Merge branch 'master' into sh/hw-breakpoints

Conflict between FPU thread flag migration and debug
thread flag addition.

Conflicts:
	arch/sh/include/asm/thread_info.h
	arch/sh/include/asm/ubc.h
	arch/sh/kernel/process_32.c

19 files changed, 1202 insertions, 375 deletions

diff --git a/arch/sh/kernel/cpu/Makefile b/arch/sh/kernel/cpu/Makefile
index 3d6b9312dc47..d97c803719ec 100644
--- a/arch/sh/kernel/cpu/Makefile
+++ b/arch/sh/kernel/cpu/Makefile
@@ -15,7 +15,6 @@ obj-$(CONFIG_ARCH_SHMOBILE)	+= shmobile/
 
 # Common interfaces.
 
-obj-$(CONFIG_UBC_WAKEUP)	+= ubc.o
 obj-$(CONFIG_SH_ADC)		+= adc.o
 obj-$(CONFIG_SH_CLK_CPG)	+= clock-cpg.o
 
diff --git a/arch/sh/kernel/cpu/init.c b/arch/sh/kernel/cpu/init.c
index e932ebef4738..89b4b76c0d76 100644
--- a/arch/sh/kernel/cpu/init.c
+++ b/arch/sh/kernel/cpu/init.c
@@ -75,16 +75,11 @@ static void __init expmask_init(void)
 	/*
 	 * Future proofing.
 	 *
-	 * Disable support for slottable sleep instruction
-	 * and non-nop instructions in the rte delay slot.
+	 * Disable support for slottable sleep instruction, non-nop
+	 * instructions in the rte delay slot, and associative writes to
+	 * the memory-mapped cache array.
 	 */
-	expmask &= ~(EXPMASK_RTEDS | EXPMASK_BRDSSLP);
-
-	/*
-	 * Enable associative writes to the memory-mapped cache array
-	 * until the cache flush ops have been rewritten.
-	 */
-	expmask |= EXPMASK_MMCAW;
+	expmask &= ~(EXPMASK_RTEDS | EXPMASK_BRDSSLP | EXPMASK_MMCAW);
 
 	__raw_writel(expmask, EXPMASK);
 	ctrl_barrier();
@@ -311,12 +306,12 @@ asmlinkage void __init sh_cpu_init(void)
 	if (fpu_disabled) {
 		printk("FPU Disabled\n");
 		current_cpu_data.flags &= ~CPU_HAS_FPU;
-		disable_fpu();
 	}
 
 	/* FPU initialization */
+	disable_fpu();
 	if ((current_cpu_data.flags & CPU_HAS_FPU)) {
-		clear_thread_flag(TIF_USEDFPU);
+		current_thread_info()->status &= ~TS_USEDFPU;
 		clear_used_math();
 	}
 
@@ -338,17 +333,6 @@ asmlinkage void __init sh_cpu_init(void)
 	}
 #endif
 
-	/*
-	 * Some brain-damaged loaders decided it would be a good idea to put
-	 * the UBC to sleep. This causes some issues when it comes to things
-	 * like PTRACE_SINGLESTEP or doing hardware watchpoints in GDB.  So ..
-	 * we wake it up and hope that all is well.
-	 */
-#ifdef CONFIG_SUPERH32
-	if (raw_smp_processor_id() == 0)
-		ubc_wakeup();
-#endif
-
 	speculative_execution_init();
 	expmask_init();
 }
diff --git a/arch/sh/kernel/cpu/sh2a/fpu.c b/arch/sh/kernel/cpu/sh2a/fpu.c
index 6df2fb98eb30..d395ce5740e7 100644
--- a/arch/sh/kernel/cpu/sh2a/fpu.c
+++ b/arch/sh/kernel/cpu/sh2a/fpu.c
@@ -25,14 +25,12 @@
 
 /*
  * Save FPU registers onto task structure.
- * Assume called with FPU enabled (SR.FD=0).
  */
 void
-save_fpu(struct task_struct *tsk, struct pt_regs *regs)
+save_fpu(struct task_struct *tsk)
 {
 	unsigned long dummy;
 
-	clear_tsk_thread_flag(tsk, TIF_USEDFPU);
 	enable_fpu();
 	asm volatile("sts.l	fpul, @-%0\n\t"
 		     "sts.l	fpscr, @-%0\n\t"
@@ -60,7 +58,6 @@ save_fpu(struct task_struct *tsk, struct pt_regs *regs)
 		     : "memory");
 
 	disable_fpu();
-	release_fpu(regs);
 }
 
 static void
@@ -598,31 +595,31 @@ BUILD_TRAP_HANDLER(fpu_error)
 	struct task_struct *tsk = current;
 	TRAP_HANDLER_DECL;
 
-	save_fpu(tsk, regs);
+	__unlazy_fpu(tsk, regs);
 	if (ieee_fpe_handler(regs)) {
 		tsk->thread.fpu.hard.fpscr &=
 			~(FPSCR_CAUSE_MASK | FPSCR_FLAG_MASK);
 		grab_fpu(regs);
 		restore_fpu(tsk);
-		set_tsk_thread_flag(tsk, TIF_USEDFPU);
+		task_thread_info(tsk)->status |= TS_USEDFPU;
 		return;
 	}
 
 	force_sig(SIGFPE, tsk);
 }
 
-BUILD_TRAP_HANDLER(fpu_state_restore)
+void fpu_state_restore(struct pt_regs *regs)
 {
 	struct task_struct *tsk = current;
-	TRAP_HANDLER_DECL;
 
 	grab_fpu(regs);
-	if (!user_mode(regs)) {
+	if (unlikely(!user_mode(regs))) {
 		printk(KERN_ERR "BUG: FPU is used in kernel mode.\n");
+		BUG();
 		return;
 	}
 
-	if (used_math()) {
+	if (likely(used_math())) {
 		/* Using the FPU again.  */
 		restore_fpu(tsk);
 	} else	{
@@ -630,5 +627,13 @@ BUILD_TRAP_HANDLER(fpu_state_restore)
 		fpu_init();
 		set_used_math();
 	}
-	set_tsk_thread_flag(tsk, TIF_USEDFPU);
+	task_thread_info(tsk)->status |= TS_USEDFPU;
+	tsk->fpu_counter++;
+}
+
+BUILD_TRAP_HANDLER(fpu_state_restore)
+{
+	TRAP_HANDLER_DECL;
+
+	fpu_state_restore(regs);
 }
diff --git a/arch/sh/kernel/cpu/sh3/entry.S b/arch/sh/kernel/cpu/sh3/entry.S
index bb407ef0b91e..3f7e2a22c7c2 100644
--- a/arch/sh/kernel/cpu/sh3/entry.S
+++ b/arch/sh/kernel/cpu/sh3/entry.S
@@ -297,41 +297,8 @@ ENTRY(vbr_base)
 !
 	.balign 	256,0,256
 general_exception:
-#ifndef CONFIG_CPU_SUBTYPE_SHX3
 	bra	handle_exception
 	 sts	pr, k3		! save original pr value in k3
-#else
-	mov.l	1f, k4
-	mov.l	@k4, k4
-
-	! Is EXPEVT larger than 0x800?
-	mov	#0x8, k0
-	shll8	k0
-	cmp/hs	k0, k4
-	bf	0f
-
-	! then add 0x580 (k2 is 0xd80 or 0xda0)
-	mov	#0x58, k0
-	shll2	k0
-	shll2	k0
-	add	k0, k4
-0:
-	! Setup stack and save DSP context (k0 contains original r15 on return)
-	bsr	prepare_stack
-	 nop
-
-	! Save registers / Switch to bank 0
-	mov		k4, k2		! keep vector in k2
-	mov.l	1f, k4		! SR bits to clear in k4
-	bsr	save_regs	! needs original pr value in k3
-	 nop
-
-	bra	handle_exception_special
-	 nop
-
-	.align	2
-1:	.long	EXPEVT
-#endif
 
 ! prepare_stack()
 ! - roll back gRB
diff --git a/arch/sh/kernel/cpu/sh4/Makefile b/arch/sh/kernel/cpu/sh4/Makefile
index 203b18347b83..3a1dbc709831 100644
--- a/arch/sh/kernel/cpu/sh4/Makefile
+++ b/arch/sh/kernel/cpu/sh4/Makefile
@@ -9,6 +9,11 @@ obj-$(CONFIG_HIBERNATION)		+= $(addprefix ../sh3/, swsusp.o)
 obj-$(CONFIG_SH_FPU)			+= fpu.o softfloat.o
 obj-$(CONFIG_SH_STORE_QUEUES)		+= sq.o
 
+# Perf events
+perf-$(CONFIG_CPU_SUBTYPE_SH7750)	:= perf_event.o
+perf-$(CONFIG_CPU_SUBTYPE_SH7750S)	:= perf_event.o
+perf-$(CONFIG_CPU_SUBTYPE_SH7091)	:= perf_event.o
+
 # CPU subtype setup
 obj-$(CONFIG_CPU_SUBTYPE_SH7750)	+= setup-sh7750.o
 obj-$(CONFIG_CPU_SUBTYPE_SH7750R)	+= setup-sh7750.o
@@ -27,4 +32,5 @@ endif
 # Additional clocks by subtype
 clock-$(CONFIG_CPU_SUBTYPE_SH4_202)	+= clock-sh4-202.o
 
-obj-y	+= $(clock-y)
+obj-y					+= $(clock-y)
+obj-$(CONFIG_PERF_EVENTS)		+= $(perf-y)
diff --git a/arch/sh/kernel/cpu/sh4/fpu.c b/arch/sh/kernel/cpu/sh4/fpu.c
index e3ea5411da6d..e97857aec8a0 100644
--- a/arch/sh/kernel/cpu/sh4/fpu.c
+++ b/arch/sh/kernel/cpu/sh4/fpu.c
@@ -41,13 +41,11 @@ static unsigned int fpu_exception_flags;
 
 /*
  * Save FPU registers onto task structure.
- * Assume called with FPU enabled (SR.FD=0).
  */
-void save_fpu(struct task_struct *tsk, struct pt_regs *regs)
+void save_fpu(struct task_struct *tsk)
 {
 	unsigned long dummy;
 
-	clear_tsk_thread_flag(tsk, TIF_USEDFPU);
 	enable_fpu();
 	asm volatile ("sts.l	fpul, @-%0\n\t"
 		      "sts.l	fpscr, @-%0\n\t"
@@ -92,7 +90,6 @@ void save_fpu(struct task_struct *tsk, struct pt_regs *regs)
 		      :"memory");
 
 	disable_fpu();
-	release_fpu(regs);
 }
 
 static void restore_fpu(struct task_struct *tsk)
@@ -285,7 +282,6 @@ static int ieee_fpe_handler(struct pt_regs *regs)
 		/* fcnvsd */
 		struct task_struct *tsk = current;
 
-		save_fpu(tsk, regs);
 		if ((tsk->thread.fpu.hard.fpscr & FPSCR_CAUSE_ERROR))
 			/* FPU error */
 			denormal_to_double(&tsk->thread.fpu.hard,
@@ -462,7 +458,7 @@ BUILD_TRAP_HANDLER(fpu_error)
 	struct task_struct *tsk = current;
 	TRAP_HANDLER_DECL;
 
