Merge with mainline to remove plat-omap/Kconfig conflict

Conflicts:
	arch/arm/plat-omap/Kconfig

8 files changed, 70 insertions, 150 deletions

diff --git a/arch/sh/kernel/cpu/sh4/clock-sh4-202.c b/arch/sh/kernel/cpu/sh4/clock-sh4-202.c
index 21421e34e7d5..6b80850294da 100644
--- a/arch/sh/kernel/cpu/sh4/clock-sh4-202.c
+++ b/arch/sh/kernel/cpu/sh4/clock-sh4-202.c
@@ -23,7 +23,7 @@ static int frqcr3_values[]   = { 0, 1, 2, 3, 4, 5, 6  };
 
 static unsigned long emi_clk_recalc(struct clk *clk)
 {
-	int idx = ctrl_inl(CPG2_FRQCR3) & 0x0007;
+	int idx = __raw_readl(CPG2_FRQCR3) & 0x0007;
 	return clk->parent->rate / frqcr3_divisors[idx];
 }
 
@@ -52,7 +52,7 @@ static struct clk sh4202_emi_clk = {
 
 static unsigned long femi_clk_recalc(struct clk *clk)
 {
-	int idx = (ctrl_inl(CPG2_FRQCR3) >> 3) & 0x0007;
+	int idx = (__raw_readl(CPG2_FRQCR3) >> 3) & 0x0007;
 	return clk->parent->rate / frqcr3_divisors[idx];
 }
 
@@ -92,7 +92,7 @@ static void shoc_clk_init(struct clk *clk)
 
 static unsigned long shoc_clk_recalc(struct clk *clk)
 {
-	int idx = (ctrl_inl(CPG2_FRQCR3) >> 6) & 0x0007;
+	int idx = (__raw_readl(CPG2_FRQCR3) >> 6) & 0x0007;
 	return clk->parent->rate / frqcr3_divisors[idx];
 }
 
@@ -122,10 +122,10 @@ static int shoc_clk_set_rate(struct clk *clk, unsigned long rate, int algo_id)
 
 	tmp = frqcr3_lookup(clk, rate);
 
-	frqcr3 = ctrl_inl(CPG2_FRQCR3);
+	frqcr3 = __raw_readl(CPG2_FRQCR3);
 	frqcr3 &= ~(0x0007 << 6);
 	frqcr3 |= tmp << 6;
-	ctrl_outl(frqcr3, CPG2_FRQCR3);
+	__raw_writel(frqcr3, CPG2_FRQCR3);
 
 	clk->rate = clk->parent->rate / frqcr3_divisors[tmp];
 
diff --git a/arch/sh/kernel/cpu/sh4/clock-sh4.c b/arch/sh/kernel/cpu/sh4/clock-sh4.c
index 73294d9cd049..5add75c1f539 100644
--- a/arch/sh/kernel/cpu/sh4/clock-sh4.c
+++ b/arch/sh/kernel/cpu/sh4/clock-sh4.c
@@ -28,7 +28,7 @@ static int pfc_divisors[] = { 2, 3, 4, 6, 8, 2, 2, 2 };
 
 static void master_clk_init(struct clk *clk)
 {
-	clk->rate *= pfc_divisors[ctrl_inw(FRQCR) & 0x0007];
+	clk->rate *= pfc_divisors[__raw_readw(FRQCR) & 0x0007];
 }
 
 static struct clk_ops sh4_master_clk_ops = {
@@ -37,7 +37,7 @@ static struct clk_ops sh4_master_clk_ops = {
 
 static unsigned long module_clk_recalc(struct clk *clk)
 {
-	int idx = (ctrl_inw(FRQCR) & 0x0007);
+	int idx = (__raw_readw(FRQCR) & 0x0007);
 	return clk->parent->rate / pfc_divisors[idx];
 }
 
@@ -47,7 +47,7 @@ static struct clk_ops sh4_module_clk_ops = {
 
 static unsigned long bus_clk_recalc(struct clk *clk)
 {
-	int idx = (ctrl_inw(FRQCR) >> 3) & 0x0007;
+	int idx = (__raw_readw(FRQCR) >> 3) & 0x0007;
 	return clk->parent->rate / bfc_divisors[idx];
 }
 
