsigned int psr, pc, npc;
+	unsigned int psr, pc, npc, ufp;
 	compat_uptr_t fpu_save;
 	compat_uptr_t rwin_save;
 	sigset_t set;
@@ -242,11 +259,16 @@ asmlinkage void do_rt_sigreturn32(struct pt_regs *regs)
 	sf = (struct rt_signal_frame32 __user *) regs->u_regs[UREG_FP];
 
 	/* 1. Make sure we are not getting garbage from the user */
-	if (!access_ok(VERIFY_READ, sf, sizeof(*sf)) ||
-	    (((unsigned long) sf) & 3))
+	if (invalid_frame_pointer(sf, sizeof(*sf)))
 		goto segv;
 
-	if (get_user(pc, &sf->regs.pc) || 
+	if (get_user(ufp, &sf->regs.u_regs[UREG_FP]))
+		goto segv;
+
+	if (ufp & 0x7)
+		goto segv;
+
+	if (__get_user(pc, &sf->regs.pc) || 
 	    __get_user(npc, &sf->regs.npc))
 		goto segv;
 
@@ -307,14 +329,6 @@ asmlinkage void do_rt_sigreturn32(struct pt_regs *regs)
 	force_sig(SIGSEGV, current);
 }
 
-/* Checks if the fp is valid */
-static int invalid_frame_pointer(void __user *fp, int fplen)
-{
-	if ((((unsigned long) fp) & 7) || ((unsigned long)fp) > 0x100000000ULL - fplen)
-		return 1;
-	return 0;
-}
-
 static void __user *get_sigframe(struct ksignal *ksig, struct pt_regs *regs, unsigned long framesize)
 {
 	unsigned long sp;
diff --git a/arch/sparc/kernel/signal_32.c b/arch/sparc/kernel/signal_32.c
index 52aa5e4ce5e77c0716e083dd41372c16e5ac8967..c3c12efe0bc004053fea6b49c2f34e51e1ffc32f 100644
--- a/arch/sparc/kernel/signal_32.c
+++ b/arch/sparc/kernel/signal_32.c
@@ -60,10 +60,22 @@ struct rt_signal_frame {
 #define SF_ALIGNEDSZ  (((sizeof(struct signal_frame) + 7) & (~7)))
 #define RT_ALIGNEDSZ  (((sizeof(struct rt_signal_frame) + 7) & (~7)))
 
+/* Checks if the fp is valid.  We always build signal frames which are
+ * 16-byte aligned, therefore we can always enforce that the restore
+ * frame has that property as well.
+ */
+static inline bool invalid_frame_pointer(void __user *fp, int fplen)
+{
+	if ((((unsigned long) fp) & 15) || !__access_ok((unsigned long)fp, fplen))
+		return true;
+
+	return false;
+}
+
 asmlinkage void do_sigreturn(struct pt_regs *regs)
 {
+	unsigned long up_psr, pc, npc, ufp;
 	struct signal_frame __user *sf;
-	unsigned long up_psr, pc, npc;
 	sigset_t set;
 	__siginfo_fpu_t __user *fpu_save;
 	__siginfo_rwin_t __user *rwin_save;
@@ -77,10 +89,13 @@ asmlinkage void do_sigreturn(struct pt_regs *regs)
 	sf = (struct signal_frame __user *) regs->u_regs[UREG_FP];
 
 	/* 1. Make sure we are not getting garbage from the user */
-	if (!access_ok(VERIFY_READ, sf, sizeof(*sf)))
+	if (!invalid_frame_pointer(sf, sizeof(*sf)))
+		goto segv_and_exit;
+
+	if (get_user(ufp, &sf->info.si_regs.u_regs[UREG_FP]))
 		goto segv_and_exit;
 