-	save_fpu(tsk, regs);
+	__unlazy_fpu(tsk, regs);
 	fpu_exception_flags = 0;
 	if (ieee_fpe_handler(regs)) {
 		tsk->thread.fpu.hard.fpscr &=
@@ -473,7 +469,7 @@ BUILD_TRAP_HANDLER(fpu_error)
 		tsk->thread.fpu.hard.fpscr |= (fpu_exception_flags >> 10);
 		grab_fpu(regs);
 		restore_fpu(tsk);
-		set_tsk_thread_flag(tsk, TIF_USEDFPU);
+		task_thread_info(tsk)->status |= TS_USEDFPU;
 		if ((((tsk->thread.fpu.hard.fpscr & FPSCR_ENABLE_MASK) >> 7) &
 		     (fpu_exception_flags >> 2)) == 0) {
 			return;
@@ -483,18 +479,18 @@ BUILD_TRAP_HANDLER(fpu_error)
 	force_sig(SIGFPE, tsk);
 }
 
-BUILD_TRAP_HANDLER(fpu_state_restore)
+void fpu_state_restore(struct pt_regs *regs)
 {
 	struct task_struct *tsk = current;
-	TRAP_HANDLER_DECL;
 
 	grab_fpu(regs);
-	if (!user_mode(regs)) {
+	if (unlikely(!user_mode(regs))) {
 		printk(KERN_ERR "BUG: FPU is used in kernel mode.\n");
+		BUG();
 		return;
 	}
 
-	if (used_math()) {
+	if (likely(used_math())) {
 		/* Using the FPU again.  */
 		restore_fpu(tsk);
 	} else {
@@ -502,5 +498,13 @@ BUILD_TRAP_HANDLER(fpu_state_restore)
 		fpu_init();
 		set_used_math();
 	}
-	set_tsk_thread_flag(tsk, TIF_USEDFPU);
+	task_thread_info(tsk)->status |= TS_USEDFPU;
+	tsk->fpu_counter++;
+}
+
+BUILD_TRAP_HANDLER(fpu_state_restore)
+{
+	TRAP_HANDLER_DECL;
+
+	fpu_state_restore(regs);
 }
diff --git a/arch/sh/kernel/cpu/sh4/perf_event.c b/arch/sh/kernel/cpu/sh4/perf_event.c
new file mode 100644
index 000000000000..7f9ecc9c2d02
--- /dev/null
+++ b/arch/sh/kernel/cpu/sh4/perf_event.c
@@ -0,0 +1,253 @@
+/*
+ * Performance events support for SH7750-style performance counters
+ *
+ *  Copyright (C) 2009  Paul Mundt
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ */
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/io.h>
+#include <linux/irq.h>
+#include <linux/perf_event.h>
+#include <asm/processor.h>
+
+#define PM_CR_BASE	0xff000084	/* 16-bit */
+#define PM_CTR_BASE	0xff100004	/* 32-bit */
+
+#define PMCR(n)		(PM_CR_BASE + ((n) * 0x04))
+#define PMCTRH(n)	(PM_CTR_BASE + 0x00 + ((n) * 0x08))
+#define PMCTRL(n)	(PM_CTR_BASE + 0x04 + ((n) * 0x08))
+
+#define PMCR_PMM_MASK	0x0000003f
+
+#define PMCR_CLKF	0x00000100
+#define PMCR_PMCLR	0x00002000
+#define PMCR_PMST	0x00004000
+#define PMCR_PMEN	0x00008000
+
+static struct sh_pmu sh7750_pmu;
+
+/*
+ * There are a number of events supported by each counter (33 in total).
+ * Since we have 2 counters, each counter will take the event code as it
+ * corresponds to the PMCR PMM setting. Each counter can be configured
+ * independently.
+ *
+ *	Event Code	Description
+ *	----------	-----------
+ *
+ *	0x01		Operand read access
+ *	0x02		Operand write access
+ *	0x03		UTLB miss
+ *	0x04		Operand cache read miss
+ *	0x05		Operand cache write miss
+ *	0x06		Instruction fetch (w/ cache)
+ *	0x07		Instruction TLB miss
+ *	0x08		Instruction cache miss
+ *	0x09		All operand accesses
+ *	0x0a		All instruction accesses
+ *	0x0b		OC RAM operand access
+ *	0x0d		On-chip I/O space access
+ *	0x0e		Operand access (r/w)
+ *	0x0f		Operand cache miss (r/w)
+ *	0x10		Branch instruction
+ *	0x11		Branch taken
+ *	0x12		BSR/BSRF/JSR
+ *	0x13		Instruction execution
+ *	0x14		Instruction execution in parallel
+ *	0x15		FPU Instruction execution
+ *	0x16		Interrupt
+ *	0x17		NMI
+ *	0x18		trapa instruction execution
+ *	0x19		UBCA match
+ *	0x1a		UBCB match
+ *	0x21		Instruction cache fill
+ *	0x22		Operand cache fill
+ *	0x23		Elapsed time
+ *	0x24		Pipeline freeze by I-cache miss
+ *	0x25		Pipeline freeze by D-cache miss
+ *	0x27		Pipeline freeze by branch instruction
+ *	0x28		Pipeline freeze by CPU register
+ *	0x29		Pipeline freeze by FPU
+ */
+
+static const int sh7750_general_events[] = {
+	[PERF_COUNT_HW_CPU_CYCLES]		= 0x0023,
+	[PERF_COUNT_HW_INSTRUCTIONS]		= 0x000a,
+	[PERF_COUNT_HW_CACHE_REFERENCES]	= 0x0006,	/* I-cache */
+	[PERF_COUNT_HW_CACHE_MISSES]		= 0x0008,	/* I-cache */
+	[PERF_COUNT_HW_BRANCH_INSTRUCTIONS]	= 0x0010,
+	[PERF_COUNT_HW_BRANCH_MISSES]		= -1,
+	[PERF_COUNT_HW_BUS_CYCLES]		= -1,
+};
+
+#define C(x)	PERF_COUNT_HW_CACHE_##x
+
+static const int sh7750_cache_events
+			[PERF_COUNT_HW_CACHE_MAX]
+			[PERF_COUNT_HW_CACHE_OP_MAX]
+			[PERF_COUNT_HW_CACHE_RESULT_MAX] =
+{
+	[ C(L1D) ] = {
+		[ C(OP_READ) ] = {
+			[ C(RESULT_ACCESS) ] = 0x0001,
+			[ C(RESULT_MISS)   ] = 0x0004,
+		},
+		[ C(OP_WRITE) ] = {
+			[ C(RESULT_ACCESS) ] = 0x0002,
+			[ C(RESULT_MISS)   ] = 0x0005,
+		},
+		[ C(OP_PREFETCH) ] = {
+			[ C(RESULT_ACCESS) ] = 0,
+			[ C(RESULT_MISS)   ] = 0,
+		},
+	},
+
+	[ C(L1I) ] = {
+		[ C(OP_READ) ] = {
+			[ C(RESULT_ACCESS) ] = 0x0006,
+			[ C(RESULT_MISS)   ] = 0x0008,
+		},
+		[ C(OP_WRITE) ] = {
+			[ C(RESULT_ACCESS) ] = -1,
+			[ C(RESULT_MISS)   ] = -1,
+		},
+		[ C(OP_PREFETCH) ] = {
+			[ C(RESULT_ACCESS) ] = 0,
+			[ C(RESULT_MISS)   ] = 0,
+		},
+	},
+
+	[ C(LL) ] = {
+		[ C(OP_READ) ] = {
+			[ C(RESULT_ACCESS) ] = 0,
+			[ C(RESULT_MISS)   ] = 0,
+		},
+		[ C(OP_WRITE) ] = {
+			[ C(RESULT_ACCESS) ] = 0,
+			[ C(RESULT_MISS)   ] = 0,
+		},
+		[ C(OP_PREFETCH) ] = {
+			[ C(RESULT_ACCESS) ] = 0,
+			[ C(RESULT_MISS)   ] = 0,
+		},
+	},
+
+	[ C(DTLB) ] = {
+		[ C(OP_READ) ] = {
+			[ C(RESULT_ACCESS) ] = 0,
+			[ C(RESULT_MISS)   ] = 0x0003,
+		},
+		[ C(OP_WRITE) ] = {
+			[ C(RESULT_ACCESS) ] = 0,
+			[ C(RESULT_MISS)   ] = 0,
+		},
+		[ C(OP_PREFETCH) ] = {
+			[ C(RESULT_ACCESS) ] = 0,
+			[ C(RESULT_MISS)   ] = 0,
+		},
+	},
+
+	[ C(ITLB) ] = {
+		[ C(OP_READ) ] = {
+			[ C(RESULT_ACCESS) ] = 0,
+			[ C(RESULT_MISS)   ] = 0x0007,
+		},
+		[ C(OP_WRITE) ] = {
+			[ C(RESULT_ACCESS) ] = -1,
+			[ C(RESULT_MISS)   ] = -1,
+		},
+		[ C(OP_PREFETCH) ] = {
+			[ C(RESULT_ACCESS) ] = -1,
+			[ C(RESULT_MISS)   ] = -1,
+		},
+	},
+
+	[ C(BPU) ] = {
+		[ C(OP_READ) ] = {
+			[ C(RESULT_ACCESS) ] = -1,
+			[ C(RESULT_MISS)   ] = -1,
+		},
+		[ C(OP_WRITE) ] = {
+			[ C(RESULT_ACCESS) ] = -1,
+			[ C(RESULT_MISS)   ] = -1,
+		},
+		[ C(OP_PREFETCH) ] = {
+			[ C(RESULT_ACCESS) ] = -1,
+			[ C(RESULT_MISS)   ] = -1,
+		},
+	},
+};
+
+static int sh7750_event_map(int event)
+{
+	return sh7750_general_events[event];
+}
+
+static u64 sh7750_pmu_read(int idx)
+{
+	return (u64)((u64)(__raw_readl(PMCTRH(idx)) & 0xffff) << 32) |
+			   __raw_readl(PMCTRL(idx));
+}
+
+static void sh7750_pmu_disable(struct hw_perf_event *hwc, int idx)
+{
+	unsigned int tmp;
+
+	tmp = __raw_readw(PMCR(idx));
+	tmp &= ~(PMCR_PMM_MASK | PMCR_PMEN);
+	__raw_writew(tmp, PMCR(idx));
+}
+
+static void sh7750_pmu_enable(struct hw_perf_event *hwc, int idx)
+{
+	__raw_writew(__raw_readw(PMCR(idx)) | PMCR_PMCLR, PMCR(idx));
+	__raw_writew(hwc->config | PMCR_PMEN | PMCR_PMST, PMCR(idx));
+}
+
+static void sh7750_pmu_disable_all(void)
+{
+	int i;
+
+	for (i = 0; i < sh7750_pmu.num_events; i++)
+		__raw_writew(__raw_readw(PMCR(i)) & ~PMCR_PMEN, PMCR(i));
+}
+
+static void sh7750_pmu_enable_all(void)
+{
+	int i;
+
+	for (i = 0; i < sh7750_pmu.num_events; i++)
+		__raw_writew(__raw_readw(PMCR(i)) | PMCR_PMEN, PMCR(i));
+}
+
+static struct sh_pmu sh7750_pmu = {
+	.name		= "SH7750",
+	.num_events	= 2,
+	.event_map	= sh7750_event_map,
+	.max_events	= ARRAY_SIZE(sh7750_general_events),
+	.raw_event_mask	= PMCR_PMM_MASK,
+	.cache_events	= &sh7750_cache_events,
+	.read		= sh7750_pmu_read,
+	.disable	= sh7750_pmu_disable,
+	.enable		= sh7750_pmu_enable,
+	.disable_all	= sh7750_pmu_disable_all,
+	.enable_all	= sh7750_pmu_enable_all,
+};
+
+static int __init sh7750_pmu_init(void)
+{
+	/*
+	 * Make sure this CPU actually has perf counters.
+	 */
+	if (!(boot_cpu_data.flags & CPU_HAS_PERF_COUNTER)) {
+		pr_notice("HW perf events unsupported, software events only.\n");
+		return -ENODEV;
+	}
+
+	return register_sh_pmu(&sh7750_pmu);
+}
+arch_initcall(sh7750_pmu_init);
diff --git a/arch/sh/kernel/cpu/sh4a/Makefile b/arch/sh/kernel/cpu/sh4a/Makefile
index 490d5dc9e372..33bab477d2e2 100644
--- a/arch/sh/kernel/cpu/sh4a/Makefile
+++ b/arch/sh/kernel/cpu/sh4a/Makefile
@@ -44,3 +44,4 @@ pinmux-$(CONFIG_CPU_SUBTYPE_SH7786)	:= pinmux-sh7786.o
 obj-y				+= $(clock-y)
 obj-$(CONFIG_SMP)		+= $(smp-y)
 obj-$(CONFIG_GENERIC_GPIO)	+= $(pinmux-y)
+obj-$(CONFIG_PERF_EVENTS)	+= perf_event.o
diff --git a/arch/sh/kernel/cpu/sh4a/clock-sh7724.c b/arch/sh/kernel/cpu/sh4a/clock-sh7724.c
index dfe9192be63e..9db743802f06 100644
--- a/arch/sh/kernel/cpu/sh4a/clock-sh7724.c
+++ b/arch/sh/kernel/cpu/sh4a/clock-sh7724.c
@@ -152,7 +152,7 @@ struct clk div6_clks[] = {
 	SH_CLK_DIV6("fsia_clk", &div3_clk, FCLKACR, 0),
 	SH_CLK_DIV6("fsib_clk", &div3_clk, FCLKBCR, 0),
 	SH_CLK_DIV6("irda_clk", &div3_clk, IRDACLKCR, 0),
-	SH_CLK_DIV6("spu_clk", &div3_clk, SPUCLKCR, 0),
+	SH_CLK_DIV6("spu_clk", &div3_clk, SPUCLKCR, CLK_ENABLE_ON_INIT),
 };
 
