Merge commit 'perf/core' into perf/hw-breakpoint

Conflicts:
	kernel/Makefile
	kernel/trace/Makefile
	kernel/trace/trace.h
	samples/Makefile

Merge reason: We need to be uptodate with the perf events development
branch because we plan to rewrite the breakpoints API on top of
perf events.

36 files changed, 4149 insertions, 335 deletions

diff --git a/arch/sh/kernel/cpu/Makefile b/arch/sh/kernel/cpu/Makefile
index eecad7cbd61..3d6b9312dc4 100644
--- a/arch/sh/kernel/cpu/Makefile
+++ b/arch/sh/kernel/cpu/Makefile
@@ -19,4 +19,4 @@ obj-$(CONFIG_UBC_WAKEUP)	+= ubc.o
 obj-$(CONFIG_SH_ADC)		+= adc.o
 obj-$(CONFIG_SH_CLK_CPG)	+= clock-cpg.o
 
-obj-y	+= irq/ init.o clock.o
+obj-y	+= irq/ init.o clock.o hwblk.o
diff --git a/arch/sh/kernel/cpu/hwblk.c b/arch/sh/kernel/cpu/hwblk.c
new file mode 100644
index 00000000000..c0ad7d46e78
--- /dev/null
+++ b/arch/sh/kernel/cpu/hwblk.c
@@ -0,0 +1,155 @@
+#include <linux/clk.h>
+#include <linux/compiler.h>
+#include <linux/slab.h>
+#include <linux/io.h>
+#include <linux/spinlock.h>
+#include <asm/suspend.h>
+#include <asm/hwblk.h>
+#include <asm/clock.h>
+
+static DEFINE_SPINLOCK(hwblk_lock);
+
+static void hwblk_area_mod_cnt(struct hwblk_info *info,
+			       int area, int counter, int value, int goal)
+{
+	struct hwblk_area *hap = info->areas + area;
+
+	hap->cnt[counter] += value;
+
+	if (hap->cnt[counter] != goal)
+		return;
+
+	if (hap->flags & HWBLK_AREA_FLAG_PARENT)
+		hwblk_area_mod_cnt(info, hap->parent, counter, value, goal);
+}
+
+
+static int __hwblk_mod_cnt(struct hwblk_info *info, int hwblk,
+			  int counter, int value, int goal)
+{
+	struct hwblk *hp = info->hwblks + hwblk;
+
+	hp->cnt[counter] += value;
+	if (hp->cnt[counter] == goal)
+		hwblk_area_mod_cnt(info, hp->area, counter, value, goal);
+
+	return hp->cnt[counter];
+}
+
+static void hwblk_mod_cnt(struct hwblk_info *info, int hwblk,
+			  int counter, int value, int goal)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&hwblk_lock, flags);
+	__hwblk_mod_cnt(info, hwblk, counter, value, goal);
+	spin_unlock_irqrestore(&hwblk_lock, flags);
+}
+
+void hwblk_cnt_inc(struct hwblk_info *info, int hwblk, int counter)
+{
+	hwblk_mod_cnt(info, hwblk, counter, 1, 1);
+}
+
+void hwblk_cnt_dec(struct hwblk_info *info, int hwblk, int counter)
+{
+	hwblk_mod_cnt(info, hwblk, counter, -1, 0);
+}
+
+void hwblk_enable(struct hwblk_info *info, int hwblk)
+{
+	struct hwblk *hp = info->hwblks + hwblk;
+	unsigned long tmp;
+	unsigned long flags;
+	int ret;
+
+	spin_lock_irqsave(&hwblk_lock, flags);
+
+	ret = __hwblk_mod_cnt(info, hwblk, HWBLK_CNT_USAGE, 1, 1);
+	if (ret == 1) {
+		tmp = __raw_readl(hp->mstp);
+		tmp &= ~(1 << hp->bit);
+		__raw_writel(tmp, hp->mstp);
+	}
+
+	spin_unlock_irqrestore(&hwblk_lock, flags);
+}
+
+void hwblk_disable(struct hwblk_info *info, int hwblk)
+{
+	struct hwblk *hp = info->hwblks + hwblk;
+	unsigned long tmp;
+	unsigned long flags;
+	int ret;
+
+	spin_lock_irqsave(&hwblk_lock, flags);
+
+	ret = __hwblk_mod_cnt(info, hwblk, HWBLK_CNT_USAGE, -1, 0);
+	if (ret == 0) {
+		tmp = __raw_readl(hp->mstp);
+		tmp |= 1 << hp->bit;
+		__raw_writel(tmp, hp->mstp);
+	}
+
+	spin_unlock_irqrestore(&hwblk_lock, flags);
+}
+
+struct hwblk_info *hwblk_info;
+
+int __init hwblk_register(struct hwblk_info *info)
+{
+	hwblk_info = info;
+	return 0;
+}
+
+int __init __weak arch_hwblk_init(void)
+{
+	return 0;
+}
+
+int __weak arch_hwblk_sleep_mode(void)
+{
+	return SUSP_SH_SLEEP;
+}
+
+int __init hwblk_init(void)
+{
+	return arch_hwblk_init();
+}
+
+/* allow clocks to enable and disable hardware blocks */
+static int sh_hwblk_clk_enable(struct clk *clk)
+{
+	if (!hwblk_info)
+		return -ENOENT;
+
+	hwblk_enable(hwblk_info, clk->arch_flags);
+	return 0;
+}
+
+static void sh_hwblk_clk_disable(struct clk *clk)
+{
+	if (hwblk_info)
+		hwblk_disable(hwblk_info, clk->arch_flags);
+}
+
+static struct clk_ops sh_hwblk_clk_ops = {
+	.enable		= sh_hwblk_clk_enable,
+	.disable	= sh_hwblk_clk_disable,
+	.recalc		= followparent_recalc,
+};
+
+int __init sh_hwblk_clk_register(struct clk *clks, int nr)
+{
+	struct clk *clkp;
+	int ret = 0;
+	int k;
+
+	for (k = 0; !ret && (k < nr); k++) {
+		clkp = clks + k;
+		clkp->ops = &sh_hwblk_clk_ops;
+		ret |= clk_register(clkp);
+	}
+
+	return ret;
+}
diff --git a/arch/sh/kernel/cpu/init.c b/arch/sh/kernel/cpu/init.c
index ad85421099c..e932ebef473 100644
--- a/arch/sh/kernel/cpu/init.c
+++ b/arch/sh/kernel/cpu/init.c
@@ -3,7 +3,7 @@
  *
  * CPU init code
  *
- * Copyright (C) 2002 - 2007  Paul Mundt
+ * Copyright (C) 2002 - 2009  Paul Mundt
  * Copyright (C) 2003  Richard Curnow
  *
  * This file is subject to the terms and conditions of the GNU General Public
@@ -62,6 +62,37 @@ static void __init speculative_execution_init(void)
 #define speculative_execution_init()	do { } while (0)
 #endif
 