@@ -57,7 +57,7 @@ static struct clk_ops sh4_bus_clk_ops = {
 
 static unsigned long cpu_clk_recalc(struct clk *clk)
 {
-	int idx = (ctrl_inw(FRQCR) >> 6) & 0x0007;
+	int idx = (__raw_readw(FRQCR) >> 6) & 0x0007;
 	return clk->parent->rate / ifc_divisors[idx];
 }
 
diff --git a/arch/sh/kernel/cpu/sh4/fpu.c b/arch/sh/kernel/cpu/sh4/fpu.c
index e97857aec8a0..447482d7f65e 100644
--- a/arch/sh/kernel/cpu/sh4/fpu.c
+++ b/arch/sh/kernel/cpu/sh4/fpu.c
@@ -85,14 +85,14 @@ void save_fpu(struct task_struct *tsk)
 		      "fmov.s	fr1, @-%0\n\t"
 		      "fmov.s	fr0, @-%0\n\t"
 		      "lds	%3, fpscr\n\t":"=r" (dummy)
-		      :"0"((char *)(&tsk->thread.fpu.hard.status)),
+		      :"0"((char *)(&tsk->thread.xstate->hardfpu.status)),
 		      "r"(FPSCR_RCHG), "r"(FPSCR_INIT)
 		      :"memory");
 
 	disable_fpu();
 }
 
-static void restore_fpu(struct task_struct *tsk)
+void restore_fpu(struct task_struct *tsk)
 {
 	unsigned long dummy;
 
@@ -135,62 +135,11 @@ static void restore_fpu(struct task_struct *tsk)
 		      "lds.l	@%0+, fpscr\n\t"
 		      "lds.l	@%0+, fpul\n\t"
 		      :"=r" (dummy)
-		      :"0"(&tsk->thread.fpu), "r"(FPSCR_RCHG)
+		      :"0" (tsk->thread.xstate), "r" (FPSCR_RCHG)
 		      :"memory");
 	disable_fpu();
 }
 
-/*
- * Load the FPU with signalling NANS.  This bit pattern we're using
- * has the property that no matter wether considered as single or as
- * double precision represents signaling NANS.
- */
-
-static void fpu_init(void)
-{
-	enable_fpu();
-	asm volatile (	"lds	%0, fpul\n\t"
-			"lds	%1, fpscr\n\t"
-			"fsts	fpul, fr0\n\t"
-			"fsts	fpul, fr1\n\t"
-			"fsts	fpul, fr2\n\t"
-			"fsts	fpul, fr3\n\t"
-			"fsts	fpul, fr4\n\t"
-			"fsts	fpul, fr5\n\t"
-			"fsts	fpul, fr6\n\t"
-			"fsts	fpul, fr7\n\t"
-			"fsts	fpul, fr8\n\t"
-			"fsts	fpul, fr9\n\t"
-			"fsts	fpul, fr10\n\t"
-			"fsts	fpul, fr11\n\t"
-			"fsts	fpul, fr12\n\t"
-			"fsts	fpul, fr13\n\t"
-			"fsts	fpul, fr14\n\t"
-			"fsts	fpul, fr15\n\t"
-			"frchg\n\t"
-			"fsts	fpul, fr0\n\t"
-			"fsts	fpul, fr1\n\t"
-			"fsts	fpul, fr2\n\t"
-			"fsts	fpul, fr3\n\t"
-			"fsts	fpul, fr4\n\t"
-			"fsts	fpul, fr5\n\t"
-			"fsts	fpul, fr6\n\t"
-			"fsts	fpul, fr7\n\t"
-			"fsts	fpul, fr8\n\t"
-			"fsts	fpul, fr9\n\t"
-			"fsts	fpul, fr10\n\t"
-			"fsts	fpul, fr11\n\t"
-			"fsts	fpul, fr12\n\t"
-			"fsts	fpul, fr13\n\t"
-			"fsts	fpul, fr14\n\t"
-			"fsts	fpul, fr15\n\t"
-			"frchg\n\t"
-			"lds	%2, fpscr\n\t"
-			:	/* no output */
-			:"r" (0), "r"(FPSCR_RCHG), "r"(FPSCR_INIT));
-	disable_fpu();
-}
-
 /**
  *      denormal_to_double - Given denormalized float number,
  *                           store double float
@@ -282,9 +231,9 @@ static int ieee_fpe_handler(struct pt_regs *regs)
 		/* fcnvsd */
 		struct task_struct *tsk = current;
 