 #define R_CLK (&r_clk)
diff --git a/arch/sh/kernel/cpu/sh4a/perf_event.c b/arch/sh/kernel/cpu/sh4a/perf_event.c
new file mode 100644
index 000000000000..eddc21973fa1
--- /dev/null
+++ b/arch/sh/kernel/cpu/sh4a/perf_event.c
@@ -0,0 +1,269 @@
+/*
+ * Performance events support for SH-4A performance counters
+ *
+ *  Copyright (C) 2009  Paul Mundt
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ */
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/io.h>
+#include <linux/irq.h>
+#include <linux/perf_event.h>
+#include <asm/processor.h>
+
+#define PPC_CCBR(idx)	(0xff200800 + (sizeof(u32) * idx))
+#define PPC_PMCTR(idx)	(0xfc100000 + (sizeof(u32) * idx))
+
+#define CCBR_CIT_MASK	(0x7ff << 6)
+#define CCBR_DUC	(1 << 3)
+#define CCBR_CMDS	(1 << 1)
+#define CCBR_PPCE	(1 << 0)
+
+#define PPC_PMCAT	0xfc100080
+
+#define PMCAT_OVF3	(1 << 27)
+#define PMCAT_CNN3	(1 << 26)
+#define PMCAT_CLR3	(1 << 25)
+#define PMCAT_OVF2	(1 << 19)
+#define PMCAT_CLR2	(1 << 17)
+#define PMCAT_OVF1	(1 << 11)
+#define PMCAT_CNN1	(1 << 10)
+#define PMCAT_CLR1	(1 << 9)
+#define PMCAT_OVF0	(1 << 3)
+#define PMCAT_CLR0	(1 << 1)
+
+static struct sh_pmu sh4a_pmu;
+
+/*
+ * Supported raw event codes:
+ *
+ *	Event Code	Description
+ *	----------	-----------
+ *
+ *	0x0000		number of elapsed cycles
+ *	0x0200		number of elapsed cycles in privileged mode
+ *	0x0280		number of elapsed cycles while SR.BL is asserted
+ *	0x0202		instruction execution
+ *	0x0203		instruction execution in parallel
+ *	0x0204		number of unconditional branches
+ *	0x0208		number of exceptions
+ *	0x0209		number of interrupts
+ *	0x0220		UTLB miss caused by instruction fetch
+ *	0x0222		UTLB miss caused by operand access
+ *	0x02a0		number of ITLB misses
+ *	0x0028		number of accesses to instruction memories
+ *	0x0029		number of accesses to instruction cache
+ *	0x002a		instruction cache miss
+ *	0x022e		number of access to instruction X/Y memory
+ *	0x0030		number of reads to operand memories
+ *	0x0038		number of writes to operand memories
+ *	0x0031		number of operand cache read accesses
+ *	0x0039		number of operand cache write accesses
+ *	0x0032		operand cache read miss
+ *	0x003a		operand cache write miss
+ *	0x0236		number of reads to operand X/Y memory
+ *	0x023e		number of writes to operand X/Y memory
+ *	0x0237		number of reads to operand U memory
+ *	0x023f		number of writes to operand U memory
+ *	0x0337		number of U memory read buffer misses
+ *	0x02b4		number of wait cycles due to operand read access
+ *	0x02bc		number of wait cycles due to operand write access
+ *	0x0033		number of wait cycles due to operand cache read miss
+ *	0x003b		number of wait cycles due to operand cache write miss
+ */
+
+/*
+ * Special reserved bits used by hardware emulators, read values will
+ * vary, but writes must always be 0.
+ */
+#define PMCAT_EMU_CLR_MASK	((1 << 24) | (1 << 16) | (1 << 8) | (1 << 0))
+
+static const int sh4a_general_events[] = {
+	[PERF_COUNT_HW_CPU_CYCLES]		= 0x0000,
+	[PERF_COUNT_HW_INSTRUCTIONS]		= 0x0202,
+	[PERF_COUNT_HW_CACHE_REFERENCES]	= 0x0029,	/* I-cache */
+	[PERF_COUNT_HW_CACHE_MISSES]		= 0x002a,	/* I-cache */
+	[PERF_COUNT_HW_BRANCH_INSTRUCTIONS]	= 0x0204,
+	[PERF_COUNT_HW_BRANCH_MISSES]		= -1,
+	[PERF_COUNT_HW_BUS_CYCLES]		= -1,
+};
+
+#define C(x)	PERF_COUNT_HW_CACHE_##x
+
+static const int sh4a_cache_events
+			[PERF_COUNT_HW_CACHE_MAX]
+			[PERF_COUNT_HW_CACHE_OP_MAX]
+			[PERF_COUNT_HW_CACHE_RESULT_MAX] =
+{
+	[ C(L1D) ] = {
+		[ C(OP_READ) ] = {
+			[ C(RESULT_ACCESS) ] = 0x0031,
+			[ C(RESULT_MISS)   ] = 0x0032,
+		},
+		[ C(OP_WRITE) ] = {
+			[ C(RESULT_ACCESS) ] = 0x0039,
+			[ C(RESULT_MISS)   ] = 0x003a,
+		},
+		[ C(OP_PREFETCH) ] = {
+			[ C(RESULT_ACCESS) ] = 0,
+			[ C(RESULT_MISS)   ] = 0,
+		},
+	},
+
+	[ C(L1I) ] = {
+		[ C(OP_READ) ] = {
+			[ C(RESULT_ACCESS) ] = 0x0029,
+			[ C(RESULT_MISS)   ] = 0x002a,
+		},
+		[ C(OP_WRITE) ] = {
+			[ C(RESULT_ACCESS) ] = -1,
+			[ C(RESULT_MISS)   ] = -1,
+		},
+		[ C(OP_PREFETCH) ] = {
+			[ C(RESULT_ACCESS) ] = 0,
+			[ C(RESULT_MISS)   ] = 0,
+		},
+	},
+
+	[ C(LL) ] = {
+		[ C(OP_READ) ] = {
+			[ C(RESULT_ACCESS) ] = 0x0030,
+			[ C(RESULT_MISS)   ] = 0,
+		},
+		[ C(OP_WRITE) ] = {
+			[ C(RESULT_ACCESS) ] = 0x0038,
+			[ C(RESULT_MISS)   ] = 0,
+		},
+		[ C(OP_PREFETCH) ] = {
+			[ C(RESULT_ACCESS) ] = 0,
+			[ C(RESULT_MISS)   ] = 0,
+		},
+	},
+
+	[ C(DTLB) ] = {
+		[ C(OP_READ) ] = {
+			[ C(RESULT_ACCESS) ] = 0x0222,
+			[ C(RESULT_MISS)   ] = 0x0220,
+		},
+		[ C(OP_WRITE) ] = {
+			[ C(RESULT_ACCESS) ] = 0,
+			[ C(RESULT_MISS)   ] = 0,
+		},
+		[ C(OP_PREFETCH) ] = {
+			[ C(RESULT_ACCESS) ] = 0,
+			[ C(RESULT_MISS)   ] = 0,
+		},
+	},
+
+	[ C(ITLB) ] = {
+		[ C(OP_READ) ] = {
+			[ C(RESULT_ACCESS) ] = 0,
+			[ C(RESULT_MISS)   ] = 0x02a0,
+		},
+		[ C(OP_WRITE) ] = {
+			[ C(RESULT_ACCESS) ] = -1,
+			[ C(RESULT_MISS)   ] = -1,
+		},
+		[ C(OP_PREFETCH) ] = {
+			[ C(RESULT_ACCESS) ] = -1,
+			[ C(RESULT_MISS)   ] = -1,
+		},
+	},
+
+	[ C(BPU) ] = {
+		[ C(OP_READ) ] = {
+			[ C(RESULT_ACCESS) ] = -1,
+			[ C(RESULT_MISS)   ] = -1,
+		},
+		[ C(OP_WRITE) ] = {
+			[ C(RESULT_ACCESS) ] = -1,
+			[ C(RESULT_MISS)   ] = -1,
+		},
+		[ C(OP_PREFETCH) ] = {
+			[ C(RESULT_ACCESS) ] = -1,
+			[ C(RESULT_MISS)   ] = -1,
+		},
+	},
+};
+
+static int sh4a_event_map(int event)
+{
+	return sh4a_general_events[event];
+}
+
+static u64 sh4a_pmu_read(int idx)
+{
+	return __raw_readl(PPC_PMCTR(idx));
+}
+
+static void sh4a_pmu_disable(struct hw_perf_event *hwc, int idx)
+{
+	unsigned int tmp;
+
+	tmp = __raw_readl(PPC_CCBR(idx));
+	tmp &= ~(CCBR_CIT_MASK | CCBR_DUC);
+	__raw_writel(tmp, PPC_CCBR(idx));
+}
+
+static void sh4a_pmu_enable(struct hw_perf_event *hwc, int idx)
+{
+	unsigned int tmp;
+
+	tmp = __raw_readl(PPC_PMCAT);
+	tmp &= ~PMCAT_EMU_CLR_MASK;
+	tmp |= idx ? PMCAT_CLR1 : PMCAT_CLR0;
+	__raw_writel(tmp, PPC_PMCAT);
+
+	tmp = __raw_readl(PPC_CCBR(idx));
+	tmp |= (hwc->config << 6) | CCBR_CMDS | CCBR_PPCE;
+	__raw_writel(tmp, PPC_CCBR(idx));
+
+	__raw_writel(__raw_readl(PPC_CCBR(idx)) | CCBR_DUC, PPC_CCBR(idx));
+}
+
+static void sh4a_pmu_disable_all(void)
+{
+	int i;
+
+	for (i = 0; i < sh4a_pmu.num_events; i++)
+		__raw_writel(__raw_readl(PPC_CCBR(i)) & ~CCBR_DUC, PPC_CCBR(i));
+}
+
+static void sh4a_pmu_enable_all(void)
+{
+	int i;
+
+	for (i = 0; i < sh4a_pmu.num_events; i++)
+		__raw_writel(__raw_readl(PPC_CCBR(i)) | CCBR_DUC, PPC_CCBR(i));
+}
+
+static struct sh_pmu sh4a_pmu = {
+	.name		= "SH-4A",
+	.num_events	= 2,
+	.event_map	= sh4a_event_map,
+	.max_events	= ARRAY_SIZE(sh4a_general_events),
+	.raw_event_mask	= 0x3ff,
+	.cache_events	= &sh4a_cache_events,
+	.read		= sh4a_pmu_read,
+	.disable	= sh4a_pmu_disable,
+	.enable		= sh4a_pmu_enable,
+	.disable_all	= sh4a_pmu_disable_all,
+	.enable_all	= sh4a_pmu_enable_all,
+};
+
+static int __init sh4a_pmu_init(void)
+{
+	/*
+	 * Make sure this CPU actually has perf counters.
+	 */
+	if (!(boot_cpu_data.flags & CPU_HAS_PERF_COUNTER)) {
+		pr_notice("HW perf events unsupported, software events only.\n");
+		return -ENODEV;
+	}
+
+	return register_sh_pmu(&sh4a_pmu);
+}
+arch_initcall(sh4a_pmu_init);
diff --git a/arch/sh/kernel/cpu/sh4a/setup-sh7724.c b/arch/sh/kernel/cpu/sh4a/setup-sh7724.c
index f3851fd757ec..845e89c936e7 100644
--- a/arch/sh/kernel/cpu/sh4a/setup-sh7724.c
+++ b/arch/sh/kernel/cpu/sh4a/setup-sh7724.c
@@ -20,6 +20,8 @@
 #include <linux/uio_driver.h>
 #include <linux/sh_timer.h>
 #include <linux/io.h>
+#include <linux/notifier.h>
+#include <asm/suspend.h>
 #include <asm/clock.h>
 #include <asm/mmzone.h>
 #include <cpu/sh7724.h>
@@ -202,7 +204,7 @@ static struct resource veu0_resources[] = {
 	[0] = {
 		.name	= "VEU3F0",
 		.start	= 0xfe920000,
-		.end	= 0xfe9200cb - 1,
+		.end	= 0xfe9200cb,
 		.flags	= IORESOURCE_MEM,
 	},
 	[1] = {
@@ -234,7 +236,7 @@ static struct resource veu1_resources[] = {
 	[0] = {
 		.name	= "VEU3F1",
 		.start	= 0xfe924000,
-		.end	= 0xfe9240cb - 1,
+		.end	= 0xfe9240cb,
 		.flags	= IORESOURCE_MEM,
 	},
 	[1] = {
@@ -523,6 +525,70 @@ static struct platform_device jpu_device = {
 	},
 };
 