+#ifdef CONFIG_CPU_SH4A
+#define EXPMASK			0xff2f0004
+#define EXPMASK_RTEDS		(1 << 0)
+#define EXPMASK_BRDSSLP		(1 << 1)
+#define EXPMASK_MMCAW		(1 << 4)
+
+static void __init expmask_init(void)
+{
+	unsigned long expmask = __raw_readl(EXPMASK);
+
+	/*
+	 * Future proofing.
+	 *
+	 * Disable support for slottable sleep instruction
+	 * and non-nop instructions in the rte delay slot.
+	 */
+	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;
+
+	__raw_writel(expmask, EXPMASK);
+	ctrl_barrier();
+}
+#else
+#define expmask_init()	do { } while (0)
+#endif
+
 /* 2nd-level cache init */
 void __uses_jump_to_uncached __attribute__ ((weak)) l2_cache_init(void)
 {
@@ -268,11 +299,9 @@ asmlinkage void __init sh_cpu_init(void)
 	cache_init();
 
 	if (raw_smp_processor_id() == 0) {
-#ifdef CONFIG_MMU
 		shm_align_mask = max_t(unsigned long,
 				       current_cpu_data.dcache.way_size - 1,
 				       PAGE_SIZE - 1);
-#endif
 
 		/* Boot CPU sets the cache shape */
 		detect_cache_shape();
@@ -321,4 +350,5 @@ asmlinkage void __init sh_cpu_init(void)
 #endif
 
 	speculative_execution_init();
+	expmask_init();
 }
diff --git a/arch/sh/kernel/cpu/irq/ipr.c b/arch/sh/kernel/cpu/irq/ipr.c
index 808d99a48ef..c1508a90fc6 100644
--- a/arch/sh/kernel/cpu/irq/ipr.c
+++ b/arch/sh/kernel/cpu/irq/ipr.c
@@ -35,6 +35,7 @@ static void disable_ipr_irq(unsigned int irq)
 	unsigned long addr = get_ipr_desc(irq)->ipr_offsets[p->ipr_idx];
 	/* Set the priority in IPR to 0 */
 	__raw_writew(__raw_readw(addr) & (0xffff ^ (0xf << p->shift)), addr);
+	(void)__raw_readw(addr);	/* Read back to flush write posting */
 }
 