-		if ((tsk->thread.fpu.hard.fpscr & FPSCR_CAUSE_ERROR))
+		if ((tsk->thread.xstate->hardfpu.fpscr & FPSCR_CAUSE_ERROR))
 			/* FPU error */
-			denormal_to_double(&tsk->thread.fpu.hard,
+			denormal_to_double(&tsk->thread.xstate->hardfpu,
 					   (finsn >> 8) & 0xf);
 		else
 			return 0;
@@ -300,9 +249,9 @@ static int ieee_fpe_handler(struct pt_regs *regs)
 
 		n = (finsn >> 8) & 0xf;
 		m = (finsn >> 4) & 0xf;
-		hx = tsk->thread.fpu.hard.fp_regs[n];
-		hy = tsk->thread.fpu.hard.fp_regs[m];
-		fpscr = tsk->thread.fpu.hard.fpscr;
+		hx = tsk->thread.xstate->hardfpu.fp_regs[n];
+		hy = tsk->thread.xstate->hardfpu.fp_regs[m];
+		fpscr = tsk->thread.xstate->hardfpu.fpscr;
 		prec = fpscr & FPSCR_DBL_PRECISION;
 
 		if ((fpscr & FPSCR_CAUSE_ERROR)
@@ -312,18 +261,18 @@ static int ieee_fpe_handler(struct pt_regs *regs)
 
 			/* FPU error because of denormal (doubles) */
 			llx = ((long long)hx << 32)
-			    | tsk->thread.fpu.hard.fp_regs[n + 1];
+			    | tsk->thread.xstate->hardfpu.fp_regs[n + 1];
 			lly = ((long long)hy << 32)
-			    | tsk->thread.fpu.hard.fp_regs[m + 1];
+			    | tsk->thread.xstate->hardfpu.fp_regs[m + 1];
 			llx = float64_mul(llx, lly);
-			tsk->thread.fpu.hard.fp_regs[n] = llx >> 32;
-			tsk->thread.fpu.hard.fp_regs[n + 1] = llx & 0xffffffff;
+			tsk->thread.xstate->hardfpu.fp_regs[n] = llx >> 32;
+			tsk->thread.xstate->hardfpu.fp_regs[n + 1] = llx & 0xffffffff;
 		} else if ((fpscr & FPSCR_CAUSE_ERROR)
 			   && (!prec && ((hx & 0x7fffffff) < 0x00800000
 					 || (hy & 0x7fffffff) < 0x00800000))) {
 			/* FPU error because of denormal (floats) */
 			hx = float32_mul(hx, hy);
-			tsk->thread.fpu.hard.fp_regs[n] = hx;
+			tsk->thread.xstate->hardfpu.fp_regs[n] = hx;
 		} else
 			return 0;
 
@@ -338,9 +287,9 @@ static int ieee_fpe_handler(struct pt_regs *regs)
 
 		n = (finsn >> 8) & 0xf;
 		m = (finsn >> 4) & 0xf;
-		hx = tsk->thread.fpu.hard.fp_regs[n];
-		hy = tsk->thread.fpu.hard.fp_regs[m];
-		fpscr = tsk->thread.fpu.hard.fpscr;
+		hx = tsk->thread.xstate->hardfpu.fp_regs[n];
+		hy = tsk->thread.xstate->hardfpu.fp_regs[m];
+		fpscr = tsk->thread.xstate->hardfpu.fpscr;
 		prec = fpscr & FPSCR_DBL_PRECISION;
 
 		if ((fpscr & FPSCR_CAUSE_ERROR)
@@ -350,15 +299,15 @@ static int ieee_fpe_handler(struct pt_regs *regs)
 