+/* SPU2DSP0 */
+static struct uio_info spu0_platform_data = {
+	.name = "SPU2DSP0",
+	.version = "0",
+	.irq = 86,
+};
+
+static struct resource spu0_resources[] = {
+	[0] = {
+		.name	= "SPU2DSP0",
+		.start	= 0xFE200000,
+		.end	= 0xFE2FFFFF,
+		.flags	= IORESOURCE_MEM,
+	},
+	[1] = {
+		/* place holder for contiguous memory */
+	},
+};
+
+static struct platform_device spu0_device = {
+	.name		= "uio_pdrv_genirq",
+	.id		= 4,
+	.dev = {
+		.platform_data	= &spu0_platform_data,
+	},
+	.resource	= spu0_resources,
+	.num_resources	= ARRAY_SIZE(spu0_resources),
+	.archdata = {
+		.hwblk_id = HWBLK_SPU,
+	},
+};
+
+/* SPU2DSP1 */
+static struct uio_info spu1_platform_data = {
+	.name = "SPU2DSP1",
+	.version = "0",
+	.irq = 87,
+};
+
+static struct resource spu1_resources[] = {
+	[0] = {
+		.name	= "SPU2DSP1",
+		.start	= 0xFE300000,
+		.end	= 0xFE3FFFFF,
+		.flags	= IORESOURCE_MEM,
+	},
+	[1] = {
+		/* place holder for contiguous memory */
+	},
+};
+
+static struct platform_device spu1_device = {
+	.name		= "uio_pdrv_genirq",
+	.id		= 5,
+	.dev = {
+		.platform_data	= &spu1_platform_data,
+	},
+	.resource	= spu1_resources,
+	.num_resources	= ARRAY_SIZE(spu1_resources),
+	.archdata = {
+		.hwblk_id = HWBLK_SPU,
+	},
+};
+
 static struct platform_device *sh7724_devices[] __initdata = {
 	&cmt_device,
 	&tmu0_device,
@@ -539,6 +605,8 @@ static struct platform_device *sh7724_devices[] __initdata = {
 	&veu0_device,
 	&veu1_device,
 	&jpu_device,
+	&spu0_device,
+	&spu1_device,
 };
 
 static int __init sh7724_devices_setup(void)
@@ -547,6 +615,8 @@ static int __init sh7724_devices_setup(void)
 	platform_resource_setup_memory(&veu0_device, "veu0", 2 << 20);
 	platform_resource_setup_memory(&veu1_device, "veu1", 2 << 20);
 	platform_resource_setup_memory(&jpu_device,  "jpu",  2 << 20);
+	platform_resource_setup_memory(&spu0_device, "spu0", 2 << 20);
+	platform_resource_setup_memory(&spu1_device, "spu1", 2 << 20);
 