 static void enable_ipr_irq(unsigned int irq)
diff --git a/arch/sh/kernel/cpu/sh2/entry.S b/arch/sh/kernel/cpu/sh2/entry.S
index becc54c4569..c8a4331d9b8 100644
--- a/arch/sh/kernel/cpu/sh2/entry.S
+++ b/arch/sh/kernel/cpu/sh2/entry.S
@@ -227,8 +227,9 @@ ENTRY(sh_bios_handler)
 	mov.l	@r15+, r14
 	add	#8,r15
 	lds.l	@r15+, pr
+	mov.l	@r15+,r15
 	rte
-	 mov.l	@r15+,r15
+	 nop
 	.align	2
 1:	.long	gdb_vbr_vector
 #endif /* CONFIG_SH_STANDARD_BIOS */
diff --git a/arch/sh/kernel/cpu/sh2/probe.c b/arch/sh/kernel/cpu/sh2/probe.c
index 5916d9096b9..1db6d888388 100644
--- a/arch/sh/kernel/cpu/sh2/probe.c
+++ b/arch/sh/kernel/cpu/sh2/probe.c
@@ -29,6 +29,7 @@ int __init detect_cpu_and_cache_system(void)
 	 */
 	boot_cpu_data.dcache.flags |= SH_CACHE_COMBINED;
 	boot_cpu_data.icache = boot_cpu_data.dcache;
+	boot_cpu_data.family = CPU_FAMILY_SH2;
 
 	return 0;
 }
diff --git a/arch/sh/kernel/cpu/sh2a/entry.S b/arch/sh/kernel/cpu/sh2a/entry.S
index ab3903eeda5..222742ddc0d 100644
--- a/arch/sh/kernel/cpu/sh2a/entry.S
+++ b/arch/sh/kernel/cpu/sh2a/entry.S
@@ -176,8 +176,9 @@ ENTRY(sh_bios_handler)
 	movml.l	@r15+,r14
 	add	#8,r15
 	lds.l	@r15+, pr
+	mov.l	@r15+,r15
 	rte
-	 mov.l	@r15+,r15
+	 nop
 	.align	2
 1:	.long	gdb_vbr_vector
 #endif /* CONFIG_SH_STANDARD_BIOS */
diff --git a/arch/sh/kernel/cpu/sh2a/probe.c b/arch/sh/kernel/cpu/sh2a/probe.c
index e098e2f6aa0..6825d650716 100644
--- a/arch/sh/kernel/cpu/sh2a/probe.c
+++ b/arch/sh/kernel/cpu/sh2a/probe.c
@@ -15,6 +15,8 @@
 
 int __init detect_cpu_and_cache_system(void)
 {
+	boot_cpu_data.family			= CPU_FAMILY_SH2A;
+
 	/* All SH-2A CPUs have support for 16 and 32-bit opcodes.. */
 	boot_cpu_data.flags			|= CPU_HAS_OP32;
 
diff --git a/arch/sh/kernel/cpu/sh3/clock-sh7709.c b/arch/sh/kernel/cpu/sh3/clock-sh7709.c
index fa30b601773..e8749505bd2 100644
--- a/arch/sh/kernel/cpu/sh3/clock-sh7709.c
+++ b/arch/sh/kernel/cpu/sh3/clock-sh7709.c
@@ -22,13 +22,6 @@ static int stc_multipliers[] = { 1, 2, 4, 8, 3, 6, 1, 1 };
 static int ifc_divisors[]    = { 1, 2, 4, 1, 3, 1, 1, 1 };
 static int pfc_divisors[]    = { 1, 2, 4, 1, 3, 6, 1, 1 };
 
-static void set_bus_parent(struct clk *clk)
-{
-	struct clk *bus_clk = clk_get(NULL, "bus_clk");
-	clk->parent = bus_clk;
-	clk_put(bus_clk);
-}
-
 static void master_clk_init(struct clk *clk)
 {
 	int frqcr = ctrl_inw(FRQCR);
@@ -50,9 +43,6 @@ static unsigned long module_clk_recalc(struct clk *clk)
 }
 
 static struct clk_ops sh7709_module_clk_ops = {
-#ifdef CLOCK_MODE_0_1_2_7
-	.init		= set_bus_parent,
-#endif
 	.recalc		= module_clk_recalc,
 };
 
