Merge bk://bk.arm.linux.org.uk/linux-2.6-rmk

into ppc970.osdl.org:/home/torvalds/v2.6/linux

Documentation/arm/Sharp-LH/SDRAM

+README on the SDRAM Controller for the LH7a40X
+==============================================
+
+The standard configuration for the SDRAM controller generates a sparse
+memory array.  The precise layout is determined by the SDRAM chips.  A
+default kernel configuration assembles the discontiguous memory
+regions into separate memory nodes via the NUMA (Non-Uniform Memory
+Architecture) facilities.  In this default configuration, the kernel
+is forgiving about the precise layout.  As long as it is given an
+accurate picture of available memory by the bootloader the kernel will
+execute correctly.
+
+The SDRC supports a mode where some of the chip select lines are
+swapped in order to make SDRAM look like a synchronous ROM.  Setting
+this bit means that the RAM will present as a contiguous array.  Some
+programmers prefer this to the discontiguous layout.  Be aware that
+may be a penalty for this feature where some some configurations of
+memory are significantly reduced; i.e. 64MiB of RAM appears as only 32
+MiB.
+
+There are a couple of configuration options to override the default
+behavior.  When the SROMLL bit is set and memory appears as a
+contiguous array, there is no reason to support NUMA.
+CONFIG_LH7A40X_CONTIGMEM disables NUMA support.  When physical memory
+is discontiguous, the memory tables are organized such that there are
+two banks per nodes with a small gap between them.  This layout wastes
+some kernel memory for page tables representing non-existent memory.
+CONFIG_LH7A40X_ONE_BANK_PER_NODE optimizes the node tables such that
+there are no gaps.  These options control the low level organization
+of the memory management tables in ways that may prevent the kernel
+from booting or may cause the kernel to allocated excessively large
+page tables.  Be warned.  Only change these options if you know what
+you are doing.  The default behavior is a reasonable compromise that
+will suit all users.
+
+--
+
+A typical 32MiB system with the default configuration options will
+find physical memory managed as follows.
+
+   node 0: 0xc0000000 4MiB
+           0xc1000000 4MiB
+   node 1: 0xc4000000 4MiB
+           0xc5000000 4MiB
+   node 2: 0xc8000000 4MiB
+           0xc9000000 4MiB
+   node 3: 0xcc000000 4MiB
+           0xcd000000 4MiB
+
+Setting CONFIG_LH7A40X_ONE_BANK_PER_NODE will put each bank into a
+separate node.

Documentation/arm/VFP/release-notes.txt

+Release notes for Linux Kernel VFP support code
+-----------------------------------------------
+
+Date: 	20 May 2004
+Author:	Russell King
+
+This is the first release of the Linux Kernel VFP support code.  It
+provides support for the exceptions bounced from VFP hardware found
+on ARM926EJ-S.
+
+This release has been validated against the SoftFloat-2b library by
+John R. Hauser using the TestFloat-2a test suite.  Details of this
+library and test suite can be found at:
+
+   http://www.cs.berkeley.edu/~jhauser/arithmetic/SoftFloat.html
+
+The operations which have been tested with this package are:
+
+ - fdiv
+ - fsub
+ - fadd
+ - fmul
+ - fcmp
+ - fcmpe
+ - fcvtd
+ - fcvts
+ - fsito
+ - ftosi
+ - fsqrt
+
+All the above pass softfloat tests with the following exceptions:
+
+- fadd/fsub shows some differences in the handling of +0 / -0 results
+  when input operands differ in signs.
+- the handling of underflow exceptions is slightly different.  If a
+  result underflows before rounding, but becomes a normalised number
+  after rounding, we do not signal an underflow exception.
+
+Other operations which have been tested by basic assembly-only tests
+are:
+
+ - fcpy
+ - fabs
+ - fneg
+ - ftoui
+ - ftosiz
+ - ftouiz
+
+The combination operations have not been tested:
+
+ - fmac
+ - fnmac
+ - fmsc
+ - fnmsc
+ - fnmul

arch/arm/Kconfig

@@ -246,10 +246,10 @@ menu "General setup"
 # Select various configuration options depending on the machine type
 config DISCONTIGMEM
 	bool
-	depends on ARCH_EDB7211 || ARCH_SA1100 || ARCH_LH7A40X
+	depends on ARCH_EDB7211 || ARCH_SA1100 || (ARCH_LH7A40X && !LH7A40X_SROMLL)
 	default y
 	help
-	  Say Y to upport efficient handling of discontiguous physical memory,
+	  Say Y to support efficient handling of discontiguous physical memory,
 	  for architectures which are either NUMA (Non-Uniform Memory Access)
 	  or have huge holes in the physical address space for other reasons.
 	  See <file:Documentation/vm/numa> for more.
@@ -421,6 +421,17 @@ config FPE_FASTFPE
 	  If you do not feel you need a faster FP emulation you should better
 	  choose NWFPE.
 
+config VFP
+	bool "VFP-format floating point maths"
+	help
+	  Say Y to include VFP support code in the kernel. This is needed
+	  if your hardware includes a VFP unit.
+
+	  Please see <file:Documentation/arm/VFP/release-notes.txt> for
+	  release notes and additional status information.
+
+	  Say N if your target does not have VFP hardware.
+
 source "fs/Kconfig.binfmt"
 
 source "drivers/base/Kconfig"

arch/arm/Makefile

@@ -119,6 +119,7 @@ core-y				+= arch/arm/mach-$(machine-y)/
 endif
 core-$(CONFIG_FPE_NWFPE)	+= arch/arm/nwfpe/
 core-$(CONFIG_FPE_FASTFPE)	+= $(FASTFPE_OBJ)
+core-$(CONFIG_VFP)		+= arch/arm/vfp/
 
 drivers-$(CONFIG_OPROFILE)      += arch/arm/oprofile/
 drivers-$(CONFIG_ARCH_CLPS7500)	+= drivers/acorn/char/

arch/arm/kernel/debug.S

@@ -591,7 +591,8 @@
 		.macro  addruart,rx
 		mrc     p15, 0, \rx, c1, c0
 		tst     \rx, #1                 @ MMU enabled?
-		ldr     \rx, =0x80000700        @ physical base address
+		mov     \rx, #0x00000700        @ offset from base
+		orreq   \rx, \rx, #0x80000000   @ physical base
 		orrne   \rx, \rx, #0xf8000000   @ virtual base
 		.endm
 

arch/arm/kernel/entry-armv.S

@@ -19,6 +19,7 @@
 #include <asm/thread_info.h>
 #include <asm/glue.h>
 #include <asm/ptrace.h>
+#include <asm/vfpmacros.h>
 
 #include "entry-header.S"
 