 			/* FPU error because of denormal (doubles) */
 			llx = ((long long)hx << 32)
-			    | tsk->thread.fpu.hard.fp_regs[n + 1];
+			    | tsk->thread.xstate->hardfpu.fp_regs[n + 1];
 			lly = ((long long)hy << 32)
-			    | tsk->thread.fpu.hard.fp_regs[m + 1];
+			    | tsk->thread.xstate->hardfpu.fp_regs[m + 1];
 			if ((finsn & 0xf00f) == 0xf000)
 				llx = float64_add(llx, lly);
 			else
 				llx = float64_sub(llx, lly);
-			tsk->thread.fpu.hard.fp_regs[n] = llx >> 32;
-			tsk->thread.fpu.hard.fp_regs[n + 1] = llx & 0xffffffff;
+			tsk->thread.xstate->hardfpu.fp_regs[n] = llx >> 32;
+			tsk->thread.xstate->hardfpu.fp_regs[n + 1] = llx & 0xffffffff;
 		} else if ((fpscr & FPSCR_CAUSE_ERROR)
 			   && (!prec && ((hx & 0x7fffffff) < 0x00800000
 					 || (hy & 0x7fffffff) < 0x00800000))) {
@@ -367,7 +316,7 @@ static int ieee_fpe_handler(struct pt_regs *regs)
 				hx = float32_add(hx, hy);
 			else
 				hx = float32_sub(hx, hy);
-			tsk->thread.fpu.hard.fp_regs[n] = hx;
+			tsk->thread.xstate->hardfpu.fp_regs[n] = hx;
 		} else
 			return 0;
 
@@ -382,9 +331,9 @@ static int ieee_fpe_handler(struct pt_regs *regs)
 
 		n = (finsn >> 8) & 0xf;
 		m = (finsn >> 4) & 0xf;
-		hx = tsk->thread.fpu.hard.fp_regs[n];
-		hy = tsk->thread.fpu.hard.fp_regs[m];
-		fpscr = tsk->thread.fpu.hard.fpscr;
+		hx = tsk->thread.xstate->hardfpu.fp_regs[n];
+		hy = tsk->thread.xstate->hardfpu.fp_regs[m];
+		fpscr = tsk->thread.xstate->hardfpu.fpscr;
 		prec = fpscr & FPSCR_DBL_PRECISION;
 
 		if ((fpscr & FPSCR_CAUSE_ERROR)
@@ -394,20 +343,20 @@ static int ieee_fpe_handler(struct pt_regs *regs)
 
 			/* FPU error because of denormal (doubles) */
 			llx = ((long long)hx << 32)
-			    | tsk->thread.fpu.hard.fp_regs[n + 1];
+			    | tsk->thread.xstate->hardfpu.fp_regs[n + 1];
 			lly = ((long long)hy << 32)
-			    | tsk->thread.fpu.hard.fp_regs[m + 1];
+			    | tsk->thread.xstate->hardfpu.fp_regs[m + 1];
 
 			llx = float64_div(llx, lly);
 
-			tsk->thread.fpu.hard.fp_regs[n] = llx >> 32;
-			tsk->thread.fpu.hard.fp_regs[n + 1] = llx & 0xffffffff;
+			tsk->thread.xstate->hardfpu.fp_regs[n] = llx >> 32;
+			tsk->thread.xstate->hardfpu.fp_regs[n + 1] = llx & 0xffffffff;
 		} else if ((fpscr & FPSCR_CAUSE_ERROR)
 			   && (!prec && ((hx & 0x7fffffff) < 0x00800000
 					 || (hy & 0x7fffffff) < 0x00800000))) {
 			/* FPU error because of denormal (floats) */
 			hx = float32_div(hx, hy);
-			tsk->thread.fpu.hard.fp_regs[n] = hx;
+			tsk->thread.xstate->hardfpu.fp_regs[n] = hx;
 		} else
 			return 0;
 
@@ -420,17 +369,17 @@ static int ieee_fpe_handler(struct pt_regs *regs)
 		unsigned int hx;
 
 		m = (finsn >> 8) & 0x7;
-		hx = tsk->thread.fpu.hard.fp_regs[m];
+		hx = tsk->thread.xstate->hardfpu.fp_regs[m];
 
-		if ((tsk->thread.fpu.hard.fpscr & FPSCR_CAUSE_ERROR)
+		if ((tsk->thread.xstate->hardfpu.fpscr & FPSCR_CAUSE_ERROR)
 			&& ((hx & 0x7fffffff) < 0x00100000)) {
 			/* subnormal double to float conversion */
 			long long llx;
 