@@ -78,7 +68,6 @@ static unsigned long cpu_clk_recalc(struct clk *clk)
 }
 
 static struct clk_ops sh7709_cpu_clk_ops = {
-	.init		= set_bus_parent,
 	.recalc		= cpu_clk_recalc,
 };
 
diff --git a/arch/sh/kernel/cpu/sh3/entry.S b/arch/sh/kernel/cpu/sh3/entry.S
index 3cb531f233f..bb407ef0b91 100644
--- a/arch/sh/kernel/cpu/sh3/entry.S
+++ b/arch/sh/kernel/cpu/sh3/entry.S
@@ -53,10 +53,6 @@
  *	syscall #
  *
  */
-#if defined(CONFIG_KGDB)
-NMI_VEC = 0x1c0			! Must catch early for debounce
-#endif
-
 /* Offsets to the stack */
 OFF_R0  =  0		/* Return value. New ABI also arg4 */
 OFF_R1  =  4     	/* New ABI: arg5 */
@@ -71,7 +67,6 @@ OFF_PC  =  (16*4)
 OFF_SR	=  (16*4+8)
 OFF_TRA	=  (16*4+6*4)
 
-
 #define k0	r0
 #define k1	r1
 #define k2	r2
@@ -113,34 +108,34 @@ OFF_TRA	=  (16*4+6*4)
 #if defined(CONFIG_MMU)
 	.align	2
 ENTRY(tlb_miss_load)
-	bra	call_dpf
+	bra	call_handle_tlbmiss
 	 mov	#0, r5
 
 	.align	2
 ENTRY(tlb_miss_store)
-	bra	call_dpf
+	bra	call_handle_tlbmiss
 	 mov	#1, r5
 
 	.align	2
 ENTRY(initial_page_write)
-	bra	call_dpf
-	 mov	#1, r5
+	bra	call_handle_tlbmiss
+	 mov	#2, r5
 
 	.align	2
 ENTRY(tlb_protection_violation_load)
-	bra	call_dpf
+	bra	call_do_page_fault
 	 mov	#0, r5
 
 	.align	2
 ENTRY(tlb_protection_violation_store)
-	bra	call_dpf
+	bra	call_do_page_fault
 	 mov	#1, r5
 
-call_dpf:
+call_handle_tlbmiss:
+	setup_frame_reg
 	mov.l	1f, r0
 	mov	r5, r8
 	mov.l	@r0, r6
-	mov	r6, r9
 	mov.l	2f, r0
 	sts	pr, r10
 	jsr	@r0
@@ -151,16 +146,23 @@ call_dpf:
 	 lds	r10, pr
 	rts
 	 nop
-0:	mov.l	3f, r0
-	mov	r9, r6
+0:
 	mov	r8, r5
+call_do_page_fault:
+	mov.l	1f, r0
+	mov.l	@r0, r6
+
+	mov.l	3f, r0
+	mov.l	4f, r1
+	mov	r15, r4
 	jmp	@r0
-	 mov	r15, r4
+	 lds	r1, pr
 
 	.align 2
 1:	.long	MMU_TEA
-2:	.long	__do_page_fault
+2:	.long	handle_tlbmiss
 3:	.long	do_page_fault
+4:	.long	ret_from_exception
 
 	.align	2
 ENTRY(address_error_load)
@@ -256,7 +258,7 @@ restore_all:
 	!
 	! Calculate new SR value
 	mov	k3, k2			! original SR value
-	mov	#0xf0, k1
+	mov	#0xfffffff0, k1
 	extu.b	k1, k1
 	not	k1, k1
 	and	k1, k2			! Mask original SR value
@@ -272,21 +274,12 @@ restore_all:
 6:	or	k0, k2			! Set the IMASK-bits
 	ldc	k2, ssr
 	!
-#if defined(CONFIG_KGDB)
-	! Clear in_nmi
-	mov.l	6f, k0
-	mov	#0, k1
-	mov.b	k1, @k0
-#endif
 	mov	k4, r15
 	rte
 	 nop
 
 	.align	2
 5:	.long	0x00001000	! DSP
-#ifdef CONFIG_KGDB
-6:	.long	in_nmi
-#endif
 7:	.long	0x30000000
 
 ! common exception handler
@@ -478,23 +471,6 @@ ENTRY(save_low_regs)
 !
 	.balign 	512,0,512
 ENTRY(handle_interrupt)
-#if defined(CONFIG_KGDB)
-	mov.l	2f, k2
-	! Debounce (filter nested NMI)
-	mov.l	@k2, k0
-	mov.l	9f, k1
-	cmp/eq	k1, k0
-	bf	11f
-	mov.l	10f, k1
-	tas.b	@k1
-	bt	11f
-	rte
-	 nop
-	.align	2
-9:	.long	NMI_VEC
-10:	.long	in_nmi
-11:
-#endif /* defined(CONFIG_KGDB) */
 	sts	pr, k3		! save original pr value in k3
 	mova	exception_data, k0
 