@@ -1198,8 +1199,13 @@ call_fpe:	enable_irq r10				@ Enable interrupts
 		mov	pc, lr				@ CP#7
 		mov	pc, lr				@ CP#8
 		mov	pc, lr				@ CP#9
+#ifdef CONFIG_VFP
+		b	do_vfp				@ CP#10 (VFP)
+		b	do_vfp				@ CP#11 (VFP)
+#else
 		mov	pc, lr				@ CP#10 (VFP)
 		mov	pc, lr				@ CP#11 (VFP)
+#endif
 		mov	pc, lr				@ CP#12
 		mov	pc, lr				@ CP#13
 		mov	pc, lr				@ CP#14 (Debug)
@@ -1260,6 +1266,13 @@ ENTRY(__switch_to)
 		ldr	r3, [r2, #TI_CPU_DOMAIN]!
 		stmia	ip, {r4 - sl, fp, sp, lr}	@ Store most regs on stack
 		mcr	p15, 0, r3, c3, c0, 0		@ Set domain register
+#ifdef CONFIG_VFP
+		@ Always disable VFP so we can lazily save/restore the old
+		@ state. This occurs in the context of the previous thread.
+		VFPFMRX	r4, FPEXC
+		bic	r4, r4, #FPEXC_ENABLE
+		VFPFMXR	FPEXC, r4
+#endif
 		ldmib	r2, {r4 - sl, fp, sp, pc}	@ Load all regs saved previously
 
 		__INIT

arch/arm/kernel/process.c

@@ -314,10 +314,16 @@ void flush_thread(void)
 	memset(thread->used_cp, 0, sizeof(thread->used_cp));
 	memset(&tsk->thread.debug, 0, sizeof(struct debug_info));
 	fp_init(&thread->fpstate);
+#if defined(CONFIG_VFP)
+	vfp_flush_thread(&thread->vfpstate);
+#endif
 }
 
 void release_thread(struct task_struct *dead_task)
 {
+#if defined(CONFIG_VFP)
+	vfp_release_thread(&dead_task->thread_info->vfpstate);
+#endif
 }
 
 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");

arch/arm/mach-integrator/Makefile

@@ -4,7 +4,7 @@
 
 # Object file lists.
 
-obj-y					:= core.o lm.o time.o
+obj-y					:= clock.o core.o lm.o time.o
 obj-$(CONFIG_ARCH_INTEGRATOR_AP)	+= integrator_ap.o
 obj-$(CONFIG_ARCH_INTEGRATOR_CP)	+= integrator_cp.o
 

arch/arm/mach-integrator/clock.c

+/*
+ *  linux/arch/arm/mach-integrator/clock.c
+ *
+ *  Copyright (C) 2004 ARM Limited.
+ *  Written by Deep Blue Solutions Limited.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/errno.h>
+#include <linux/err.h>
+
+#include <asm/semaphore.h>
+#include <asm/hardware/clock.h>
+#include <asm/hardware/icst525.h>
+
+#include "clock.h"
+
+static LIST_HEAD(clocks);
+static DECLARE_MUTEX(clocks_sem);
+
+struct clk *clk_get(struct device *dev, const char *id)
+{
+	struct clk *p, *clk = ERR_PTR(-ENOENT);
+
+	down(&clocks_sem);
+	list_for_each_entry(p, &clocks, node) {
+		if (strcmp(id, p->name) == 0 && try_module_get(p->owner)) {
+			clk = p;
+			break;
+		}
+	}
+	up(&clocks_sem);
+
+	return clk;
+}
+EXPORT_SYMBOL(clk_get);
+
+void clk_put(struct clk *clk)
+{
+	module_put(clk->owner);
+}
+EXPORT_SYMBOL(clk_put);
+
+int clk_enable(struct clk *clk)
+{
+	return 0;
+}
+EXPORT_SYMBOL(clk_enable);
+
+void clk_disable(struct clk *clk)
+{
+}
+EXPORT_SYMBOL(clk_disable);
+
+int clk_use(struct clk *clk)
+{
+	return 0;
+}
+EXPORT_SYMBOL(clk_use);
+
+void clk_unuse(struct clk *clk)
+{
+}
+EXPORT_SYMBOL(clk_unuse);
+
+unsigned long clk_get_rate(struct clk *clk)
+{
+	return clk->rate;
+}
+EXPORT_SYMBOL(clk_get_rate);
+
+long clk_round_rate(struct clk *clk, unsigned long rate)
+{
+	return rate;
+}
+EXPORT_SYMBOL(clk_round_rate);
+
+int clk_set_rate(struct clk *clk, unsigned long rate)
+{
+	int ret = -EIO;
+	if (clk->setvco) {
+		struct icst525_vco vco;
+
+		vco = icst525_khz_to_vco(clk->params, rate);
+		clk->rate = icst525_khz(clk->params, vco);
+
+		printk("Clock %s: setting VCO reg params: S=%d R=%d V=%d\n",
+			clk->name, vco.s, vco.r, vco.v);
+
+		clk->setvco(clk, vco);
+		ret = 0;
+	}
+	return 0;
+}
+EXPORT_SYMBOL(clk_set_rate);
+
+/*
+ * These are fixed clocks.
+ */
+static struct clk kmi_clk = {
+	.name	= "KMIREFCLK",
+	.rate	= 24000000,
+};
+
+static struct clk uart_clk = {
+	.name	= "UARTCLK",
+	.rate	= 14745600,
+};
+
+int clk_register(struct clk *clk)
+{
+	down(&clocks_sem);
+	list_add(&clk->node, &clocks);
+	up(&clocks_sem);
+	return 0;
+}
+EXPORT_SYMBOL(clk_register);
+
+void clk_unregister(struct clk *clk)
+{
+	down(&clocks_sem);
+	list_del(&clk->node);
+	up(&clocks_sem);
+}
+EXPORT_SYMBOL(clk_unregister);
+
+static int __init clk_init(void)
+{
+	clk_register(&kmi_clk);
+	clk_register(&uart_clk);
+	return 0;
+}
+arch_initcall(clk_init);

arch/arm/mach-integrator/clock.h

+/*
+ *  linux/arch/arm/mach-integrator/clock.h
+ *
+ *  Copyright (C) 2004 ARM Limited.
+ *  Written by Deep Blue Solutions Limited.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+struct module;
+struct icst525_params;
+
+struct clk {
+	struct list_head	node;
+	unsigned long		rate;
+	struct module		*owner;
+	const char		*name;
+	const struct icst525_params *params;
+	void			*data;
+	void			(*setvco)(struct clk *, struct icst525_vco vco);
+};
+
+int clk_register(struct clk *clk);
+void clk_unregister(struct clk *clk);

arch/arm/mach-integrator/impd1.c

@@ -25,6 +25,8 @@
 #include <asm/arch/impd1.h>
 #include <asm/sizes.h>
 
+#include "clock.h"
+
 static int module_id;
 
 module_param_named(lmid, module_id, int, 0444);
@@ -32,6 +34,7 @@ MODULE_PARM_DESC(lmid, "logic module stack position");
 
 struct impd1_module {
 	void		*base;
+	struct clk	vcos[2];
 };
 
 static const struct icst525_params impd1_vco_params = {
@@ -43,25 +46,20 @@ static const struct icst525_params impd1_vco_params = {
 	.rd_max		= 120,
 };
 