 	return platform_add_devices(sh7724_devices,
 				    ARRAY_SIZE(sh7724_devices));
@@ -827,3 +897,193 @@ void __init plat_irq_setup(void)
 {
 	register_intc_controller(&intc_desc);
 }
+
+static struct {
+	/* BSC */
+	unsigned long mmselr;
+	unsigned long cs0bcr;
+	unsigned long cs4bcr;
+	unsigned long cs5abcr;
+	unsigned long cs5bbcr;
+	unsigned long cs6abcr;
+	unsigned long cs6bbcr;
+	unsigned long cs4wcr;
+	unsigned long cs5awcr;
+	unsigned long cs5bwcr;
+	unsigned long cs6awcr;
+	unsigned long cs6bwcr;
+	/* INTC */
+	unsigned short ipra;
+	unsigned short iprb;
+	unsigned short iprc;
+	unsigned short iprd;
+	unsigned short ipre;
+	unsigned short iprf;
+	unsigned short iprg;
+	unsigned short iprh;
+	unsigned short ipri;
+	unsigned short iprj;
+	unsigned short iprk;
+	unsigned short iprl;
+	unsigned char imr0;
+	unsigned char imr1;
+	unsigned char imr2;
+	unsigned char imr3;
+	unsigned char imr4;
+	unsigned char imr5;
+	unsigned char imr6;
+	unsigned char imr7;
+	unsigned char imr8;
+	unsigned char imr9;
+	unsigned char imr10;
+	unsigned char imr11;
+	unsigned char imr12;
+	/* RWDT */
+	unsigned short rwtcnt;
+	unsigned short rwtcsr;
+	/* CPG */
+	unsigned long irdaclk;
+	unsigned long spuclk;
+} sh7724_rstandby_state;
+
+static int sh7724_pre_sleep_notifier_call(struct notifier_block *nb,
+					  unsigned long flags, void *unused)
+{
+	if (!(flags & SUSP_SH_RSTANDBY))
+		return NOTIFY_DONE;
+
+	/* BCR */
+	sh7724_rstandby_state.mmselr = __raw_readl(0xff800020); /* MMSELR */
+	sh7724_rstandby_state.mmselr |= 0xa5a50000;
+	sh7724_rstandby_state.cs0bcr = __raw_readl(0xfec10004); /* CS0BCR */
+	sh7724_rstandby_state.cs4bcr = __raw_readl(0xfec10010); /* CS4BCR */
+	sh7724_rstandby_state.cs5abcr = __raw_readl(0xfec10014); /* CS5ABCR */
+	sh7724_rstandby_state.cs5bbcr = __raw_readl(0xfec10018); /* CS5BBCR */
+	sh7724_rstandby_state.cs6abcr = __raw_readl(0xfec1001c); /* CS6ABCR */
+	sh7724_rstandby_state.cs6bbcr = __raw_readl(0xfec10020); /* CS6BBCR */
+	sh7724_rstandby_state.cs4wcr = __raw_readl(0xfec10030); /* CS4WCR */
+	sh7724_rstandby_state.cs5awcr = __raw_readl(0xfec10034); /* CS5AWCR */
+	sh7724_rstandby_state.cs5bwcr = __raw_readl(0xfec10038); /* CS5BWCR */
+	sh7724_rstandby_state.cs6awcr = __raw_readl(0xfec1003c); /* CS6AWCR */
+	sh7724_rstandby_state.cs6bwcr = __raw_readl(0xfec10040); /* CS6BWCR */
+
+	/* INTC */
+	sh7724_rstandby_state.ipra = __raw_readw(0xa4080000); /* IPRA */
+	sh7724_rstandby_state.iprb = __raw_readw(0xa4080004); /* IPRB */
+	sh7724_rstandby_state.iprc = __raw_readw(0xa4080008); /* IPRC */
+	sh7724_rstandby_state.iprd = __raw_readw(0xa408000c); /* IPRD */
+	sh7724_rstandby_state.ipre = __raw_readw(0xa4080010); /* IPRE */
+	sh7724_rstandby_state.iprf = __raw_readw(0xa4080014); /* IPRF */
+	sh7724_rstandby_state.iprg = __raw_readw(0xa4080018); /* IPRG */
+	sh7724_rstandby_state.iprh = __raw_readw(0xa408001c); /* IPRH */
+	sh7724_rstandby_state.ipri = __raw_readw(0xa4080020); /* IPRI */
+	sh7724_rstandby_state.iprj = __raw_readw(0xa4080024); /* IPRJ */
+	sh7724_rstandby_state.iprk = __raw_readw(0xa4080028); /* IPRK */
+	sh7724_rstandby_state.iprl = __raw_readw(0xa408002c); /* IPRL */
+	sh7724_rstandby_state.imr0 = __raw_readb(0xa4080080); /* IMR0 */
+	sh7724_rstandby_state.imr1 = __raw_readb(0xa4080084); /* IMR1 */
+	sh7724_rstandby_state.imr2 = __raw_readb(0xa4080088); /* IMR2 */
+	sh7724_rstandby_state.imr3 = __raw_readb(0xa408008c); /* IMR3 */
+	sh7724_rstandby_state.imr4 = __raw_readb(0xa4080090); /* IMR4 */
+	sh7724_rstandby_state.imr5 = __raw_readb(0xa4080094); /* IMR5 */
+	sh7724_rstandby_state.imr6 = __raw_readb(0xa4080098); /* IMR6 */
+	sh7724_rstandby_state.imr7 = __raw_readb(0xa408009c); /* IMR7 */
+	sh7724_rstandby_state.imr8 = __raw_readb(0xa40800a0); /* IMR8 */
+	sh7724_rstandby_state.imr9 = __raw_readb(0xa40800a4); /* IMR9 */
+	sh7724_rstandby_state.imr10 = __raw_readb(0xa40800a8); /* IMR10 */
+	sh7724_rstandby_state.imr11 = __raw_readb(0xa40800ac); /* IMR11 */
+	sh7724_rstandby_state.imr12 = __raw_readb(0xa40800b0); /* IMR12 */
+
+	/* RWDT */
+	sh7724_rstandby_state.rwtcnt = __raw_readb(0xa4520000); /* RWTCNT */
+	sh7724_rstandby_state.rwtcnt |= 0x5a00;
+	sh7724_rstandby_state.rwtcsr = __raw_readb(0xa4520004); /* RWTCSR */
+	sh7724_rstandby_state.rwtcsr |= 0xa500;
+	__raw_writew(sh7724_rstandby_state.rwtcsr & 0x07, 0xa4520004);
+
+	/* CPG */
+	sh7724_rstandby_state.irdaclk = __raw_readl(0xa4150018); /* IRDACLKCR */
+	sh7724_rstandby_state.spuclk = __raw_readl(0xa415003c); /* SPUCLKCR */
+
+	return NOTIFY_DONE;
+}
+
+static int sh7724_post_sleep_notifier_call(struct notifier_block *nb,
+					   unsigned long flags, void *unused)
+{
+	if (!(flags & SUSP_SH_RSTANDBY))
+		return NOTIFY_DONE;
+
+	/* BCR */
+	__raw_writel(sh7724_rstandby_state.mmselr, 0xff800020); /* MMSELR */
+	__raw_writel(sh7724_rstandby_state.cs0bcr, 0xfec10004); /* CS0BCR */
+	__raw_writel(sh7724_rstandby_state.cs4bcr, 0xfec10010); /* CS4BCR */
+	__raw_writel(sh7724_rstandby_state.cs5abcr, 0xfec10014); /* CS5ABCR */
+	__raw_writel(sh7724_rstandby_state.cs5bbcr, 0xfec10018); /* CS5BBCR */
+	__raw_writel(sh7724_rstandby_state.cs6abcr, 0xfec1001c); /* CS6ABCR */
+	__raw_writel(sh7724_rstandby_state.cs6bbcr, 0xfec10020); /* CS6BBCR */
+	__raw_writel(sh7724_rstandby_state.cs4wcr, 0xfec10030); /* CS4WCR */
+	__raw_writel(sh7724_rstandby_state.cs5awcr, 0xfec10034); /* CS5AWCR */
+	__raw_writel(sh7724_rstandby_state.cs5bwcr, 0xfec10038); /* CS5BWCR */
+	__raw_writel(sh7724_rstandby_state.cs6awcr, 0xfec1003c); /* CS6AWCR */
+	__raw_writel(sh7724_rstandby_state.cs6bwcr, 0xfec10040); /* CS6BWCR */
+
+	/* INTC */
+	__raw_writew(sh7724_rstandby_state.ipra, 0xa4080000); /* IPRA */
+	__raw_writew(sh7724_rstandby_state.iprb, 0xa4080004); /* IPRB */
+	__raw_writew(sh7724_rstandby_state.iprc, 0xa4080008); /* IPRC */
+	__raw_writew(sh7724_rstandby_state.iprd, 0xa408000c); /* IPRD */
+	__raw_writew(sh7724_rstandby_state.ipre, 0xa4080010); /* IPRE */
+	__raw_writew(sh7724_rstandby_state.iprf, 0xa4080014); /* IPRF */
+	__raw_writew(sh7724_rstandby_state.iprg, 0xa4080018); /* IPRG */
+	__raw_writew(sh7724_rstandby_state.iprh, 0xa408001c); /* IPRH */
+	__raw_writew(sh7724_rstandby_state.ipri, 0xa4080020); /* IPRI */
+	__raw_writew(sh7724_rstandby_state.iprj, 0xa4080024); /* IPRJ */
+	__raw_writew(sh7724_rstandby_state.iprk, 0xa4080028); /* IPRK */
+	__raw_writew(sh7724_rstandby_state.iprl, 0xa408002c); /* IPRL */
+	__raw_writeb(sh7724_rstandby_state.imr0, 0xa4080080); /* IMR0 */
+	__raw_writeb(sh7724_rstandby_state.imr1, 0xa4080084); /* IMR1 */
+	__raw_writeb(sh7724_rstandby_state.imr2, 0xa4080088); /* IMR2 */
+	__raw_writeb(sh7724_rstandby_state.imr3, 0xa408008c); /* IMR3 */
+	__raw_writeb(sh7724_rstandby_state.imr4, 0xa4080090); /* IMR4 */
+	__raw_writeb(sh7724_rstandby_state.imr5, 0xa4080094); /* IMR5 */
+	__raw_writeb(sh7724_rstandby_state.imr6, 0xa4080098); /* IMR6 */
+	__raw_writeb(sh7724_rstandby_state.imr7, 0xa408009c); /* IMR7 */
+	__raw_writeb(sh7724_rstandby_state.imr8, 0xa40800a0); /* IMR8 */
+	__raw_writeb(sh7724_rstandby_state.imr9, 0xa40800a4); /* IMR9 */
+	__raw_writeb(sh7724_rstandby_state.imr10, 0xa40800a8); /* IMR10 */
+	__raw_writeb(sh7724_rstandby_state.imr11, 0xa40800ac); /* IMR11 */
+	__raw_writeb(sh7724_rstandby_state.imr12, 0xa40800b0); /* IMR12 */
+
+	/* RWDT */
+	__raw_writew(sh7724_rstandby_state.rwtcnt, 0xa4520000); /* RWTCNT */
+	__raw_writew(sh7724_rstandby_state.rwtcsr, 0xa4520004); /* RWTCSR */
+
+	/* CPG */
+	__raw_writel(sh7724_rstandby_state.irdaclk, 0xa4150018); /* IRDACLKCR */
+	__raw_writel(sh7724_rstandby_state.spuclk, 0xa415003c); /* SPUCLKCR */
+
+	return NOTIFY_DONE;
+}
+
+static struct notifier_block sh7724_pre_sleep_notifier = {
+	.notifier_call = sh7724_pre_sleep_notifier_call,
+	.priority = SH_MOBILE_PRE(SH_MOBILE_SLEEP_CPU),
+};
+
+static struct notifier_block sh7724_post_sleep_notifier = {
+	.notifier_call = sh7724_post_sleep_notifier_call,
+	.priority = SH_MOBILE_POST(SH_MOBILE_SLEEP_CPU),
+};
+
+static int __init sh7724_sleep_setup(void)
+{
+	atomic_notifier_chain_register(&sh_mobile_pre_sleep_notifier_list,
+				       &sh7724_pre_sleep_notifier);
+
+	atomic_notifier_chain_register(&sh_mobile_post_sleep_notifier_list,
+				       &sh7724_post_sleep_notifier);
+	return 0;
+}
+arch_initcall(sh7724_sleep_setup);
+
diff --git a/arch/sh/kernel/cpu/sh4a/setup-shx3.c b/arch/sh/kernel/cpu/sh4a/setup-shx3.c
index e848443deeb9..c7ba9166e18a 100644
--- a/arch/sh/kernel/cpu/sh4a/setup-shx3.c
+++ b/arch/sh/kernel/cpu/sh4a/setup-shx3.c
@@ -15,6 +15,15 @@
 #include <linux/sh_timer.h>
 #include <asm/mmzone.h>
 
+/*
+ * This intentionally only registers SCIF ports 0, 1, and 3. SCIF 2
+ * INTEVT values overlap with the FPU EXPEVT ones, requiring special
+ * demuxing in the exception dispatch path.
+ *
+ * As this overlap is something that never should have made it in to
+ * silicon in the first place, we just refuse to deal with the port at
+ * all rather than adding infrastructure to hack around it.
+ */
 static struct plat_sci_port sci_platform_data[] = {
 	{
 		.mapbase	= 0xffc30000,
@@ -27,11 +36,6 @@ static struct plat_sci_port sci_platform_data[] = {
 		.type		= PORT_SCIF,
 		.irqs		= { 44, 45, 47, 46 },
 	}, {
-		.mapbase	= 0xffc50000,
-		.flags		= UPF_BOOT_AUTOCONF,
-		.type		= PORT_SCIF,
-		.irqs		= { 48, 49, 51, 50 },
-	}, {
 		.mapbase	= 0xffc60000,
 		.flags		= UPF_BOOT_AUTOCONF,
 		.type		= PORT_SCIF,
@@ -268,7 +272,11 @@ enum {
 	UNUSED = 0,
 
 	/* interrupt sources */
-	IRL, IRQ0, IRQ1, IRQ2, IRQ3,
+	IRL_LLLL, IRL_LLLH, IRL_LLHL, IRL_LLHH,
+	IRL_LHLL, IRL_LHLH, IRL_LHHL, IRL_LHHH,
+	IRL_HLLL, IRL_HLLH, IRL_HLHL, IRL_HLHH,
+	IRL_HHLL, IRL_HHLH, IRL_HHHL,
+	IRQ0, IRQ1, IRQ2, IRQ3,
 	HUDII,
 	TMU0, TMU1, TMU2, TMU3, TMU4, TMU5,
 	PCII0, PCII1, PCII2, PCII3, PCII4,
@@ -291,7 +299,7 @@ enum {
 	INTICI4, INTICI5, INTICI6, INTICI7,
 
 	/* interrupt groups */
-	PCII56789, SCIF0, SCIF1, SCIF2, SCIF3,
+	IRL, PCII56789, SCIF0, SCIF1, SCIF2, SCIF3,
 	DMAC0, DMAC1,
 };
 
@@ -309,8 +317,6 @@ static struct intc_vect vectors[] __initdata = {
 	INTC_VECT(SCIF0_BRI, 0x740), INTC_VECT(SCIF0_TXI, 0x760),
 	INTC_VECT(SCIF1_ERI, 0x780), INTC_VECT(SCIF1_RXI, 0x7a0),
 	INTC_VECT(SCIF1_BRI, 0x7c0), INTC_VECT(SCIF1_TXI, 0x7e0),
-	INTC_VECT(SCIF2_ERI, 0x800), INTC_VECT(SCIF2_RXI, 0x820),
-	INTC_VECT(SCIF2_BRI, 0x840), INTC_VECT(SCIF2_TXI, 0x860),
 	INTC_VECT(SCIF3_ERI, 0x880), INTC_VECT(SCIF3_RXI, 0x8a0),
 	INTC_VECT(SCIF3_BRI, 0x8c0), INTC_VECT(SCIF3_TXI, 0x8e0),
 	INTC_VECT(DMAC0_DMINT0, 0x900), INTC_VECT(DMAC0_DMINT1, 0x920),
@@ -344,10 +350,13 @@ static struct intc_vect vectors[] __initdata = {
 };
 
 static struct intc_group groups[] __initdata = {
+	INTC_GROUP(IRL, IRL_LLLL, IRL_LLLH, IRL_LLHL, IRL_LLHH,
+		   IRL_LHLL, IRL_LHLH, IRL_LHHL, IRL_LHHH,
+		   IRL_HLLL, IRL_HLLH, IRL_HLHL, IRL_HLHH,
+		   IRL_HHLL, IRL_HHLH, IRL_HHHL),
 	INTC_GROUP(PCII56789, PCII5, PCII6, PCII7, PCII8, PCII9),
 	INTC_GROUP(SCIF0, SCIF0_ERI, SCIF0_RXI, SCIF0_BRI, SCIF0_TXI),
 	INTC_GROUP(SCIF1, SCIF1_ERI, SCIF1_RXI, SCIF1_BRI, SCIF1_TXI),
-	INTC_GROUP(SCIF2, SCIF2_ERI, SCIF2_RXI, SCIF2_BRI, SCIF2_TXI),
 	INTC_GROUP(SCIF3, SCIF3_ERI, SCIF3_RXI, SCIF3_BRI, SCIF3_TXI),
 	INTC_GROUP(DMAC0, DMAC0_DMINT0, DMAC0_DMINT1, DMAC0_DMINT2,
 		   DMAC0_DMINT3, DMAC0_DMINT4, DMAC0_DMINT5, DMAC0_DMAE),
@@ -419,14 +428,14 @@ static DECLARE_INTC_DESC(intc_desc_irq, "shx3-irq", vectors_irq, groups,
 