@@ -507,13 +483,49 @@ ENTRY(handle_interrupt)
 	bsr	save_regs	! needs original pr value in k3
 	 mov	#-1, k2		! default vector kept in k2
 
+	setup_frame_reg
+
+	stc	sr, r0	! get status register
+	shlr2	r0
+	and	#0x3c, r0
+	cmp/eq	#0x3c, r0
+	bf	9f
+	TRACE_IRQS_OFF
+9:
+
 	! Setup return address and jump to do_IRQ
 	mov.l	4f, r9		! fetch return address
 	lds	r9, pr		! put return address in pr
 	mov.l	2f, r4
 	mov.l	3f, r9
 	mov.l	@r4, r4		! pass INTEVT vector as arg0
+
+	shlr2	r4
+	shlr	r4
+	mov	r4, r0		! save vector->jmp table offset for later
+
+	shlr2	r4		! vector to IRQ# conversion
+	add	#-0x10, r4
+
+	cmp/pz	r4		! is it a valid IRQ?
+	bt	10f
+
+	/*
+	 * We got here as a result of taking the INTEVT path for something
+	 * that isn't a valid hard IRQ, therefore we bypass the do_IRQ()
+	 * path and special case the event dispatch instead.  This is the
+	 * expected path for the NMI (and any other brilliantly implemented
+	 * exception), which effectively wants regular exception dispatch
+	 * but is unfortunately reported through INTEVT rather than
+	 * EXPEVT.  Grr.
+	 */
+	mov.l	6f, r9
+	mov.l	@(r0, r9), r9
 	jmp	@r9
+	 mov	r15, r8		! trap handlers take saved regs in r8
+
+10:
+	jmp	@r9		! Off to do_IRQ() we go.
 	 mov	r15, r5		! pass saved registers as arg1
 
 ENTRY(exception_none)
diff --git a/arch/sh/kernel/cpu/sh3/ex.S b/arch/sh/kernel/cpu/sh3/ex.S
index e5a0de39a2d..46610c35c23 100644
--- a/arch/sh/kernel/cpu/sh3/ex.S
+++ b/arch/sh/kernel/cpu/sh3/ex.S
@@ -48,9 +48,7 @@ ENTRY(exception_handling_table)
 	.long	system_call	! Unconditional Trap	 /* 160 */
 	.long	exception_error	! reserved_instruction (filled by trap_init) /* 180 */
 	.long	exception_error	! illegal_slot_instruction (filled by trap_init) /*1A0*/
-ENTRY(nmi_slot)
-	.long	kgdb_handle_exception	/* 1C0 */	! Allow trap to debugger
-ENTRY(user_break_point_trap)
+	.long	nmi_trap_handler	/* 1C0 */	! Allow trap to debugger
 	.long	break_point_trap	/* 1E0 */
 
 	/*
diff --git a/arch/sh/kernel/cpu/sh3/probe.c b/arch/sh/kernel/cpu/sh3/probe.c
index 10f2a760c5e..f9c7df64eb0 100644
--- a/arch/sh/kernel/cpu/sh3/probe.c
+++ b/arch/sh/kernel/cpu/sh3/probe.c
@@ -107,5 +107,7 @@ int __uses_jump_to_uncached detect_cpu_and_cache_system(void)
 	boot_cpu_data.dcache.flags |= SH_CACHE_COMBINED;
 	boot_cpu_data.icache = boot_cpu_data.dcache;
 
+	boot_cpu_data.family = CPU_FAMILY_SH3;
+
 	return 0;
 }
diff --git a/arch/sh/kernel/cpu/sh4/probe.c b/arch/sh/kernel/cpu/sh4/probe.c
index 6c78d0a9c85..d36f0c45f55 100644
--- a/arch/sh/kernel/cpu/sh4/probe.c
+++ b/arch/sh/kernel/cpu/sh4/probe.c
@@ -57,8 +57,12 @@ int __init detect_cpu_and_cache_system(void)
 	 * Setup some generic flags we can probe on SH-4A parts
 	 */
 	if (((pvr >> 16) & 0xff) == 0x10) {
-		if ((cvr & 0x10000000) == 0)
+		boot_cpu_data.family = CPU_FAMILY_SH4A;
+
+		if ((cvr & 0x10000000) == 0) {
 			boot_cpu_data.flags |= CPU_HAS_DSP;
+			boot_cpu_data.family = CPU_FAMILY_SH4AL_DSP;
+		}
 
 		boot_cpu_data.flags |= CPU_HAS_LLSC | CPU_HAS_PERF_COUNTER;
 		boot_cpu_data.cut_major = pvr & 0x7f;
@@ -68,6 +72,7 @@ int __init detect_cpu_and_cache_system(void)
 	} else {
 		/* And some SH-4 defaults.. */
 		boot_cpu_data.flags |= CPU_HAS_PTEA;
+		boot_cpu_data.family = CPU_FAMILY_SH4;
 	}
 