-void impd1_set_vco(struct device *dev, int vconr, unsigned long period)
+static void impd1_setvco(struct clk *clk, struct icst525_vco vco)
 {
-	struct impd1_module *impd1 = dev_get_drvdata(dev);
-	struct icst525_vco vco;
+	struct impd1_module *impd1 = clk->data;
+	int vconr = clk - impd1->vcos;
 	u32 val;
 
-	vco = icst525_ps_to_vco(&impd1_vco_params, period);
-
-	pr_debug("Guessed VCO reg params: S=%d R=%d V=%d\n",
-		vco.s, vco.r, vco.v);
-
 	val = vco.v | (vco.r << 9) | (vco.s << 16);
 
 	writel(0xa05f, impd1->base + IMPD1_LOCK);
 	switch (vconr) {
-	case 1:
+	case 0:
 		writel(val, impd1->base + IMPD1_OSC1);
 		break;
-	case 2:
+	case 1:
 		writel(val, impd1->base + IMPD1_OSC2);
 		break;
 	}
@@ -77,8 +75,6 @@ void impd1_set_vco(struct device *dev, int vconr, unsigned long period)
 #endif
 }
 
-EXPORT_SYMBOL(impd1_set_vco);
-
 void impd1_tweak_control(struct device *dev, u32 mask, u32 val)
 {
 	struct impd1_module *impd1 = dev_get_drvdata(dev);
@@ -140,6 +136,11 @@ static struct impd1_device impd1_devs[] = {
 	}
 };
 
+static const char *impd1_vconames[2] = {
+	"CLCDCLK",
+	"AUXVCO2",
+};
+
 static int impd1_probe(struct lm_device *dev)
 {
 	struct impd1_module *impd1;
@@ -168,6 +169,16 @@ static int impd1_probe(struct lm_device *dev)
 
 	printk("IM-PD1 found at 0x%08lx\n", dev->resource.start);
 
+	for (i = 0; i < ARRAY_SIZE(impd1->vcos); i++) {
+		impd1->vcos[i].owner = THIS_MODULE,
+		impd1->vcos[i].name = impd1_vconames[i],
+		impd1->vcos[i].params = &impd1_vco_params,
+		impd1->vcos[i].data = impd1,
+		impd1->vcos[i].setvco = impd1_setvco;
+
+		clk_register(&impd1->vcos[i]);
+	}
+
 	for (i = 0; i < ARRAY_SIZE(impd1_devs); i++) {
 		struct impd1_device *idev = impd1_devs + i;
 		struct amba_device *d;
@@ -216,6 +227,7 @@ static void impd1_remove(struct lm_device *dev)
 {
 	struct impd1_module *impd1 = lm_get_drvdata(dev);
 	struct list_head *l, *n;
+	int i;
 
 	list_for_each_safe(l, n, &dev->dev.children) {
 		struct device *d = list_to_dev(l);
@@ -223,6 +235,9 @@ static void impd1_remove(struct lm_device *dev)
 		device_unregister(d);
 	}
 
+	for (i = 0; i < ARRAY_SIZE(impd1->vcos); i++)
+		clk_unregister(&impd1->vcos[i]);
+
 	lm_set_drvdata(dev, NULL);
 
 	iounmap(impd1->base);

arch/arm/mach-integrator/integrator_cp.c

@@ -23,6 +23,7 @@
 #include <asm/mach-types.h>
 #include <asm/hardware/amba.h>
 #include <asm/hardware/amba_kmi.h>
+#include <asm/hardware/icst525.h>
 
 #include <asm/arch/lm.h>
 
@@ -32,12 +33,16 @@
 #include <asm/mach/mmc.h>
 #include <asm/mach/map.h>
 
+#include "clock.h"
+
 #define INTCP_PA_MMC_BASE		0x1c000000
 #define INTCP_PA_AACI_BASE		0x1d000000
 
 #define INTCP_PA_FLASH_BASE		0x24000000
 #define INTCP_FLASH_SIZE		SZ_32M
 
+#define INTCP_PA_CLCD_BASE		0xc0000000
+
 #define INTCP_VA_CIC_BASE		0xf1000040
 #define INTCP_VA_PIC_BASE		0xf1400000
 #define INTCP_VA_SIC_BASE		0xfca00000
@@ -209,6 +214,44 @@ static void __init intcp_init_irq(void)
 	pic_unmask_irq(IRQ_CP_CPPLDINT);
 }
 
+/*
+ * Clock handling
+ */
+#define CM_LOCK (IO_ADDRESS(INTEGRATOR_HDR_BASE)+INTEGRATOR_HDR_LOCK_OFFSET)
+#define CM_AUXOSC (IO_ADDRESS(INTEGRATOR_HDR_BASE)+0x1c)
+
+static const struct icst525_params cp_auxvco_params = {
+	.ref		= 24000,
+	.vco_max	= 320000,
+	.vd_min 	= 8,
+	.vd_max 	= 263,
+	.rd_min 	= 3,
+	.rd_max 	= 65,
+};
+
+static void cp_auxvco_set(struct clk *clk, struct icst525_vco vco)
+{
+	u32 val;
+
+	val = readl(CM_AUXOSC) & ~0x7ffff;
+	val |= vco.v | (vco.r << 9) | (vco.s << 16);
+
+	writel(0xa05f, CM_LOCK);
+	writel(val, CM_AUXOSC);
+	writel(0, CM_LOCK);
+}
+
+static struct clk cp_clcd_clk = {
+	.name	= "CLCDCLK",
+	.params	= &cp_auxvco_params,
+	.setvco = cp_auxvco_set,
+};
+
+static struct clk cp_mmci_clk = {
+	.name	= "MCLK",
+	.rate	= 33000000,
+};
+
 /*
  * Flash handling.
  */
@@ -340,15 +383,34 @@ static struct amba_device aaci_device = {
 	.periphid	= 0,
 };
 