-			llx = ((long long)tsk->thread.fpu.hard.fp_regs[m] << 32)
-			    | tsk->thread.fpu.hard.fp_regs[m + 1];
+			llx = ((long long)tsk->thread.xstate->hardfpu.fp_regs[m] << 32)
+			    | tsk->thread.xstate->hardfpu.fp_regs[m + 1];
 
-			tsk->thread.fpu.hard.fpul = float64_to_float32(llx);
+			tsk->thread.xstate->hardfpu.fpul = float64_to_float32(llx);
 		} else
 			return 0;
 
@@ -449,7 +398,7 @@ void float_raise(unsigned int flags)
 int float_rounding_mode(void)
 {
 	struct task_struct *tsk = current;
-	int roundingMode = FPSCR_ROUNDING_MODE(tsk->thread.fpu.hard.fpscr);
+	int roundingMode = FPSCR_ROUNDING_MODE(tsk->thread.xstate->hardfpu.fpscr);
 	return roundingMode;
 }
 
@@ -461,16 +410,16 @@ BUILD_TRAP_HANDLER(fpu_error)
 	__unlazy_fpu(tsk, regs);
 	fpu_exception_flags = 0;
 	if (ieee_fpe_handler(regs)) {
-		tsk->thread.fpu.hard.fpscr &=
+		tsk->thread.xstate->hardfpu.fpscr &=
 		    ~(FPSCR_CAUSE_MASK | FPSCR_FLAG_MASK);
-		tsk->thread.fpu.hard.fpscr |= fpu_exception_flags;
+		tsk->thread.xstate->hardfpu.fpscr |= fpu_exception_flags;
 		/* Set the FPSCR flag as well as cause bits - simply
 		 * replicate the cause */
-		tsk->thread.fpu.hard.fpscr |= (fpu_exception_flags >> 10);
+		tsk->thread.xstate->hardfpu.fpscr |= (fpu_exception_flags >> 10);
 		grab_fpu(regs);
 		restore_fpu(tsk);
 		task_thread_info(tsk)->status |= TS_USEDFPU;
-		if ((((tsk->thread.fpu.hard.fpscr & FPSCR_ENABLE_MASK) >> 7) &
+		if ((((tsk->thread.xstate->hardfpu.fpscr & FPSCR_ENABLE_MASK) >> 7) &
 		     (fpu_exception_flags >> 2)) == 0) {
 			return;
 		}
@@ -478,33 +427,3 @@ BUILD_TRAP_HANDLER(fpu_error)
 
 	force_sig(SIGFPE, tsk);
 }
-
-void fpu_state_restore(struct pt_regs *regs)
-{
-	struct task_struct *tsk = current;
-
-	grab_fpu(regs);
-	if (unlikely(!user_mode(regs))) {
-		printk(KERN_ERR "BUG: FPU is used in kernel mode.\n");
-		BUG();
-		return;
-	}
-
-	if (likely(used_math())) {
-		/* Using the FPU again.  */
-		restore_fpu(tsk);
-	} else {
-		/* First time FPU user.  */
-		fpu_init();
-		set_used_math();
-	}
-	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/probe.c b/arch/sh/kernel/cpu/sh4/probe.c
index d36f0c45f55f..822977a06d84 100644
--- a/arch/sh/kernel/cpu/sh4/probe.c
+++ b/arch/sh/kernel/cpu/sh4/probe.c
@@ -28,9 +28,9 @@ int __init detect_cpu_and_cache_system(void)
 		[9] = (1 << 16)
 	};
 
-	pvr = (ctrl_inl(CCN_PVR) >> 8) & 0xffffff;
-	prr = (ctrl_inl(CCN_PRR) >> 4) & 0xff;
-	cvr = (ctrl_inl(CCN_CVR));
+	pvr = (__raw_readl(CCN_PVR) >> 8) & 0xffffff;
+	prr = (__raw_readl(CCN_PRR) >> 4) & 0xff;
+	cvr = (__raw_readl(CCN_CVR));
 