 	/* FPU detection works for everyone */
@@ -139,8 +144,15 @@ int __init detect_cpu_and_cache_system(void)
 		}
 		break;
 	case 0x300b:
-		boot_cpu_data.type = CPU_SH7724;
-		boot_cpu_data.flags |= CPU_HAS_L2_CACHE;
+		switch (prr) {
+		case 0x20:
+			boot_cpu_data.type = CPU_SH7724;
+			boot_cpu_data.flags |= CPU_HAS_L2_CACHE;
+			break;
+		case 0x50:
+			boot_cpu_data.type = CPU_SH7757;
+			break;
+		}
 		break;
 	case 0x4000:	/* 1st cut */
 	case 0x4001:	/* 2nd cut */
@@ -173,9 +185,6 @@ int __init detect_cpu_and_cache_system(void)
 		boot_cpu_data.dcache.ways = 2;
 
 		break;
-	default:
-		boot_cpu_data.type = CPU_SH_NONE;
-		break;
 	}
 
 	/*
diff --git a/arch/sh/kernel/cpu/sh4a/Makefile b/arch/sh/kernel/cpu/sh4a/Makefile
index ebdd391d5f4..490d5dc9e37 100644
--- a/arch/sh/kernel/cpu/sh4a/Makefile
+++ b/arch/sh/kernel/cpu/sh4a/Makefile
@@ -3,6 +3,7 @@
 #
 
 # CPU subtype setup
+obj-$(CONFIG_CPU_SUBTYPE_SH7757)	+= setup-sh7757.o
 obj-$(CONFIG_CPU_SUBTYPE_SH7763)	+= setup-sh7763.o
 obj-$(CONFIG_CPU_SUBTYPE_SH7770)	+= setup-sh7770.o
 obj-$(CONFIG_CPU_SUBTYPE_SH7780)	+= setup-sh7780.o
@@ -19,15 +20,16 @@ obj-$(CONFIG_CPU_SUBTYPE_SHX3)		+= setup-shx3.o
 smp-$(CONFIG_CPU_SHX3)			:= smp-shx3.o
 
 # Primary on-chip clocks (common)
+clock-$(CONFIG_CPU_SUBTYPE_SH7757)	:= clock-sh7757.o
 clock-$(CONFIG_CPU_SUBTYPE_SH7763)	:= clock-sh7763.o
 clock-$(CONFIG_CPU_SUBTYPE_SH7770)	:= clock-sh7770.o
 clock-$(CONFIG_CPU_SUBTYPE_SH7780)	:= clock-sh7780.o
 clock-$(CONFIG_CPU_SUBTYPE_SH7785)	:= clock-sh7785.o
 clock-$(CONFIG_CPU_SUBTYPE_SH7786)	:= clock-sh7786.o
 clock-$(CONFIG_CPU_SUBTYPE_SH7343)	:= clock-sh7343.o
-clock-$(CONFIG_CPU_SUBTYPE_SH7722)	:= clock-sh7722.o
-clock-$(CONFIG_CPU_SUBTYPE_SH7723)	:= clock-sh7723.o
-clock-$(CONFIG_CPU_SUBTYPE_SH7724)	:= clock-sh7724.o
+clock-$(CONFIG_CPU_SUBTYPE_SH7722)	:= clock-sh7722.o hwblk-sh7722.o
+clock-$(CONFIG_CPU_SUBTYPE_SH7723)	:= clock-sh7723.o hwblk-sh7723.o
+clock-$(CONFIG_CPU_SUBTYPE_SH7724)	:= clock-sh7724.o hwblk-sh7724.o
 clock-$(CONFIG_CPU_SUBTYPE_SH7366)	:= clock-sh7366.o
 clock-$(CONFIG_CPU_SUBTYPE_SHX3)	:= clock-shx3.o
 