+static struct amba_device clcd_device = {
+	.dev		= {
+		.bus_id	= "mb:c0",
+		.coherent_dma_mask = ~0,
+	},
+	.res		= {
+		.start	= INTCP_PA_CLCD_BASE,
+		.end	= INTCP_PA_CLCD_BASE + SZ_4K - 1,
+		.flags	= IORESOURCE_MEM,
+	},
+	.dma_mask	= ~0,
+	.irq		= { IRQ_CP_CLCDCINT, NO_IRQ },
+	.periphid	= 0,
+};
+
 static struct amba_device *amba_devs[] __initdata = {
 	&mmc_device,
 	&aaci_device,
+	&clcd_device,
 };
 
 static void __init intcp_init(void)
 {
 	int i;
 
+	clk_register(&cp_clcd_clk);
+	clk_register(&cp_mmci_clk);
+
 	platform_add_devices(intcp_devs, ARRAY_SIZE(intcp_devs));
 
 	for (i = 0; i < ARRAY_SIZE(amba_devs); i++) {

arch/arm/mach-lh7a40x/Kconfig

@@ -34,6 +34,37 @@ config ARCH_LH7A400
 config ARCH_LH7A404
 	bool
 
+config LH7A40X_CONTIGMEM
+	bool "Disable NUMA Support"
+	depends on ARCH_LH7A40X
+	help
+	  Say Y here if your bootloader sets the SROMLL bit(s) in
+	  the SDRAM controller, organizing memory as a contiguous
+	  array.  This option will disable CONFIG_DISCONTIGMEM and
+          force the kernel to manage all memory in one node.
+
+	  Setting this option incorrectly may prevent the kernel from
+	  booting.  It is OK to leave it N.
+
+	  For more information, consult
+	    <file:Documentation/arm/Sharp-LH/SDRAM>.
+
+config LH7A40X_ONE_BANK_PER_NODE
+	bool "Optimize NUMA Node Tables for Size"
+	depends on ARCH_LH7A40X && !LH7A40X_CONTIGMEM
+	help
+	  Say Y here to produce compact memory node tables.  By
+	  default pairs of adjacent physical RAM banks are managed
+	  together in a single node, incurring some wasted overhead
+	  in the node tables, however also maintaining compatibility
+	  with systems where physical memory is truly contiguous.
+
+	  Setting this option incorrectly may prevent the kernel from
+	  booting.  It is OK to leave it N.
+
+	  For more information, consult
+	    <file:Documentation/arm/Sharp-LH/SDRAM>.
+
 endmenu
 
 endif

arch/arm/mach-versatile/Makefile

@@ -2,4 +2,4 @@
 # Makefile for the linux kernel.
 #
 
-obj-y					:= core.o
+obj-y					:= core.o clock.o

arch/arm/mach-versatile/clock.c

+/*
+ *  linux/arch/arm/mach-versatile/clock.c
+ *
+ *  Copyright (C) 2004 ARM Limited.
+ *  Written by Deep Blue Solutions Limited.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/errno.h>
+#include <linux/err.h>
+
+#include <asm/semaphore.h>
+#include <asm/hardware/clock.h>
+#include <asm/hardware/icst525.h>
+
+#include "clock.h"
+
+static LIST_HEAD(clocks);
+static DECLARE_MUTEX(clocks_sem);
+
+struct clk *clk_get(struct device *dev, const char *id)
+{
+	struct clk *p, *clk = ERR_PTR(-ENOENT);
+
+	down(&clocks_sem);
+	list_for_each_entry(p, &clocks, node) {
+		if (strcmp(id, p->name) == 0 && try_module_get(p->owner)) {
+			clk = p;
+			break;
+		}
+	}
+	up(&clocks_sem);
+
+	return clk;
+}
+EXPORT_SYMBOL(clk_get);
+
+void clk_put(struct clk *clk)
+{
+	module_put(clk->owner);
+}
+EXPORT_SYMBOL(clk_put);
+
+int clk_enable(struct clk *clk)
+{
+	return 0;
+}
+EXPORT_SYMBOL(clk_enable);
+
+void clk_disable(struct clk *clk)
+{
+}
+EXPORT_SYMBOL(clk_disable);
+
+int clk_use(struct clk *clk)
+{
+	return 0;
+}
+EXPORT_SYMBOL(clk_use);
+
+void clk_unuse(struct clk *clk)
+{
+}
+EXPORT_SYMBOL(clk_unuse);
+
+unsigned long clk_get_rate(struct clk *clk)
+{
+	return clk->rate;
+}
+EXPORT_SYMBOL(clk_get_rate);
+
+long clk_round_rate(struct clk *clk, unsigned long rate)
+{
+	return rate;
+}
+EXPORT_SYMBOL(clk_round_rate);
+
+int clk_set_rate(struct clk *clk, unsigned long rate)
+{
+	int ret = -EIO;
+#if 0 // Not yet
+	if (clk->setvco) {
+		struct icst525_vco vco;
+
+		vco = icst525_khz_to_vco(clk->params, rate);
+		clk->rate = icst525_khz(clk->params, vco);
+
+		printk("Clock %s: setting VCO reg params: S=%d R=%d V=%d\n",
+			clk->name, vco.s, vco.r, vco.v);
+
+		clk->setvco(clk, vco);
+		ret = 0;
+	}
+#endif
+	return 0;
+}
+EXPORT_SYMBOL(clk_set_rate);
+
+/*
+ * These are fixed clocks.
+ */
+static struct clk kmi_clk = {
+	.name	= "KMIREFCLK",
+	.rate	= 24000000,
+};
+
+static struct clk uart_clk = {
+	.name	= "UARTCLK",
+	.rate	= 24000000,
+};
+
+static struct clk mmci_clk = {
+	.name	= "MCLK",
+	.rate	= 33000000,
+};
+
+int clk_register(struct clk *clk)
+{
+	down(&clocks_sem);
+	list_add(&clk->node, &clocks);
+	up(&clocks_sem);
+	return 0;
+}
+EXPORT_SYMBOL(clk_register);
+
+void clk_unregister(struct clk *clk)
+{
+	down(&clocks_sem);
+	list_del(&clk->node);
+	up(&clocks_sem);
+}
+EXPORT_SYMBOL(clk_unregister);
+
+static int __init clk_init(void)
+{
+	clk_register(&kmi_clk);
+	clk_register(&uart_clk);
+	clk_register(&mmci_clk);
+	return 0;
+}
+arch_initcall(clk_init);

arch/arm/mach-versatile/clock.h

+/*
+ *  linux/arch/arm/mach-versatile/clock.h
+ *
+ *  Copyright (C) 2004 ARM Limited.
+ *  Written by Deep Blue Solutions Limited.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+struct module;
+struct icst525_params;
+
+struct clk {
+	struct list_head	node;
+	unsigned long		rate;
+	struct module		*owner;
+	const char		*name;
+	const struct icst525_params *params;
+	void			*data;
+	void			(*setvco)(struct clk *, struct icst525_vco vco);
+};
+
+int clk_register(struct clk *clk);
+void clk_unregister(struct clk *clk);