-	if (((unsigned long) sf) & 3)
+	if (ufp & 0x7)
 		goto segv_and_exit;
 
 	err = __get_user(pc,  &sf->info.si_regs.pc);
@@ -127,7 +142,7 @@ asmlinkage void do_sigreturn(struct pt_regs *regs)
 asmlinkage void do_rt_sigreturn(struct pt_regs *regs)
 {
 	struct rt_signal_frame __user *sf;
-	unsigned int psr, pc, npc;
+	unsigned int psr, pc, npc, ufp;
 	__siginfo_fpu_t __user *fpu_save;
 	__siginfo_rwin_t __user *rwin_save;
 	sigset_t set;
@@ -135,8 +150,13 @@ asmlinkage void do_rt_sigreturn(struct pt_regs *regs)
 
 	synchronize_user_stack();
 	sf = (struct rt_signal_frame __user *) regs->u_regs[UREG_FP];
-	if (!access_ok(VERIFY_READ, sf, sizeof(*sf)) ||
-	    (((unsigned long) sf) & 0x03))
+	if (!invalid_frame_pointer(sf, sizeof(*sf)))
+		goto segv;
+
+	if (get_user(ufp, &sf->regs.u_regs[UREG_FP]))
+		goto segv;
+
+	if (ufp & 0x7)
 		goto segv;
 
 	err = __get_user(pc, &sf->regs.pc);
@@ -178,15 +198,6 @@ asmlinkage void do_rt_sigreturn(struct pt_regs *regs)
 	force_sig(SIGSEGV, current);
 }
 
-/* Checks if the fp is valid */
-static inline int invalid_frame_pointer(void __user *fp, int fplen)
-{
-	if ((((unsigned long) fp) & 7) || !__access_ok((unsigned long)fp, fplen))
-		return 1;
-
-	return 0;
-}
-
 static inline void __user *get_sigframe(struct ksignal *ksig, struct pt_regs *regs, unsigned long framesize)
 {
 	unsigned long sp = regs->u_regs[UREG_FP];
diff --git a/arch/sparc/kernel/signal_64.c b/arch/sparc/kernel/signal_64.c
index 39aaec173f66ebce88fe7953c99b412f5b07cc3d..5ee930c48f4c952953a9ea584f08915901f3c83a 100644
--- a/arch/sparc/kernel/signal_64.c
+++ b/arch/sparc/kernel/signal_64.c
@@ -234,6 +234,17 @@ asmlinkage void sparc64_get_context(struct pt_regs *regs)
 	goto out;
 }
 
+/* Checks if the fp is valid.  We always build rt signal frames which
+ * are 16-byte aligned, therefore we can always enforce that the
+ * restore frame has that property as well.
+ */
+static bool invalid_frame_pointer(void __user *fp)
+{
+	if (((unsigned long) fp) & 15)
+		return true;
+	return false;
+}
+
 struct rt_signal_frame {
 	struct sparc_stackf	ss;
 	siginfo_t		info;
@@ -246,8 +257,8 @@ struct rt_signal_frame {
 
 void do_rt_sigreturn(struct pt_regs *regs)
 {
+	unsigned long tpc, tnpc, tstate, ufp;
 	struct rt_signal_frame __user *sf;
-	unsigned long tpc, tnpc, tstate;
 	__siginfo_fpu_t __user *fpu_save;
 	__siginfo_rwin_t __user *rwin_save;
 	sigset_t set;
@@ -261,10 +272,16 @@ void do_rt_sigreturn(struct pt_regs *regs)
 		(regs->u_regs [UREG_FP] + STACK_BIAS);
 
 	/* 1. Make sure we are not getting garbage from the user */
-	if (((unsigned long) sf) & 3)
+	if (invalid_frame_pointer(sf))
+		goto segv;
+
+	if (get_user(ufp, &sf->regs.u_regs[UREG_FP]))
 		goto segv;
 
-	err = get_user(tpc, &sf->regs.tpc);
+	if ((ufp + STACK_BIAS) & 0x7)
+		goto segv;
+
+	err = __get_user(tpc, &sf->regs.tpc);
 	err |= __get_user(tnpc, &sf->regs.tnpc);
 	if (test_thread_flag(TIF_32BIT)) {
 		tpc &= 0xffffffff;
@@ -308,14 +325,6 @@ void do_rt_sigreturn(struct pt_regs *regs)
 	force_sig(SIGSEGV, current);
 }
 