@@ -35,6 +37,7 @@ clock-$(CONFIG_CPU_SUBTYPE_SHX3)	:= clock-shx3.o
 pinmux-$(CONFIG_CPU_SUBTYPE_SH7722)	:= pinmux-sh7722.o
 pinmux-$(CONFIG_CPU_SUBTYPE_SH7723)	:= pinmux-sh7723.o
 pinmux-$(CONFIG_CPU_SUBTYPE_SH7724)	:= pinmux-sh7724.o
+pinmux-$(CONFIG_CPU_SUBTYPE_SH7757)	:= pinmux-sh7757.o
 pinmux-$(CONFIG_CPU_SUBTYPE_SH7785)	:= pinmux-sh7785.o
 pinmux-$(CONFIG_CPU_SUBTYPE_SH7786)	:= pinmux-sh7786.o
 
diff --git a/arch/sh/kernel/cpu/sh4a/clock-sh7722.c b/arch/sh/kernel/cpu/sh4a/clock-sh7722.c
index 40f859354f7..ea38b554dc0 100644
--- a/arch/sh/kernel/cpu/sh4a/clock-sh7722.c
+++ b/arch/sh/kernel/cpu/sh4a/clock-sh7722.c
@@ -22,6 +22,8 @@
 #include <linux/kernel.h>
 #include <linux/io.h>
 #include <asm/clock.h>
+#include <asm/hwblk.h>
+#include <cpu/sh7722.h>
 
 /* SH7722 registers */
 #define FRQCR		0xa4150000
@@ -30,9 +32,6 @@
 #define SCLKBCR		0xa415000c
 #define IRDACLKCR	0xa4150018
 #define PLLCR		0xa4150024
-#define MSTPCR0		0xa4150030
-#define MSTPCR1		0xa4150034
-#define MSTPCR2		0xa4150038
 #define DLLFRQ		0xa4150050
 
 /* Fixed 32 KHz root clock for RTC and Power Management purposes */
@@ -140,35 +139,37 @@ struct clk div6_clks[] = {
 	SH_CLK_DIV6("video_clk", &pll_clk, VCLKCR, 0),
 };
 
-#define MSTP(_str, _parent, _reg, _bit, _flags) \
-  SH_CLK_MSTP32(_str, -1, _parent, _reg, _bit, _flags)
+#define R_CLK &r_clk
+#define P_CLK &div4_clks[DIV4_P]
+#define B_CLK &div4_clks[DIV4_B]
+#define U_CLK &div4_clks[DIV4_U]
 
 static struct clk mstp_clks[] = {
-	MSTP("uram0", &div4_clks[DIV4_U], MSTPCR0, 28, CLK_ENABLE_ON_INIT),
-	MSTP("xymem0", &div4_clks[DIV4_B], MSTPCR0, 26, CLK_ENABLE_ON_INIT),
-	MSTP("tmu0", &div4_clks[DIV4_P], MSTPCR0, 15, 0),
-	MSTP("cmt0", &r_clk, MSTPCR0, 14, 0),
-	MSTP("rwdt0", &r_clk, MSTPCR0, 13, 0),
-	MSTP("flctl0", &div4_clks[DIV4_P], MSTPCR0, 10, 0),
-	MSTP("scif0", &div4_clks[DIV4_P], MSTPCR0, 7, 0),
-	MSTP("scif1", &div4_clks[DIV4_P], MSTPCR0, 6, 0),
-	MSTP("scif2", &div4_clks[DIV4_P], MSTPCR0, 5, 0),
-
-	MSTP("i2c0", &div4_clks[DIV4_P], MSTPCR1, 9, 0),
-	MSTP("rtc0", &r_clk, MSTPCR1, 8, 0),
-
-	MSTP("sdhi0", &div4_clks[DIV4_P], MSTPCR2, 18, 0),
-	MSTP("keysc0", &r_clk, MSTPCR2, 14, 0),
-	MSTP("usbf0", &div4_clks[DIV4_P], MSTPCR2, 11, 0),
-	MSTP("2dg0", &div4_clks[DIV4_B], MSTPCR2, 9, 0),
-	MSTP("siu0", &div4_clks[DIV4_B], MSTPCR2, 8, 0),
-	MSTP("vou0", &div4_clks[DIV4_B], MSTPCR2, 5, 0),
-	MSTP("jpu0", &div4_clks[DIV4_B], MSTPCR2, 6, CLK_ENABLE_ON_INIT),
-	MSTP("beu0", &div4_clks[DIV4_B], MSTPCR2, 4, 0),
-	MSTP("ceu0", &div4_clks[DIV4_B], MSTPCR2, 3, 0),
-	MSTP("veu0", &div4_clks[DIV4_B], MSTPCR2, 2, CLK_ENABLE_ON_INIT),
-	MSTP("vpu0", &div4_clks[DIV4_B], MSTPCR2, 1, CLK_ENABLE_ON_INIT),
-	MSTP("lcdc0", &div4_clks[DIV4_B], MSTPCR2, 0, 0),
+	SH_HWBLK_CLK("uram0", -1, U_CLK, HWBLK_URAM, CLK_ENABLE_ON_INIT),
+	SH_HWBLK_CLK("xymem0", -1, B_CLK, HWBLK_XYMEM, CLK_ENABLE_ON_INIT),
+	SH_HWBLK_CLK("tmu0", -1, P_CLK, HWBLK_TMU, 0),
+	SH_HWBLK_CLK("cmt0", -1, R_CLK, HWBLK_CMT, 0),
+	SH_HWBLK_CLK("rwdt0", -1, R_CLK, HWBLK_RWDT, 0),
+	SH_HWBLK_CLK("flctl0", -1, P_CLK, HWBLK_FLCTL, 0),
+	SH_HWBLK_CLK("scif0", -1, P_CLK, HWBLK_SCIF0, 0),
+	SH_HWBLK_CLK("scif1", -1, P_CLK, HWBLK_SCIF1, 0),
+	SH_HWBLK_CLK("scif2", -1, P_CLK, HWBLK_SCIF2, 0),
+
+	SH_HWBLK_CLK("i2c0", -1, P_CLK, HWBLK_IIC, 0),
+	SH_HWBLK_CLK("rtc0", -1, R_CLK, HWBLK_RTC, 0),
+
+	SH_HWBLK_CLK("sdhi0", -1, P_CLK, HWBLK_SDHI, 0),
+	SH_HWBLK_CLK("keysc0", -1, R_CLK, HWBLK_KEYSC, 0),
+	SH_HWBLK_CLK("usbf0", -1, P_CLK, HWBLK_USBF, 0),
+	SH_HWBLK_CLK("2dg0", -1, B_CLK, HWBLK_2DG, 0),
+	SH_HWBLK_CLK("siu0", -1, B_CLK, HWBLK_SIU, 0),
+	SH_HWBLK_CLK("vou0", -1, B_CLK, HWBLK_VOU, 0),
+	SH_HWBLK_CLK("jpu0", -1, B_CLK, HWBLK_JPU, 0),
+	SH_HWBLK_CLK("beu0", -1, B_CLK, HWBLK_BEU, 0),
+	SH_HWBLK_CLK("ceu0", -1, B_CLK, HWBLK_CEU, 0),
+	SH_HWBLK_CLK("veu0", -1, B_CLK, HWBLK_VEU, 0),
+	SH_HWBLK_CLK("vpu0", -1, B_CLK, HWBLK_VPU, 0),
+	SH_HWBLK_CLK("lcdc0", -1, P_CLK, HWBLK_LCDC, 0),
 };
 