arch/arm/vfp/Makefile

+#
+# linux/arch/arm/vfp/Makefile
+#
+# Copyright (C) 2001 ARM Limited
+#
+
+# EXTRA_CFLAGS := -DDEBUG
+# EXTRA_AFLAGS := -DDEBUG
+
+obj-y			+= vfp.o
+
+vfp-$(CONFIG_VFP)	+= entry.o vfpmodule.o vfphw.o vfpsingle.o vfpdouble.o

arch/arm/vfp/entry.S

+/*
+ *  linux/arch/arm/vfp/entry.S
+ *
+ *  Copyright (C) 2004 ARM Limited.
+ *  Written by Deep Blue Solutions Limited.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * Basic entry code, called from the kernel's undefined instruction trap.
+ *  r0  = faulted instruction
+ *  r5  = faulted PC+4
+ *  r9  = successful return
+ *  r10 = thread_info structure
+ *  lr  = failure return
+ */
+#include <linux/linkage.h>
+#include <linux/init.h>
+#include <asm/thread_info.h>
+#include <asm/vfpmacros.h>
+
+	.globl	do_vfp
+do_vfp:
+ 	ldr	r4, .LCvfp
+	add	r10, r10, #TI_VFPSTATE	@ r10 = workspace
+	ldr	pc, [r4]		@ call VFP entry point
+
+.LCvfp:
+	.word	vfp_vector
+
+@ This code is called if the VFP does not exist. It needs to flag the
+@ failure to the VFP initialisation code.
+
+	__INIT
+	.globl	vfp_testing_entry
+vfp_testing_entry:
+	ldr	r0, VFP_arch_address
+	str	r5, [r0]		@ known non-zero value
+	mov	pc, r9			@ we have handled the fault
+
+VFP_arch_address:
+	.word	VFP_arch
+
+	__FINIT