 	/*
 	 * Setup some sane SH-4 defaults for the icache
@@ -71,11 +71,11 @@ int __init detect_cpu_and_cache_system(void)
 		boot_cpu_data.dcache.ways = 4;
 	} else {
 		/* And some SH-4 defaults.. */
-		boot_cpu_data.flags |= CPU_HAS_PTEA;
+		boot_cpu_data.flags |= CPU_HAS_PTEA | CPU_HAS_FPU;
 		boot_cpu_data.family = CPU_FAMILY_SH4;
 	}
 
-	/* FPU detection works for everyone */
+	/* FPU detection works for almost everyone */
 	if ((cvr & 0x20000000))
 		boot_cpu_data.flags |= CPU_HAS_FPU;
 
@@ -124,6 +124,7 @@ int __init detect_cpu_and_cache_system(void)
 		boot_cpu_data.type = CPU_SH7785;
 		break;
 	case 0x4004:
+	case 0x4005:
 		boot_cpu_data.type = CPU_SH7786;
 		boot_cpu_data.flags |= CPU_HAS_PTEAEX | CPU_HAS_L2_CACHE;
 		break;
@@ -160,6 +161,7 @@ int __init detect_cpu_and_cache_system(void)
 		break;
 	case 0x700:
 		boot_cpu_data.type = CPU_SH4_501;
+		boot_cpu_data.flags &= ~CPU_HAS_FPU;
 		boot_cpu_data.icache.ways = 2;
 		boot_cpu_data.dcache.ways = 2;
 		break;
@@ -227,7 +229,7 @@ int __init detect_cpu_and_cache_system(void)
 			 * Size calculation is much more sensible
 			 * than it is for the L1.
 			 *
-			 * Sizes are 128KB, 258KB, 512KB, and 1MB.
+			 * Sizes are 128KB, 256KB, 512KB, and 1MB.
 			 */
 			size = (cvr & 0xf) << 17;
 
diff --git a/arch/sh/kernel/cpu/sh4/setup-sh4-202.c b/arch/sh/kernel/cpu/sh4/setup-sh4-202.c
index 4b733715cdb5..b9b7e10ad68f 100644
--- a/arch/sh/kernel/cpu/sh4/setup-sh4-202.c
+++ b/arch/sh/kernel/cpu/sh4/setup-sh4-202.c
@@ -198,7 +198,7 @@ void __init plat_irq_setup_pins(int mode)
 {
 	switch (mode) {
 	case IRQ_MODE_IRQ: /* individual interrupt mode for IRL3-0 */
-		ctrl_outw(ctrl_inw(INTC_ICR) | INTC_ICR_IRLM, INTC_ICR);
+		__raw_writew(__raw_readw(INTC_ICR) | INTC_ICR_IRLM, INTC_ICR);
 		register_intc_controller(&intc_desc_irlm);
 		break;
 	default:
diff --git a/arch/sh/kernel/cpu/sh4/setup-sh7750.c b/arch/sh/kernel/cpu/sh4/setup-sh7750.c
index b2a9df1af64c..ffd79e57254f 100644
--- a/arch/sh/kernel/cpu/sh4/setup-sh7750.c
+++ b/arch/sh/kernel/cpu/sh4/setup-sh7750.c
@@ -442,7 +442,7 @@ void __init plat_irq_setup_pins(int mode)
 
 	switch (mode) {
 	case IRQ_MODE_IRQ: /* individual interrupt mode for IRL3-0 */
-		ctrl_outw(ctrl_inw(INTC_ICR) | INTC_ICR_IRLM, INTC_ICR);
+		__raw_writew(__raw_readw(INTC_ICR) | INTC_ICR_IRLM, INTC_ICR);
 		register_intc_controller(&intc_desc_irlm);
 		break;
 	default:
diff --git a/arch/sh/kernel/cpu/sh4/setup-sh7760.c b/arch/sh/kernel/cpu/sh4/setup-sh7760.c
index 5b74cc0b43da..a16eb3656f4b 100644
--- a/arch/sh/kernel/cpu/sh4/setup-sh7760.c
+++ b/arch/sh/kernel/cpu/sh4/setup-sh7760.c
@@ -319,7 +319,7 @@ void __init plat_irq_setup_pins(int mode)
 {
 	switch (mode) {
 	case IRQ_MODE_IRQ:
-		ctrl_outw(ctrl_inw(INTC_ICR) | INTC_ICR_IRLM, INTC_ICR);
+		__raw_writew(__raw_readw(INTC_ICR) | INTC_ICR_IRLM, INTC_ICR);
 		register_intc_controller(&intc_desc_irq);
 		break;
 	default:
diff --git a/arch/sh/kernel/cpu/sh4/sq.c b/arch/sh/kernel/cpu/sh4/sq.c
index 8a8a993f55ea..fc065f9da6e5 100644
--- a/arch/sh/kernel/cpu/sh4/sq.c
+++ b/arch/sh/kernel/cpu/sh4/sq.c
@@ -43,9 +43,9 @@ static unsigned long *sq_bitmap;
 