 int __init arch_clk_init(void)
@@ -191,7 +192,7 @@ int __init arch_clk_init(void)
 		ret = sh_clk_div6_register(div6_clks, ARRAY_SIZE(div6_clks));
 
 	if (!ret)
-		ret = sh_clk_mstp32_register(mstp_clks, ARRAY_SIZE(mstp_clks));
+		ret = sh_hwblk_clk_register(mstp_clks, ARRAY_SIZE(mstp_clks));
 
 	return ret;
 }
diff --git a/arch/sh/kernel/cpu/sh4a/clock-sh7723.c b/arch/sh/kernel/cpu/sh4a/clock-sh7723.c
index e67c2678b8a..20a31c2255a 100644
--- a/arch/sh/kernel/cpu/sh4a/clock-sh7723.c
+++ b/arch/sh/kernel/cpu/sh4a/clock-sh7723.c
@@ -22,6 +22,8 @@
 #include <linux/kernel.h>
 #include <linux/io.h>
 #include <asm/clock.h>
+#include <asm/hwblk.h>
+#include <cpu/sh7723.h>
 
 /* SH7723 registers */
 #define FRQCR		0xa4150000
@@ -30,9 +32,6 @@
 #define SCLKBCR		0xa415000c
 #define IRDACLKCR	0xa4150018
 #define PLLCR		0xa4150024
-#define MSTPCR0		0xa4150030
-#define MSTPCR1		0xa4150034
-#define MSTPCR2		0xa4150038
 #define DLLFRQ		0xa4150050
 
 /* Fixed 32 KHz root clock for RTC and Power Management purposes */
@@ -140,60 +139,64 @@ struct clk div6_clks[] = {
 	SH_CLK_DIV6("video_clk", &pll_clk, VCLKCR, 0),
 };
 
-#define MSTP(_str, _parent, _reg, _bit, _force_on, _need_cpg, _need_ram) \
-  SH_CLK_MSTP32(_str, -1, _parent, _reg, _bit, _force_on * CLK_ENABLE_ON_INIT)
+#define R_CLK (&r_clk)
+#define P_CLK (