arch/arm/vfp/vfp.h

+/*
+ *  linux/arch/arm/vfp/vfp.h
+ *
+ *  Copyright (C) 2004 ARM Limited.
+ *  Written by Deep Blue Solutions Limited.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+static inline u32 vfp_shiftright32jamming(u32 val, unsigned int shift)
+{
+	if (shift) {
+		if (shift < 32)
+			val = val >> shift | ((val << (32 - shift)) != 0);
+		else
+			val = val != 0;
+	}
+	return val;
+}
+
+static inline u64 vfp_shiftright64jamming(u64 val, unsigned int shift)
+{
+	if (shift) {
+		if (shift < 64)
+			val = val >> shift | ((val << (64 - shift)) != 0);
+		else
+			val = val != 0;
+	}
+	return val;
+}
+
+static inline u32 vfp_hi64to32jamming(u64 val)
+{
+	u32 v;
+
+	asm(
+	"cmp	%Q1, #1		@ vfp_hi64to32jamming\n\t"
+	"movcc	%0, %R1\n\t"
+	"orrcs	%0, %R1, #1"
+	: "=r" (v) : "r" (val) : "cc");
+
+	return v;
+}
+
+static inline void add128(u64 *resh, u64 *resl, u64 nh, u64 nl, u64 mh, u64 ml)
+{
+	asm(	"adds	%Q0, %Q2, %Q4\n\t"
+		"adcs	%R0, %R2, %R4\n\t"
+		"adcs	%Q1, %Q3, %Q5\n\t"
+		"adc	%R1, %R3, %R5"
+	    : "=r" (nl), "=r" (nh)
+	    : "0" (nl), "1" (nh), "r" (ml), "r" (mh)
+	    : "cc");
+	*resh = nh;
+	*resl = nl;
+}
+
+static inline void sub128(u64 *resh, u64 *resl, u64 nh, u64 nl, u64 mh, u64 ml)
+{
+	asm(	"subs	%Q0, %Q2, %Q4\n\t"
+		"sbcs	%R0, %R2, %R4\n\t"
+		"sbcs	%Q1, %Q3, %Q5\n\t"
+		"sbc	%R1, %R3, %R5\n\t"
+	    : "=r" (nl), "=r" (nh)
+	    : "0" (nl), "1" (nh), "r" (ml), "r" (mh)
+	    : "cc");
+	*resh = nh;
+	*resl = nl;
+}
+
+static inline void mul64to128(u64 *resh, u64 *resl, u64 n, u64 m)
+{
+	u32 nh, nl, mh, ml;
+	u64 rh, rma, rmb, rl;
+
+	nl = n;
+	ml = m;
+	rl = (u64)nl * ml;
+
+	nh = n >> 32;
+	rma = (u64)nh * ml;
+
+	mh = m >> 32;
+	rmb = (u64)nl * mh;
+	rma += rmb;
+
+	rh = (u64)nh * mh;
+	rh += ((u64)(rma < rmb) << 32) + (rma >> 32);
+
+	rma <<= 32;
+	rl += rma;
+	rh += (rl < rma);
+
+	*resl = rl;
+	*resh = rh;
+}
+
+static inline void shift64left(u64 *resh, u64 *resl, u64 n)
+{
+	*resh = n >> 63;
+	*resl = n << 1;
+}
+
+static inline u64 vfp_hi64multiply64(u64 n, u64 m)
+{
+	u64 rh, rl;
+	mul64to128(&rh, &rl, n, m);
+	return rh | (rl != 0);
+}
+
+static inline u64 vfp_estimate_div128to64(u64 nh, u64 nl, u64 m)
+{
+	u64 mh, ml, remh, reml, termh, terml, z;
+
+	if (nh >= m)
+		return ~0ULL;
+	mh = m >> 32;
+	z = (mh << 32 <= nh) ? 0xffffffff00000000ULL : (nh / mh) << 32;
+	mul64to128(&termh, &terml, m, z);
+	sub128(&remh, &reml, nh, nl, termh, terml);
+	ml = m << 32;
+	while ((s64)remh < 0) {
+		z -= 0x100000000ULL;
+		add128(&remh, &reml, remh, reml, mh, ml);
+	}
+	remh = (remh << 32) | (reml >> 32);
+	z |= (mh << 32 <= remh) ? 0xffffffff : remh / mh;
+	return z;
+}
+
+/*
+ * Operations on unpacked elements
+ */
+#define vfp_sign_negate(sign)	(sign ^ 0x8000)
+
+/*
+ * Single-precision
+ */
+struct vfp_single {
+	s16	exponent;
+	u16	sign;
+	u32	significand;
+};
+
+extern s32 vfp_get_float(unsigned int reg);
+extern void vfp_put_float(unsigned int reg, s32 val);
+
+/*
+ * VFP_SINGLE_MANTISSA_BITS - number of bits in the mantissa
+ * VFP_SINGLE_EXPONENT_BITS - number of bits in the exponent
+ * VFP_SINGLE_LOW_BITS - number of low bits in the unpacked significand
+ *  which are not propagated to the float upon packing.
+ */
+#define VFP_SINGLE_MANTISSA_BITS	(23)
+#define VFP_SINGLE_EXPONENT_BITS	(8)
+#define VFP_SINGLE_LOW_BITS		(32 - VFP_SINGLE_MANTISSA_BITS - 2)
+#define VFP_SINGLE_LOW_BITS_MASK	((1 << VFP_SINGLE_LOW_BITS) - 1)
+
+/*
+ * The bit in an unpacked float which indicates that it is a quiet NaN
+ */
+#define VFP_SINGLE_SIGNIFICAND_QNAN	(1 << (VFP_SINGLE_MANTISSA_BITS - 1 + VFP_SINGLE_LOW_BITS))
+
+/*
+ * Operations on packed single-precision numbers
+ */
+#define vfp_single_packed_sign(v)	((v) & 0x80000000)
+#define vfp_single_packed_negate(v)	((v) ^ 0x80000000)
+#define vfp_single_packed_abs(v)	((v) & ~0x80000000)
+#define vfp_single_packed_exponent(v)	(((v) >> VFP_SINGLE_MANTISSA_BITS) & ((1 << VFP_SINGLE_EXPONENT_BITS) - 1))
+#define vfp_single_packed_mantissa(v)	((v) & ((1 << VFP_SINGLE_MANTISSA_BITS) - 1))
+
+/*
+ * Unpack a single-precision float.  Note that this returns the magnitude
+ * of the single-precision float mantissa with the 1. if necessary,
+ * aligned to bit 30.
+ */
+static inline void vfp_single_unpack(struct vfp_single *s, s32 val)
+{
+	u32 significand;
+
+	s->sign = vfp_single_packed_sign(val) >> 16,