-/* Checks if the fp is valid */
-static int invalid_frame_pointer(void __user *fp)
-{
-	if (((unsigned long) fp) & 15)
-		return 1;
-	return 0;
-}
-
 static inline void __user *get_sigframe(struct ksignal *ksig, struct pt_regs *regs, unsigned long framesize)
 {
 	unsigned long sp = regs->u_regs[UREG_FP] + STACK_BIAS;
diff --git a/arch/sparc/kernel/sigutil_32.c b/arch/sparc/kernel/sigutil_32.c
index 0f6eebe71e6c91560585580a31621f0fb59a6f81..e5fe8cef9a699a0ee645b879abd3a5bbc6fc712b 100644
--- a/arch/sparc/kernel/sigutil_32.c
+++ b/arch/sparc/kernel/sigutil_32.c
@@ -48,6 +48,10 @@ int save_fpu_state(struct pt_regs *regs, __siginfo_fpu_t __user *fpu)
 int restore_fpu_state(struct pt_regs *regs, __siginfo_fpu_t __user *fpu)
 {
 	int err;
+
+	if (((unsigned long) fpu) & 3)
+		return -EFAULT;
+
 #ifdef CONFIG_SMP
 	if (test_tsk_thread_flag(current, TIF_USEDFPU))
 		regs->psr &= ~PSR_EF;
@@ -97,7 +101,10 @@ int restore_rwin_state(__siginfo_rwin_t __user *rp)
 	struct thread_info *t = current_thread_info();
 	int i, wsaved, err;
 
-	__get_user(wsaved, &rp->wsaved);
+	if (((unsigned long) rp) & 3)
+		return -EFAULT;
+
+	get_user(wsaved, &rp->wsaved);
 	if (wsaved > NSWINS)
 		return -EFAULT;
 
diff --git a/arch/sparc/kernel/sigutil_64.c b/arch/sparc/kernel/sigutil_64.c
index 387834a9c56a5143a0a28b4e49018e43f4d95212..36aadcbeac694667b78199d2dc37607e09b63233 100644
--- a/arch/sparc/kernel/sigutil_64.c
+++ b/arch/sparc/kernel/sigutil_64.c
@@ -37,7 +37,10 @@ int restore_fpu_state(struct pt_regs *regs, __siginfo_fpu_t __user *fpu)
 	unsigned long fprs;
 	int err;
 
-	err = __get_user(fprs, &fpu->si_fprs);
+	if (((unsigned long) fpu) & 7)
+		return -EFAULT;
+
+	err = get_user(fprs, &fpu->si_fprs);
 	fprs_write(0);
 	regs->tstate &= ~TSTATE_PEF;
 	if (fprs & FPRS_DL)
@@ -72,7 +75,10 @@ int restore_rwin_state(__siginfo_rwin_t __user *rp)
 	struct thread_info *t = current_thread_info();
 	int i, wsaved, err;
 
-	__get_user(wsaved, &rp->wsaved);
+	if (((unsigned long) rp) & 7)
+		return -EFAULT;
+
+	get_user(wsaved, &rp->wsaved);
 	if (wsaved > NSWINS)
 		return -EFAULT;
 