 #define store_queue_barrier()			\
 do {						\
-	(void)ctrl_inl(P4SEG_STORE_QUE);	\
-	ctrl_outl(0, P4SEG_STORE_QUE + 0);	\
-	ctrl_outl(0, P4SEG_STORE_QUE + 8);	\
+	(void)__raw_readl(P4SEG_STORE_QUE);	\
+	__raw_writel(0, P4SEG_STORE_QUE + 0);	\
+	__raw_writel(0, P4SEG_STORE_QUE + 8);	\
 } while (0);
 
 /**
@@ -100,7 +100,7 @@ static inline void sq_mapping_list_del(struct sq_mapping *map)
 	spin_unlock_irq(&sq_mapping_lock);
 }
 
-static int __sq_remap(struct sq_mapping *map, unsigned long flags)
+static int __sq_remap(struct sq_mapping *map, pgprot_t prot)
 {
 #if defined(CONFIG_MMU)
 	struct vm_struct *vma;
@@ -113,7 +113,7 @@ static int __sq_remap(struct sq_mapping *map, unsigned long flags)
 
 	if (ioremap_page_range((unsigned long)vma->addr,
 			       (unsigned long)vma->addr + map->size,
-			       vma->phys_addr, __pgprot(flags))) {
+			       vma->phys_addr, prot)) {
 		vunmap(vma->addr);
 		return -EAGAIN;
 	}
@@ -123,8 +123,8 @@ static int __sq_remap(struct sq_mapping *map, unsigned long flags)
 	 * straightforward, as we can just load up each queue's QACR with
 	 * the physical address appropriately masked.
 	 */
-	ctrl_outl(((map->addr >> 26) << 2) & 0x1c, SQ_QACR0);
-	ctrl_outl(((map->addr >> 26) << 2) & 0x1c, SQ_QACR1);
+	__raw_writel(((map->addr >> 26) << 2) & 0x1c, SQ_QACR0);
+	__raw_writel(((map->addr >> 26) << 2) & 0x1c, SQ_QACR1);
 #endif
 
 	return 0;
@@ -135,14 +135,14 @@ static int __sq_remap(struct sq_mapping *map, unsigned long flags)
  * @phys: Physical address of mapping.
  * @size: Length of mapping.
  * @name: User invoking mapping.
- * @flags: Protection flags.
+ * @prot: Protection bits.
  *
  * Remaps the physical address @phys through the next available store queue
  * address of @size length. @name is logged at boot time as well as through
  * the sysfs interface.
  */
 unsigned long sq_remap(unsigned long phys, unsigned int size,
-		       const char *name, unsigned long flags)
+		       const char *name, pgprot_t prot)
 {
 	struct sq_mapping *map;
 	unsigned long end;
@@ -177,7 +177,7 @@ unsigned long sq_remap(unsigned long phys, unsigned int size,
 
 	map->sq_addr = P4SEG_STORE_QUE + (page << PAGE_SHIFT);
 
-	ret = __sq_remap(map, pgprot_val(PAGE_KERNEL_NOCACHE) | flags);
+	ret = __sq_remap(map, prot);
 	if (unlikely(ret != 0))
 		goto out;
 
@@ -309,8 +309,7 @@ static ssize_t mapping_store(const char *buf, size_t count)
 		return -EIO;
 
 	if (likely(len)) {
-		int ret = sq_remap(base, len, "Userspace",
-				   pgprot_val(PAGE_SHARED));
+		int ret = sq_remap(base, len, "Userspace", PAGE_SHARED);
 		if (ret < 0)
 			return ret;
 	} else