+	s->exponent = vfp_single_packed_exponent(val);
+
+	significand = (u32) val;
+	significand = (significand << (32 - VFP_SINGLE_MANTISSA_BITS)) >> 2;
+	if (s->exponent && s->exponent != 255)
+		significand |= 0x40000000;
+	s->significand = significand;
+}
+
+/*
+ * Re-pack a single-precision float.  This assumes that the float is
+ * already normalised such that the MSB is bit 30, _not_ bit 31.
+ */
+static inline s32 vfp_single_pack(struct vfp_single *s)
+{
+	u32 val;
+	val = (s->sign << 16) +
+	      (s->exponent << VFP_SINGLE_MANTISSA_BITS) +
+	      (s->significand >> VFP_SINGLE_LOW_BITS);
+	return (s32)val;
+}
+
+#define VFP_NUMBER		(1<<0)
+#define VFP_ZERO		(1<<1)
+#define VFP_DENORMAL		(1<<2)
+#define VFP_INFINITY		(1<<3)
+#define VFP_NAN			(1<<4)
+#define VFP_NAN_SIGNAL		(1<<5)
+
+#define VFP_QNAN		(VFP_NAN)
+#define VFP_SNAN		(VFP_NAN|VFP_NAN_SIGNAL)
+
+static inline int vfp_single_type(struct vfp_single *s)
+{
+	int type = VFP_NUMBER;
+	if (s->exponent == 255) {
+		if (s->significand == 0)
+			type = VFP_INFINITY;
+		else if (s->significand & VFP_SINGLE_SIGNIFICAND_QNAN)
+			type = VFP_QNAN;
+		else
+			type = VFP_SNAN;
+	} else if (s->exponent == 0) {
+		if (s->significand == 0)
+			type |= VFP_ZERO;
+		else
+			type |= VFP_DENORMAL;
+	}
+	return type;
+}
+
+#ifndef DEBUG
+#define vfp_single_normaliseround(sd,vsd,fpscr,except,func) __vfp_single_normaliseround(sd,vsd,fpscr,except)
+u32 __vfp_single_normaliseround(int sd, struct vfp_single *vs, u32 fpscr, u32 exceptions);
+#else
+u32 vfp_single_normaliseround(int sd, struct vfp_single *vs, u32 fpscr, u32 exceptions, const char *func);
+#endif
+
+/*
+ * Double-precision
+ */
+struct vfp_double {
+	s16	exponent;
+	u16	sign;
+	u64	significand;
+};
+
+extern u64 vfp_get_double(unsigned int reg);
+extern void vfp_put_double(unsigned int reg, u64 val);
+
+#define VFP_DOUBLE_MANTISSA_BITS	(52)
+#define VFP_DOUBLE_EXPONENT_BITS	(11)
+#define VFP_DOUBLE_LOW_BITS		(64 - VFP_DOUBLE_MANTISSA_BITS - 2)
+#define VFP_DOUBLE_LOW_BITS_MASK	((1 << VFP_DOUBLE_LOW_BITS) - 1)
+
+/*
+ * The bit in an unpacked double which indicates that it is a quiet NaN
+ */
+#define VFP_DOUBLE_SIGNIFICAND_QNAN	(1ULL << (VFP_DOUBLE_MANTISSA_BITS - 1 + VFP_DOUBLE_LOW_BITS))
+
+/*
+ * Operations on packed single-precision numbers
+ */
+#define vfp_double_packed_sign(v)	((v) & (1ULL << 63))
+#define vfp_double_packed_negate(v)	((v) ^ (1ULL << 63))
+#define vfp_double_packed_abs(v)	((v) & ~(1ULL << 63))
+#define vfp_double_packed_exponent(v)	(((v) >> VFP_DOUBLE_MANTISSA_BITS) & ((1 << VFP_DOUBLE_EXPONENT_BITS) - 1))
+#define vfp_double_packed_mantissa(v)	((v) & ((1ULL << VFP_DOUBLE_MANTISSA_BITS) - 1))
+
+/*
+ * Unpack a double-precision float.  Note that this returns the magnitude
+ * of the double-precision float mantissa with the 1. if necessary,
+ * aligned to bit 62.
+ */
+static inline void vfp_double_unpack(struct vfp_double *s, s64 val)
+{
+	u64 significand;
+
+	s->sign = vfp_double_packed_sign(val) >> 48;
+	s->exponent = vfp_double_packed_exponent(val);
+
+	significand = (u64) val;
+	significand = (significand << (64 - VFP_DOUBLE_MANTISSA_BITS)) >> 2;
+	if (s->exponent && s->exponent != 2047)
+		significand |= (1ULL << 62);
+	s->significand = significand;
+}
+
+/*
+ * Re-pack a double-precision float.  This assumes that the float is
+ * already normalised such that the MSB is bit 30, _not_ bit 31.
+ */
+static inline s64 vfp_double_pack(struct vfp_double *s)
+{
+	u64 val;
+	val = ((u64)s->sign << 48) +
+	      ((u64)s->exponent << VFP_DOUBLE_MANTISSA_BITS) +
+	      (s->significand >> VFP_DOUBLE_LOW_BITS);
+	return (s64)val;
+}
+
+static inline int vfp_double_type(struct vfp_double *s)
+{
+	int type = VFP_NUMBER;
+	if (s->exponent == 2047) {
+		if (s->significand == 0)
+			type = VFP_INFINITY;
+		else if (s->significand & VFP_DOUBLE_SIGNIFICAND_QNAN)
+			type = VFP_QNAN;
+		else
+			type = VFP_SNAN;
+	} else if (s->exponent == 0) {
+		if (s->significand == 0)
+			type |= VFP_ZERO;
+		else
+			type |= VFP_DENORMAL;
+	}
+	return type;
+}
+
+u32 vfp_double_normaliseround(int dd, struct vfp_double *vd, u32 fpscr, u32 exceptions, const char *func);
+
+/*
+ * System registers
+ */
+extern u32 vfp_get_sys(unsigned int reg);
+extern void vfp_put_sys(unsigned int reg, u32 val);
+
+u32 vfp_estimate_sqrt_significand(u32 exponent, u32 significand);