 /* External interrupt pins in IRL mode */
 static struct intc_vect vectors_irl[] __initdata = {
-	INTC_VECT(IRL, 0x200), INTC_VECT(IRL, 0x220),
-	INTC_VECT(IRL, 0x240), INTC_VECT(IRL, 0x260),
-	INTC_VECT(IRL, 0x280), INTC_VECT(IRL, 0x2a0),
-	INTC_VECT(IRL, 0x2c0), INTC_VECT(IRL, 0x2e0),
-	INTC_VECT(IRL, 0x300), INTC_VECT(IRL, 0x320),
-	INTC_VECT(IRL, 0x340), INTC_VECT(IRL, 0x360),
-	INTC_VECT(IRL, 0x380), INTC_VECT(IRL, 0x3a0),
-	INTC_VECT(IRL, 0x3c0),
+	INTC_VECT(IRL_LLLL, 0x200), INTC_VECT(IRL_LLLH, 0x220),
+	INTC_VECT(IRL_LLHL, 0x240), INTC_VECT(IRL_LLHH, 0x260),
+	INTC_VECT(IRL_LHLL, 0x280), INTC_VECT(IRL_LHLH, 0x2a0),
+	INTC_VECT(IRL_LHHL, 0x2c0), INTC_VECT(IRL_LHHH, 0x2e0),
+	INTC_VECT(IRL_HLLL, 0x300), INTC_VECT(IRL_HLLH, 0x320),
+	INTC_VECT(IRL_HLHL, 0x340), INTC_VECT(IRL_HLHH, 0x360),
+	INTC_VECT(IRL_HHLL, 0x380), INTC_VECT(IRL_HHLH, 0x3a0),
+	INTC_VECT(IRL_HHHL, 0x3c0),
 };
 
 static DECLARE_INTC_DESC(intc_desc_irl, "shx3-irl", vectors_irl, groups,
diff --git a/arch/sh/kernel/cpu/sh4a/smp-shx3.c b/arch/sh/kernel/cpu/sh4a/smp-shx3.c
index 185ec3976a25..5863e0c4d02f 100644
--- a/arch/sh/kernel/cpu/sh4a/smp-shx3.c
+++ b/arch/sh/kernel/cpu/sh4a/smp-shx3.c
@@ -14,6 +14,13 @@
 #include <linux/interrupt.h>
 #include <linux/io.h>
 
+#define STBCR_REG(phys_id) (0xfe400004 | (phys_id << 12))
+#define RESET_REG(phys_id) (0xfe400008 | (phys_id << 12))
+
+#define STBCR_MSTP	0x00000001
+#define STBCR_RESET	0x00000002
+#define STBCR_LTSLP	0x80000000
+
 static irqreturn_t ipi_interrupt_handler(int irq, void *arg)
 {
 	unsigned int message = (unsigned int)(long)arg;
@@ -21,9 +28,9 @@ static irqreturn_t ipi_interrupt_handler(int irq, void *arg)
 	unsigned int offs = 4 * cpu;
 	unsigned int x;
 
-	x = ctrl_inl(0xfe410070 + offs); /* C0INITICI..CnINTICI */
+	x = __raw_readl(0xfe410070 + offs); /* C0INITICI..CnINTICI */
 	x &= (1 << (message << 2));
-	ctrl_outl(x, 0xfe410080 + offs); /* C0INTICICLR..CnINTICICLR */
+	__raw_writel(x, 0xfe410080 + offs); /* C0INTICICLR..CnINTICICLR */
 
 	smp_message_recv(message);
 
@@ -37,6 +44,9 @@ void __init plat_smp_setup(void)
 
 	init_cpu_possible(cpumask_of(cpu));
 
+	/* Enable light sleep for the boot CPU */
+	__raw_writel(__raw_readl(STBCR_REG(cpu)) | STBCR_LTSLP, STBCR_REG(cpu));
+
 	__cpu_number_map[0] = 0;
 	__cpu_logical_map[0] = 0;
 
@@ -66,32 +76,23 @@ void __init plat_prepare_cpus(unsigned int max_cpus)
 			    "IPI", (void *)(long)i);
 }
 
-#define STBCR_REG(phys_id) (0xfe400004 | (phys_id << 12))
-#define RESET_REG(phys_id) (0xfe400008 | (phys_id << 12))
-
-#define STBCR_MSTP	0x00000001
-#define STBCR_RESET	0x00000002
-#define STBCR_LTSLP	0x80000000
-
-#define STBCR_AP_VAL	(STBCR_RESET | STBCR_LTSLP)
-
 void plat_start_cpu(unsigned int cpu, unsigned long entry_point)
 {
-	ctrl_outl(entry_point, RESET_REG(cpu));
+	__raw_writel(entry_point, RESET_REG(cpu));
 
-	if (!(ctrl_inl(STBCR_REG(cpu)) & STBCR_MSTP))
-		ctrl_outl(STBCR_MSTP, STBCR_REG(cpu));
+	if (!(__raw_readl(STBCR_REG(cpu)) & STBCR_MSTP))
+		__raw_writel(STBCR_MSTP, STBCR_REG(cpu));
 
-	while (!(ctrl_inl(STBCR_REG(cpu)) & STBCR_MSTP))
+	while (!(__raw_readl(STBCR_REG(cpu)) & STBCR_MSTP))
 		cpu_relax();
 
 	/* Start up secondary processor by sending a reset */
-	ctrl_outl(STBCR_AP_VAL, STBCR_REG(cpu));
+	__raw_writel(STBCR_RESET | STBCR_LTSLP, STBCR_REG(cpu));
 }
 
 int plat_smp_processor_id(void)
 {
-	return ctrl_inl(0xff000048); /* CPIDR */
+	return __raw_readl(0xff000048); /* CPIDR */
 }
 
 void plat_send_ipi(unsigned int cpu, unsigned int message)
@@ -100,5 +101,5 @@ void plat_send_ipi(unsigned int cpu, unsigned int message)
 
 	BUG_ON(cpu >= 4);
 
-	ctrl_outl(1 << (message << 2), addr); /* C0INTICI..CnINTICI */
+	__raw_writel(1 << (message << 2), addr); /* C0INTICI..CnINTICI */
 }
diff --git a/arch/sh/kernel/cpu/sh5/entry.S b/arch/sh/kernel/cpu/sh5/entry.S
index b0aacf675258..8f13f73cb2cb 100644
--- a/arch/sh/kernel/cpu/sh5/entry.S
+++ b/arch/sh/kernel/cpu/sh5/entry.S
@@ -933,7 +933,7 @@ ret_with_reschedule:
 
 	pta	restore_all, tr1
 
-	movi	(_TIF_SIGPENDING | _TIF_RESTORE_SIGMASK), r8
+	movi	_TIF_SIGPENDING, r8
 	and	r8, r7, r8
 	pta	work_notifysig, tr0
 	bne	r8, ZERO, tr0
diff --git a/arch/sh/kernel/cpu/shmobile/cpuidle.c b/arch/sh/kernel/cpu/shmobile/cpuidle.c
index 1c504bd972c3..83972aa319c2 100644
--- a/arch/sh/kernel/cpu/shmobile/cpuidle.c
+++ b/arch/sh/kernel/cpu/shmobile/cpuidle.c
@@ -87,25 +87,31 @@ void sh_mobile_setup_cpuidle(void)
 
 	dev->safe_state = state;
 
-	state = &dev->states[i++];
-	snprintf(state->name, CPUIDLE_NAME_LEN, "C1");
-	strncpy(state->desc, "SuperH Sleep Mode [SF]", CPUIDLE_DESC_LEN);
-	state->exit_latency = 100;
-	state->target_residency = 1 * 2;
-	state->power_usage = 1;
-	state->flags = 0;
-	state->flags |= CPUIDLE_FLAG_TIME_VALID;
-	state->enter = cpuidle_sleep_enter;
+	if (sh_mobile_sleep_supported & SUSP_SH_SF) {
+		state = &dev->states[i++];
+		snprintf(state->name, CPUIDLE_NAME_LEN, "C1");
+		strncpy(state->desc, "SuperH Sleep Mode [SF]",
+			CPUIDLE_DESC_LEN);
+		state->exit_latency = 100;
+		state->target_residency = 1 * 2;
+		state->power_usage = 1;
+		state->flags = 0;
+		state->flags |= CPUIDLE_FLAG_TIME_VALID;
+		state->enter = cpuidle_sleep_enter;
+	}
 
-	state = &dev->states[i++];
-	snprintf(state->name, CPUIDLE_NAME_LEN, "C2");
-	strncpy(state->desc, "SuperH Mobile Standby Mode [SF]", CPUIDLE_DESC_LEN);
-	state->exit_latency = 2300;
-	state->target_residency = 1 * 2;
-	state->power_usage = 1;
-	state->flags = 0;
-	state->flags |= CPUIDLE_FLAG_TIME_VALID;
-	state->enter = cpuidle_sleep_enter;
+	if (sh_mobile_sleep_supported & SUSP_SH_STANDBY) {
+		state = &dev->states[i++];
+		snprintf(state->name, CPUIDLE_NAME_LEN, "C2");
+		strncpy(state->desc, "SuperH Mobile Standby Mode [SF]",
+			CPUIDLE_DESC_LEN);
+		state->exit_latency = 2300;
+		state->target_residency = 1 * 2;
+		state->power_usage = 1;
+		state->flags = 0;
+		state->flags |= CPUIDLE_FLAG_TIME_VALID;
+		state->enter = cpuidle_sleep_enter;
+	}
 
 	dev->state_count = i;
 
diff --git a/arch/sh/kernel/cpu/shmobile/pm.c b/arch/sh/kernel/cpu/shmobile/pm.c
index ee3c2aaf66fb..ca029a44743c 100644
--- a/arch/sh/kernel/cpu/shmobile/pm.c
+++ b/arch/sh/kernel/cpu/shmobile/pm.c
@@ -15,6 +15,13 @@
 #include <linux/suspend.h>
 #include <asm/suspend.h>
 #include <asm/uaccess.h>
+#include <asm/cacheflush.h>
+
+/*
+ * Notifier lists for pre/post sleep notification
+ */
+ATOMIC_NOTIFIER_HEAD(sh_mobile_pre_sleep_notifier_list);
+ATOMIC_NOTIFIER_HEAD(sh_mobile_post_sleep_notifier_list);
 
 /*
  * Sleep modes available on SuperH Mobile:
@@ -26,30 +33,105 @@
 #define SUSP_MODE_SLEEP		(SUSP_SH_SLEEP)
 #define SUSP_MODE_SLEEP_SF	(SUSP_SH_SLEEP | SUSP_SH_SF)
 #define SUSP_MODE_STANDBY_SF	(SUSP_SH_STANDBY | SUSP_SH_SF)
+#define SUSP_MODE_RSTANDBY	(SUSP_SH_RSTANDBY | SUSP_SH_MMU | SUSP_SH_SF)
+ /*
+  * U-standby mode is unsupported since it needs bootloader hacks
+  */
 