diff --git a/arch/sparc/kernel/urtt_fill.S b/arch/sparc/kernel/urtt_fill.S
new file mode 100644
index 0000000000000000000000000000000000000000..5604a2b051d46bb822873a31b9b1f88a85253562
--- /dev/null
+++ b/arch/sparc/kernel/urtt_fill.S
@@ -0,0 +1,98 @@
+#include <asm/thread_info.h>
+#include <asm/trap_block.h>
+#include <asm/spitfire.h>
+#include <asm/ptrace.h>
+#include <asm/head.h>
+
+		.text
+		.align	8
+		.globl	user_rtt_fill_fixup_common
+user_rtt_fill_fixup_common:
+		rdpr	%cwp, %g1
+		add	%g1, 1, %g1
+		wrpr	%g1, 0x0, %cwp
+
+		rdpr	%wstate, %g2
+		sll	%g2, 3, %g2
+		wrpr	%g2, 0x0, %wstate
+
+		/* We know %canrestore and %otherwin are both zero.  */
+
+		sethi	%hi(sparc64_kern_pri_context), %g2
+		ldx	[%g2 + %lo(sparc64_kern_pri_context)], %g2
+		mov	PRIMARY_CONTEXT, %g1
+
+661:		stxa	%g2, [%g1] ASI_DMMU
+		.section .sun4v_1insn_patch, "ax"
+		.word	661b
+		stxa	%g2, [%g1] ASI_MMU
+		.previous
+
+		sethi	%hi(KERNBASE), %g1
+		flush	%g1
+
+		mov	%g4, %l4
+		mov	%g5, %l5
+		brnz,pn	%g3, 1f
+		 mov	%g3, %l3
+
+		or	%g4, FAULT_CODE_WINFIXUP, %g4
+		stb	%g4, [%g6 + TI_FAULT_CODE]
+		stx	%g5, [%g6 + TI_FAULT_ADDR]
+1:
+		mov	%g6, %l1
+		wrpr	%g0, 0x0, %tl
+
+661:		nop
+		.section		.sun4v_1insn_patch, "ax"
+		.word			661b
+		SET_GL(0)
+		.previous
+
+		wrpr	%g0, RTRAP_PSTATE, %pstate
+
+		mov	%l1, %g6
+		ldx	[%g6 + TI_TASK], %g4
+		LOAD_PER_CPU_BASE(%g5, %g6, %g1, %g2, %g3)
+
+		brnz,pn	%l3, 1f
+		 nop
+
+		call	do_sparc64_fault
+		 add	%sp, PTREGS_OFF, %o0
+		ba,pt	%xcc, rtrap
+		 nop
+
+1:		cmp	%g3, 2
+		bne,pn	%xcc, 2f
+		 nop
+
+		sethi	%hi(tlb_type), %g1
+		lduw	[%g1 + %lo(tlb_type)], %g1
+		cmp	%g1, 3
+		bne,pt	%icc, 1f
+		 add	%sp, PTREGS_OFF, %o0
+		mov	%l4, %o2
+		call	sun4v_do_mna
+		 mov	%l5, %o1
+		ba,a,pt	%xcc, rtrap
+1:		mov	%l4, %o1
+		mov	%l5, %o2
+		call	mem_address_unaligned
+		 nop
+		ba,a,pt	%xcc, rtrap
+
+2:		sethi	%hi(tlb_type), %g1
+		mov	%l4, %o1
+		lduw	[%g1 + %lo(tlb_type)], %g1
+		mov	%l5, %o2
+		cmp	%g1, 3
+		bne,pt	%icc, 1f
+		 add	%sp, PTREGS_OFF, %o0
+		call	sun4v_data_access_exception
+		 nop
+		ba,a,pt	%xcc, rtrap
+
+1:		call	spitfire_data_access_exception
+		 nop
+		ba,a,pt	%xcc, rtrap
diff --git a/arch/sparc/mm/init_64.c b/arch/sparc/mm/init_64.c
index 652683cb4b4bcc53f663dd424ed81249c7130e1f..14bb0d5ed3c605cc71680d95102c832295abbbb8 100644
--- a/arch/sparc/mm/init_64.c
+++ b/arch/sparc/mm/init_64.c
@@ -2824,9 +2824,10 @@ void hugetlb_setup(struct pt_regs *regs)
 	 * the Data-TLB for huge pages.
 	 */
 	if (tlb_type == cheetah_plus) {
+		bool need_context_reload = false;
 		unsigned long ctx;
 
-		spin_lock(&ctx_alloc_lock);
+		spin_lock_irq(&ctx_alloc_lock);
 		ctx = mm->context.sparc64_ctx_val;
 		ctx &= ~CTX_PGSZ_MASK;
 		ctx |= CTX_PGSZ_BASE << CTX_PGSZ0_SHIFT;
@@ -2845,9 +2846,12 @@ void hugetlb_setup(struct pt_regs *regs)
 			 * also executing in this address space.
 			 */
 			mm->context.sparc64_ctx_val = ctx;
-			on_each_cpu(context_reload, mm, 0);
+			need_context_reload = true;
 		}
-		spin_unlock(&ctx_alloc_lock);
+		spin_unlock_irq(&ctx_alloc_lock);
+
+		if (need_context_reload)
+			on_each_cpu(context_reload, mm, 0);
 	}
 }
 #endif