-/*
- * The following modes are not there yet:
- *
- * R-standby mode is unsupported, but will be added in the future
- * U-standby mode is low priority since it needs bootloader hacks
- */
-
-#define ILRAM_BASE 0xe5200000
-
-extern const unsigned char sh_mobile_standby[];
-extern const unsigned int sh_mobile_standby_size;
+#ifdef CONFIG_CPU_SUBTYPE_SH7724
+#define RAM_BASE 0xfd800000 /* RSMEM */
+#else
+#define RAM_BASE 0xe5200000 /* ILRAM */
+#endif
 
 void sh_mobile_call_standby(unsigned long mode)
 {
-	void *onchip_mem = (void *)ILRAM_BASE;
-	void (*standby_onchip_mem)(unsigned long, unsigned long) = onchip_mem;
+	void *onchip_mem = (void *)RAM_BASE;
+	struct sh_sleep_data *sdp = onchip_mem;
+	void (*standby_onchip_mem)(unsigned long, unsigned long);
+
+	/* code located directly after data structure */
+	standby_onchip_mem = (void *)(sdp + 1);
+
+	atomic_notifier_call_chain(&sh_mobile_pre_sleep_notifier_list,
+				   mode, NULL);
+
+	/* flush the caches if MMU flag is set */
+	if (mode & SUSP_SH_MMU)
+		flush_cache_all();
 
 	/* Let assembly snippet in on-chip memory handle the rest */
-	standby_onchip_mem(mode, ILRAM_BASE);
+	standby_onchip_mem(mode, RAM_BASE);
+
+	atomic_notifier_call_chain(&sh_mobile_post_sleep_notifier_list,
+				   mode, NULL);
+}
+
+extern char sh_mobile_sleep_enter_start;
+extern char sh_mobile_sleep_enter_end;
+
+extern char sh_mobile_sleep_resume_start;
+extern char sh_mobile_sleep_resume_end;
+
+unsigned long sh_mobile_sleep_supported = SUSP_SH_SLEEP;
+
+void sh_mobile_register_self_refresh(unsigned long flags,
+				     void *pre_start, void *pre_end,
+				     void *post_start, void *post_end)
+{
+	void *onchip_mem = (void *)RAM_BASE;
+	void *vp;
+	struct sh_sleep_data *sdp;
+	int n;
+
+	/* part 0: data area */
+	sdp = onchip_mem;
+	sdp->addr.stbcr = 0xa4150020; /* STBCR */
+	sdp->addr.bar = 0xa4150040; /* BAR */
+	sdp->addr.pteh = 0xff000000; /* PTEH */
+	sdp->addr.ptel = 0xff000004; /* PTEL */
+	sdp->addr.ttb = 0xff000008; /* TTB */
+	sdp->addr.tea = 0xff00000c; /* TEA */
+	sdp->addr.mmucr = 0xff000010; /* MMUCR */
+	sdp->addr.ptea = 0xff000034; /* PTEA */
+	sdp->addr.pascr = 0xff000070; /* PASCR */
+	sdp->addr.irmcr = 0xff000078; /* IRMCR */
+	sdp->addr.ccr = 0xff00001c; /* CCR */
+	sdp->addr.ramcr = 0xff000074; /* RAMCR */
+	vp = sdp + 1;
+
+	/* part 1: common code to enter sleep mode */
+	n = &sh_mobile_sleep_enter_end - &sh_mobile_sleep_enter_start;
+	memcpy(vp, &sh_mobile_sleep_enter_start, n);
+	vp += roundup(n, 4);
+
+	/* part 2: board specific code to enter self-refresh mode */
+	n = pre_end - pre_start;
+	memcpy(vp, pre_start, n);
+	sdp->sf_pre = (unsigned long)vp;
+	vp += roundup(n, 4);
+
+	/* part 3: board specific code to resume from self-refresh mode */
+	n = post_end - post_start;
+	memcpy(vp, post_start, n);
+	sdp->sf_post = (unsigned long)vp;
+	vp += roundup(n, 4);
+
+	/* part 4: common code to resume from sleep mode */
+	WARN_ON(vp > (onchip_mem + 0x600));
+	vp = onchip_mem + 0x600; /* located at interrupt vector */
+	n = &sh_mobile_sleep_resume_end - &sh_mobile_sleep_resume_start;
+	memcpy(vp, &sh_mobile_sleep_resume_start, n);
+	sdp->resume = (unsigned long)vp;
+
+	sh_mobile_sleep_supported |= flags;
 }
 
 static int sh_pm_enter(suspend_state_t state)
 {
+	if (!(sh_mobile_sleep_supported & SUSP_MODE_STANDBY_SF))
+		return -ENXIO;
+
 	local_irq_disable();
 	set_bl_bit();
 	sh_mobile_call_standby(SUSP_MODE_STANDBY_SF);
@@ -65,13 +147,6 @@ static struct platform_suspend_ops sh_pm_ops = {
 
 static int __init sh_pm_init(void)
 {
-	void *onchip_mem = (void *)ILRAM_BASE;
-
-	/* Copy the assembly snippet to the otherwise ununsed ILRAM */
-	memcpy(onchip_mem, sh_mobile_standby, sh_mobile_standby_size);
-	wmb();
-	ctrl_barrier();
-
 	suspend_set_ops(&sh_pm_ops);
 	sh_mobile_setup_cpuidle();
 	return 0;
diff --git a/arch/sh/kernel/cpu/shmobile/pm_runtime.c b/arch/sh/kernel/cpu/shmobile/pm_runtime.c
index 7c615b17e209..6dcb8166a64d 100644
--- a/arch/sh/kernel/cpu/shmobile/pm_runtime.c
+++ b/arch/sh/kernel/cpu/shmobile/pm_runtime.c
@@ -45,12 +45,14 @@ static int __platform_pm_runtime_resume(struct platform_device *pdev)
 
 	dev_dbg(d, "__platform_pm_runtime_resume() [%d]\n", hwblk);
 
-	if (d->driver && d->driver->pm && d->driver->pm->runtime_resume) {
+	if (d->driver) {
 		hwblk_enable(hwblk_info, hwblk);
 		ret = 0;
 
 		if (test_bit(PDEV_ARCHDATA_FLAG_SUSP, &ad->flags)) {
-			ret = d->driver->pm->runtime_resume(d);
+			if (d->driver->pm && d->driver->pm->runtime_resume)
+				ret = d->driver->pm->runtime_resume(d);
+
 			if (!ret)
 				clear_bit(PDEV_ARCHDATA_FLAG_SUSP, &ad->flags);
 			else
@@ -73,12 +75,15 @@ static int __platform_pm_runtime_suspend(struct platform_device *pdev)
 
 	dev_dbg(d, "__platform_pm_runtime_suspend() [%d]\n", hwblk);
 
-	if (d->driver && d->driver->pm && d->driver->pm->runtime_suspend) {
+	if (d->driver) {
 		BUG_ON(!test_bit(PDEV_ARCHDATA_FLAG_IDLE, &ad->flags));
+		ret = 0;
 
-		hwblk_enable(hwblk_info, hwblk);
-		ret = d->driver->pm->runtime_suspend(d);
-		hwblk_disable(hwblk_info, hwblk);
+		if (d->driver->pm && d->driver->pm->runtime_suspend) {
+			hwblk_enable(hwblk_info, hwblk);
+			ret = d->driver->pm->runtime_suspend(d);
+			hwblk_disable(hwblk_info, hwblk);
+		}
 
 		if (!ret) {
 			set_bit(PDEV_ARCHDATA_FLAG_SUSP, &ad->flags);
diff --git a/arch/sh/kernel/cpu/shmobile/sleep.S b/arch/sh/kernel/cpu/shmobile/sleep.S
index a439e6c7824f..e9dd7fa0abd2 100644
--- a/arch/sh/kernel/cpu/shmobile/sleep.S
+++ b/arch/sh/kernel/cpu/shmobile/sleep.S
@@ -20,79 +20,103 @@
  * Kernel mode register usage, see entry.S:
  *	k0	scratch
  *	k1	scratch
- *	k4	scratch
  */
 #define k0	r0
 #define k1	r1
-#define k4	r4
 
-/* manage self-refresh and enter standby mode.
+/* manage self-refresh and enter standby mode. must be self-contained.
  * this code will be copied to on-chip memory and executed from there.
  */
+	.balign 4
+ENTRY(sh_mobile_sleep_enter_start)
 
-	.balign 	4096,0,4096
-ENTRY(sh_mobile_standby)
+	/* save mode flags */
+	mov.l	r4, @(SH_SLEEP_MODE, r5)
 
 	/* save original vbr */
-	stc	vbr, r1
-	mova	saved_vbr, r0
-	mov.l	r1, @r0
+	stc	vbr, r0
+	mov.l	r0, @(SH_SLEEP_VBR, r5)
 
 	/* point vbr to our on-chip memory page */
 	ldc	r5, vbr
 
 	/* save return address */
-	mova	saved_spc, r0
-	sts	pr, r5
-	mov.l	r5, @r0
+	sts	pr, r0
+	mov.l	r0, @(SH_SLEEP_SPC, r5)
 
 	/* save sr */
-	mova	saved_sr, r0
-	stc	sr, r5
-	mov.l	r5, @r0
+	stc	sr, r0
+	mov.l	r0, @(SH_SLEEP_SR, r5)
 
-	/* save mode flags */
-	mova	saved_mode, r0
-	mov.l	r4, @r0
+	/* save sp */
+	mov.l	r15, @(SH_SLEEP_SP, r5)
+
+	/* save stbcr */
+	bsr     save_register
+	 mov    #SH_SLEEP_REG_STBCR, r0
+
+	/* save mmu and cache context if needed */
+	mov.l	@(SH_SLEEP_MODE, r5), r0
+	tst	#SUSP_SH_MMU, r0
+	bt	skip_mmu_save_disable
+
+       /* save mmu state */
+	bsr	save_register
+	 mov	#SH_SLEEP_REG_PTEH, r0
+
+	bsr	save_register
+	 mov	#SH_SLEEP_REG_PTEL, r0
+
+	bsr	save_register
+	 mov	#SH_SLEEP_REG_TTB, r0
+
+	bsr	save_register
+	 mov	#SH_SLEEP_REG_TEA, r0
+
+	bsr	save_register
+	 mov	#SH_SLEEP_REG_MMUCR, r0
+
+	bsr	save_register
+	 mov	#SH_SLEEP_REG_PTEA, r0
+
+	bsr	save_register
+	 mov	#SH_SLEEP_REG_PASCR, r0
 
-	/* put mode flags in r0 */
-	mov	r4, r0
+	bsr	save_register
+	 mov	#SH_SLEEP_REG_IRMCR, r0
 
+	/* invalidate TLBs and disable the MMU */
+	bsr	get_register
+	 mov	#SH_SLEEP_REG_MMUCR, r0
+	mov	#4, r1
+	mov.l	r1, @r0
+	icbi	@r0
+
+	/* save cache registers and disable caches */
+	bsr	save_register
+	 mov	#SH_SLEEP_REG_CCR, r0
+
+	bsr	save_register
+	 mov	#SH_SLEEP_REG_RAMCR, r0
+
+	bsr	get_register
+	 mov	#SH_SLEEP_REG_CCR, r0
+	mov	#0, r1
+	mov.l	r1, @r0
+	icbi	@r0
+
+skip_mmu_save_disable:
+	/* call self-refresh entering code if needed */
+	mov.l	@(SH_SLEEP_MODE, r5), r0
 	tst	#SUSP_SH_SF, r0
 	bt	skip_set_sf
-#ifdef CONFIG_CPU_SUBTYPE_SH7724
-	/* DBSC: put memory in self-refresh mode */
-	mov.l	dben_reg, r4
-	mov.l	dben_data0, r1
-	mov.l	r1, @r4
-
-	mov.l	dbrfpdn0_reg, r4
-	mov.l	dbrfpdn0_data0, r1
-	mov.l	r1, @r4
-
-	mov.l	dbcmdcnt_reg, r4
-	mov.l	dbcmdcnt_data0, r1
-	mov.l	r1, @r4
-
-	mov.l	dbcmdcnt_reg, r4
-	mov.l	dbcmdcnt_data1, r1
-	mov.l	r1, @r4
-
-	mov.l	dbrfpdn0_reg, r4
-	mov.l	dbrfpdn0_data1, r1
-	mov.l	r1, @r4
-#else
-	/* SBSC: disable power down and put in self-refresh mode */
-	mov.l	1f, r4
-	mov.l	2f, r1
-	mov.l	@r4, r2
-	or	r1, r2
-	mov.l   3f, r3
-	and	r3, r2
-	mov.l	r2, @r4
-#endif
+
+	mov.l	@(SH_SLEEP_SF_PRE, r5), r0
+	jsr	@r0
+	 nop
 
 skip_set_sf:
+	mov.l	@(SH_SLEEP_MODE, r5), r0
 	tst	#SUSP_SH_STANDBY, r0
 	bt	test_rstandby
 
@@ -104,6 +128,12 @@ test_rstandby:
 	tst	#SUSP_SH_RSTANDBY, r0
 	bt	test_ustandby
 
+	/* setup BAR register */
+	bsr	get_register
+	 mov	#SH_SLEEP_REG_BAR, r0
+	mov.l	@(SH_SLEEP_RESUME, r5), r1
+	mov.l	r1, @r0
+
 	/* set mode to "r-standby mode" */
 	bra	do_sleep
 	 mov	#0x20, r1
@@ -123,124 +153,136 @@ force_sleep:
 
 do_sleep:
 	/* setup and enter selected standby mode */
-	mov.l	5f, r4
-	mov.l	r1, @r4
+	bsr     get_register
+	 mov    #SH_SLEEP_REG_STBCR, r0
+	mov.l	r1, @r0
 again:
 	sleep
 	bra	again
 	 nop
 
-restore_jump_vbr:
+save_register:
+	add	#SH_SLEEP_BASE_ADDR, r0
+	mov.l	@(r0, r5), r1
+	add	#-SH_SLEEP_BASE_ADDR, r0
+	mov.l	@r1, r1
+	add	#SH_SLEEP_BASE_DATA, r0
+	mov.l	r1, @(r0, r5)
+	add	#-SH_SLEEP_BASE_DATA, r0
+	rts
+	 nop
+
+get_register:
+	add	#SH_SLEEP_BASE_ADDR, r0
+	mov.l	@(r0, r5), r0
+	rts
+	 nop
+ENTRY(sh_mobile_sleep_enter_end)
+
+	.balign 4
+ENTRY(sh_mobile_sleep_resume_start)
+
+	/* figure out start address */
+	bsr	0f
+	 nop
+0:
+	sts	pr, k1
+	mov.l	1f, k0
+	and	k0, k1
+
+	/* store pointer to data area in VBR */
+	ldc	k1, vbr
+
+	/* setup sr with saved sr */
+	mov.l	@(SH_SLEEP_SR, k1), k0
+	ldc	k0, sr
+
+	/* now: user register set! */
+	stc	vbr, r5
+
 	/* setup spc with return address to c code */
-	mov.l	saved_spc, k0
-	ldc	k0, spc
+	mov.l	@(SH_SLEEP_SPC, r5), r0
+	ldc	r0, spc
 
 	/* restore vbr */
-	mov.l	saved_vbr, k0
-	ldc	k0, vbr
+	mov.l	@(SH_SLEEP_VBR, r5), r0
+	ldc	r0, vbr
 
 	/* setup ssr with saved sr */
-	mov.l	saved_sr, k0
-	ldc	k0, ssr
+	mov.l	@(SH_SLEEP_SR, r5), r0
+	ldc	r0, ssr
 
-	/* get mode flags */
-	mov.l	saved_mode, k0
+	/* restore sp */
+	mov.l   @(SH_SLEEP_SP, r5), r15
 
-done_sleep:
-	/* reset standby mode to sleep mode */
-	mov.l	5f, k4
-	mov	#0x00, k1
-	mov.l	k1, @k4
+	/* restore sleep mode register */
+	bsr     restore_register
+	 mov    #SH_SLEEP_REG_STBCR, r0
 
-	tst	#SUSP_SH_SF, k0
+	/* call self-refresh resume code if needed */
+	mov.l	@(SH_SLEEP_MODE, r5), r0
+	tst	#SUSP_SH_SF, r0
 	bt	skip_restore_sf
 
-#ifdef CONFIG_CPU_SUBTYPE_SH7724
-	/* DBSC: put memory in auto-refresh mode */
-	mov.l	dbrfpdn0_reg, k4
-	mov.l	dbrfpdn0_data0, k1
-	mov.l	k1, @k4
-
-	nop /* sleep 140 ns */
-	nop
-	nop
-	nop
-
-	mov.l	dbcmdcnt_reg, k4
-	mov.l	dbcmdcnt_data0, k1
-	mov.l	k1, @k4
-
-	mov.l	dbcmdcnt_reg, k4
-	mov.l	dbcmdcnt_data1, k1
-	mov.l	k1, @k4
-
-	mov.l	dben_reg, k4
-	mov.l	dben_data1, k1
-	mov.l	k1, @k4
-
-	mov.l	dbrfpdn0_reg, k4
-	mov.l	dbrfpdn0_data2, k1
-	mov.l	k1, @k4
-#else
-	/* SBSC: set auto-refresh mode */
-	mov.l	1f, k4
-	mov.l	@k4, k0
-	mov.l   4f, k1
-	and	k1, k0
-	mov.l	k0, @k4
-	mov.l	6f, k4
-	mov.l	8f, k0
-	mov.l	@k4, k1
-	mov	#-1, k4
-	add	k4, k1
-	or	k1, k0
-	mov.l	7f, k1
-	mov.l	k0, @k1
-#endif
+	mov.l	@(SH_SLEEP_SF_POST, r5), r0
+	jsr	@r0
+	 nop
+
 skip_restore_sf:
-	/* jump to vbr vector */
-	mov.l	saved_vbr, k0
-	mov.l	offset_vbr, k4
-	add	k4, k0
-	jmp	@k0
+	/* restore mmu and cache state if needed */
+	mov.l	@(SH_SLEEP_MODE, r5), r0
+	tst	#SUSP_SH_MMU, r0
+	bt	skip_restore_mmu
+
+	/* restore mmu state */
+	bsr	restore_register
+	 mov	#SH_SLEEP_REG_PTEH, r0
+
+	bsr	restore_register
+	 mov	#SH_SLEEP_REG_PTEL, r0
+
+	bsr	restore_register
+	 mov	#SH_SLEEP_REG_TTB, r0
+
+	bsr	restore_register
+	 mov	#SH_SLEEP_REG_TEA, r0
+
+	bsr	restore_register
+	 mov	#SH_SLEEP_REG_PTEA, r0
+
+	bsr	restore_register
+	 mov	#SH_SLEEP_REG_PASCR, r0
+
+	bsr	restore_register
+	 mov	#SH_SLEEP_REG_IRMCR, r0
+
+	bsr	restore_register
+	 mov	#SH_SLEEP_REG_MMUCR, r0
+	icbi	@r0
+
+	/* restore cache settings */
+	bsr	restore_register
+	 mov	#SH_SLEEP_REG_RAMCR, r0
+	icbi	@r0
+
+	bsr	restore_register
+	 mov	#SH_SLEEP_REG_CCR, r0
+	icbi	@r0
+
+skip_restore_mmu:
+	rte
 	 nop
 
-	.balign 4
-saved_mode:	.long	0
-saved_spc:	.long	0
-saved_sr:	.long	0
-saved_vbr:	.long	0
-offset_vbr:	.long	0x600
-#ifdef CONFIG_CPU_SUBTYPE_SH7724
-dben_reg:	.long	0xfd000010 /* DBEN */
-dben_data0:	.long	0
-dben_data1:	.long	1
-dbrfpdn0_reg:	.long	0xfd000040 /* DBRFPDN0 */
-dbrfpdn0_data0:	.long	0
-dbrfpdn0_data1:	.long	1
-dbrfpdn0_data2:	.long	0x00010000
-dbcmdcnt_reg:	.long	0xfd000014 /* DBCMDCNT */
-dbcmdcnt_data0:	.long	2
-dbcmdcnt_data1:	.long	4
-#else
-1:	.long	0xfe400008 /* SDCR0 */
-2:	.long	0x00000400
-3:	.long	0xffff7fff
-4:	.long	0xfffffbff
-#endif
-5:	.long	0xa4150020 /* STBCR */
-6:	.long   0xfe40001c /* RTCOR */
-7:	.long   0xfe400018 /* RTCNT */
-8:	.long   0xa55a0000
-
-
-/* interrupt vector @ 0x600 */
-	.balign 	0x400,0,0x400
-	.long	0xdeadbeef
-	.balign 	0x200,0,0x200
-	bra	restore_jump_vbr
+restore_register:
+	add	#SH_SLEEP_BASE_DATA, r0
+	mov.l	@(r0, r5), r1
+	add	#-SH_SLEEP_BASE_DATA, r0
+	add	#SH_SLEEP_BASE_ADDR, r0
+	mov.l	@(r0, r5), r0
+	mov.l	r1, @r0
+	rts
 	 nop
-sh_mobile_standby_end:
 
-ENTRY(sh_mobile_standby_size)
-	.long sh_mobile_standby_end - sh_mobile_standby
+	.balign 4
+1:	.long	~0x7ff
+ENTRY(sh_mobile_sleep_resume_end)
diff --git a/arch/sh/kernel/cpu/ubc.S b/arch/sh/kernel/cpu/ubc.S
deleted file mode 100644
index 81923079fa12..000000000000
--- a/arch/sh/kernel/cpu/ubc.S
+++ /dev/null
@@ -1,59 +0,0 @@
-/*
- * arch/sh/kernel/cpu/ubc.S
- *
- * Set of management routines for the User Break Controller (UBC)
- *
- * Copyright (C) 2002 Paul Mundt
- *
- * This program is free software; you can redistribute it and/or modify it
- * under the terms of the GNU General Public License as published by the
- * Free Software Foundation; either version 2 of the License, or (at your
- * option) any later version.
- */
-#include <linux/linkage.h>
-#include <asm/ubc.h>
-
-#define STBCR2		0xffc00010
-
-ENTRY(ubc_sleep)
-	mov	#0, r0
-
-	mov.l	1f, r1		! Zero out UBC_BBRA ..
-	mov.w	r0, @r1
-
-	mov.l	2f, r1		! .. same for BBRB ..
-	mov.w	r0, @r1
-
-	mov.l	3f, r1		! .. and again for BRCR.
-	mov.w	r0, @r1
-
-	mov.w	@r1, r0		! Dummy read BRCR
-
-	mov.l	4f, r1		! Set MSTP5 in STBCR2
-	mov.b	@r1, r0
-	or	#0x01, r0
-	mov.b	r0, @r1
-
-	mov.b	@r1, r0		! Two dummy reads ..
-	mov.b	@r1, r0
-
-	rts
-	nop
-
-ENTRY(ubc_wakeup)
-	mov.l	4f, r1		! Clear MSTP5
-	mov.b	@r1, r0
-	and	#0xfe, r0
-	mov.b	r0, @r1
-
-	mov.b	@r1, r0		! Two more dummy reads ..
-	mov.b	@r1, r0
-
-	rts
-	nop
-
-1:	.long	UBC_BBRA
-2:	.long	UBC_BBRB
-3:	.long	UBC_BRCR
-4:	.long	STBCR2
-