arch/arm/vfp/vfpdouble.c

+/*
+ *  linux/arch/arm/vfp/vfpdouble.c
+ *
+ * This code is derived in part from John R. Housers softfloat library, which
+ * carries the following notice:
+ *
+ * ===========================================================================
+ * This C source file is part of the SoftFloat IEC/IEEE Floating-point
+ * Arithmetic Package, Release 2.
+ *
+ * Written by John R. Hauser.  This work was made possible in part by the
+ * International Computer Science Institute, located at Suite 600, 1947 Center
+ * Street, Berkeley, California 94704.  Funding was partially provided by the
+ * National Science Foundation under grant MIP-9311980.  The original version
+ * of this code was written as part of a project to build a fixed-point vector
+ * processor in collaboration with the University of California at Berkeley,
+ * overseen by Profs. Nelson Morgan and John Wawrzynek.  More information
+ * is available through the web page `http://HTTP.CS.Berkeley.EDU/~jhauser/
+ * arithmetic/softfloat.html'.
+ *
+ * THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE.  Although reasonable effort
+ * has been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT
+ * TIMES RESULT IN INCORRECT BEHAVIOR.  USE OF THIS SOFTWARE IS RESTRICTED TO
+ * PERSONS AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ANY
+ * AND ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM ITS USE.
+ *
+ * Derivative works are acceptable, even for commercial purposes, so long as
+ * (1) they include prominent notice that the work is derivative, and (2) they
+ * include prominent notice akin to these three paragraphs for those parts of
+ * this code that are retained.
+ * ===========================================================================
+ */
+#include <linux/kernel.h>
+#include <asm/bitops.h>
+#include <asm/ptrace.h>
+#include <asm/vfp.h>
+
+#include "vfpinstr.h"
+#include "vfp.h"
+
+static struct vfp_double vfp_double_default_qnan = {
+	.exponent	= 2047,
+	.sign		= 0,
+	.significand	= VFP_DOUBLE_SIGNIFICAND_QNAN,
+};
+
+static void vfp_double_dump(const char *str, struct vfp_double *d)
+{
+	pr_debug("VFP: %s: sign=%d exponent=%d significand=%016llx\n",
+		 str, d->sign != 0, d->exponent, d->significand);
+}
+
+static void vfp_double_normalise_denormal(struct vfp_double *vd)
+{
+	int bits = 31 - fls(vd->significand >> 32);
+	if (bits == 31)
+		bits = 62 - fls(vd->significand);
+
+	vfp_double_dump("normalise_denormal: in", vd);
+
+	if (bits) {
+		vd->exponent -= bits - 1;
+		vd->significand <<= bits;
+	}
+
+	vfp_double_dump("normalise_denormal: out", vd);
+}
+
+u32 vfp_double_normaliseround(int dd, struct vfp_double *vd, u32 fpscr, u32 exceptions, const char *func)
+{
+	u64 significand, incr;
+	int exponent, shift, underflow;
+	u32 rmode;
+
+	vfp_double_dump("pack: in", vd);
+
+	/*
+	 * Infinities and NaNs are a special case.
+	 */
+	if (vd->exponent == 2047 && (vd->significand == 0 || exceptions))
+		goto pack;
+
+	/*
+	 * Special-case zero.
+	 */
+	if (vd->significand == 0) {
+		vd->exponent = 0;
+		goto pack;
+	}
+
+	exponent = vd->exponent;
+	significand = vd->significand;
+
+	shift = 32 - fls(significand >> 32);
+	if (shift == 32)
+		shift = 64 - fls(significand);
+	if (shift) {
+		exponent -= shift;
+		significand <<= shift;
+	}
+
+#ifdef DEBUG
+	vd->exponent = exponent;
+	vd->significand = significand;
+	vfp_double_dump("pack: normalised", vd);
+#endif
+
+	/*
+	 * Tiny number?
+	 */
+	underflow = exponent < 0;
+	if (underflow) {
+		significand = vfp_shiftright64jamming(significand, -exponent);
+		exponent = 0;
+#ifdef DEBUG
+		vd->exponent = exponent;
+		vd->significand = significand;
+		vfp_double_dump("pack: tiny number", vd);
+#endif
+		if (!(significand & ((1ULL << (VFP_DOUBLE_LOW_BITS + 1)) - 1)))
+			underflow = 0;
+	}
+
+	/*
+	 * Select rounding increment.
+	 */
+	incr = 0;
+	rmode = fpscr & FPSCR_RMODE_MASK;
+
+	if (rmode == FPSCR_ROUND_NEAREST) {
+		incr = 1ULL << VFP_DOUBLE_LOW_BITS;
+		if ((significand & (1ULL << (VFP_DOUBLE_LOW_BITS + 1))) == 0)
+			incr -= 1;
+	} else if (rmode == FPSCR_ROUND_TOZERO) {
+		incr = 0;
+	} else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vd->sign != 0))
+		incr = (1ULL << (VFP_DOUBLE_LOW_BITS + 1)) - 1;
+
+	pr_debug("VFP: rounding increment = 0x%08llx\n", incr);
+
+	/*
+	 * Is our rounding going to overflow?
+	 */
+	if ((significand + incr) < significand) {
+		exponent += 1;
+		significand = (significand >> 1) | (significand & 1);
+		incr >>= 1;
+#ifdef DEBUG
+		vd->exponent = exponent;
+		vd->significand = significand;
+		vfp_double_dump("pack: overflow", vd);
+#endif
+	}
+
+	/*
+	 * If any of the low bits (which will be shifted out of the
+	 * number) are non-zero, the result is inexact.
+	 */
+	if (significand & ((1 << (VFP_DOUBLE_LOW_BITS + 1)) - 1))
+		exceptions |= FPSCR_IXC;
+
+	/*
+	 * Do our rounding.
+	 */
+	significand += incr;
+
+	/*
+	 * Infinity?
+	 */
+	if (exponent >= 2046) {
+		exceptions |= FPSCR_OFC | FPSCR_IXC;
+		if (incr == 0) {
+			vd->exponent = 2045;
+			vd->significand = 0x7fffffffffffffffULL;
+		} else {
+			vd->exponent = 2047;		/* infinity */
+			vd->significand = 0;
+		}
+	} else {
+		if (significand >> (VFP_DOUBLE_LOW_BITS + 1) == 0)
+			exponent = 0;
+		if (exponent || significand > 0x8000000000000000ULL)
+			underflow = 0;
+		if (underflow)
+			exceptions |= FPSCR_UFC;
+		vd->exponent = exponent;
+		vd->significand = significand >> 1;
+	}
+
+ pack:
+	vfp_double_dump("pack: final", vd);
+	{
+		s64 d = vfp_double_pack(vd);
+		pr_debug("VFP: %s: d(d%d)=%016llx exceptions=%08x\n", func,
+			 dd, d, exceptions);
+		vfp_put_double(dd, d);
+	}
+	return exceptions;
+}
+
+/*
+ * Propagate the NaN, setting exceptions if it is signalling.
+ * 'n' is always a NaN.  'm' may be a number, NaN or infinity.
+ */
+static u32
+vfp_propagate_nan(struct vfp_double *vdd, struct vfp_double *vdn,
+		  struct vfp_double *vdm, u32 fpscr)
+{
+	struct vfp_double *nan;
+	int tn, tm = 0;
+
+	tn = vfp_double_type(vdn);
+
+	if (vdm)
+		tm = vfp_double_type(vdm);
+
+	if (fpscr & FPSCR_DEFAULT_NAN)
+		/*
+		 * Default NaN mode - always returns a quiet NaN
+		 */
+		nan = &vfp_double_default_qnan;
+	else {
+		/*
+		 * Contemporary mode - select the first signalling
+		 * NAN, or if neither are signalling, the first
+		 * quiet NAN.
+		 */
+		if (tn == VFP_SNAN || (tm != VFP_SNAN && tn == VFP_QNAN))
+			nan = vdn;
+		else
+			nan = vdm;
+		/*
+		 * Make the NaN quiet.
+		 */
+		nan->significand |= VFP_DOUBLE_SIGNIFICAND_QNAN;
+	}
+
+	*vdd = *nan;
+
+	/*
+	 * If one was a signalling NAN, raise invalid operation.
+	 */
+	return tn == VFP_SNAN || tm == VFP_SNAN ? FPSCR_IOC : 0x100;
+}
+
+/*
+ * Extended operations
+ */
+static u32 vfp_double_fabs(int dd, int unused, int dm, u32 fpscr)
+{
+	vfp_put_double(dd, vfp_double_packed_abs(vfp_get_double(dm)));
+	return 0;
+}
+
+static u32 vfp_double_fcpy(int dd, int unused, int dm, u32 fpscr)
+{
+	vfp_put_double(dd, vfp_get_double(dm));
+	return 0;
+}
+
+static u32 vfp_double_fneg(int dd, int unused, int dm, u32 fpscr)
+{
+	vfp_put_double(dd, vfp_double_packed_negate(vfp_get_double(dm)));
+	return 0;
+}
+
+static u32 vfp_double_fsqrt(int dd, int unused, int dm, u32 fpscr)
+{
+	struct vfp_double vdm, vdd;
+	int ret, tm;
+
+	vfp_double_unpack(&vdm, vfp_get_double(dm));
+	tm = vfp_double_type(&vdm);
+	if (tm & (VFP_NAN|VFP_INFINITY)) {
+		struct vfp_double *vdp = &vdd;
+
+		if (tm & VFP_NAN)
+			ret = vfp_propagate_nan(vdp, &vdm, NULL, fpscr);
+		else if (vdm.sign == 0) {
+ sqrt_copy:
+			vdp = &vdm;
+			ret = 0;
+		} else {
+ sqrt_invalid:
+			vdp = &vfp_double_default_qnan;
+			ret = FPSCR_IOC;
+		}
+		vfp_put_double(dd, vfp_double_pack(vdp));
+		return ret;
+	}
+
+	/*
+	 * sqrt(+/- 0) == +/- 0
+	 */
+	if (tm & VFP_ZERO)
+		goto sqrt_copy;
+
+	/*
+	 * Normalise a denormalised number
+	 */
+	if (tm & VFP_DENORMAL)
+		vfp_double_normalise_denormal(&vdm);
+
+	/*
+	 * sqrt(<0) = invalid
+	 */
+	if (vdm.sign)
+		goto sqrt_invalid;
+
+	vfp_double_dump("sqrt", &vdm);
+
+	/*
+	 * Estimate the square root.
+	 */
+	vdd.sign = 0;
+	vdd.exponent = ((vdm.exponent - 1023) >> 1) + 1023;
+	vdd.significand = (u64)vfp_estimate_sqrt_significand(vdm.exponent, vdm.significand >> 32) << 31;
+
+	vfp_double_dump("sqrt estimate1", &vdd);
+
+	vdm.significand >>= 1 + (vdm.exponent & 1);
+	vdd.significand += 2 + vfp_estimate_div128to64(vdm.significand, 0, vdd.significand);
+
+	vfp_double_dump("sqrt estimate2", &vdd);
+
+	/*
+	 * And now adjust.
+	 */
+	if ((vdd.significand & VFP_DOUBLE_LOW_BITS_MASK) <= 5) {
+		if (vdd.significand < 2) {
+			vdd.significand = ~0ULL;
+		} else {
+			u64 termh, terml, remh, reml;
+			vdm.significand <<= 2;
+			mul64to128(&termh, &terml, vdd.significand, vdd.significand);
+			sub128(&remh, &reml, vdm.significand, 0, termh, terml);
+			while ((s64)remh < 0) {
+				vdd.significand -= 1;
+				shift64left(&termh, &terml, vdd.significand);
+				terml |= 1;
+				add128(&remh, &reml, remh, reml, termh, terml);
+			}
+			vdd.significand |= (remh | reml) != 0;
+		}
+	}
+	vdd.significand = vfp_shiftright64jamming(vdd.significand, 1);
+
+	return vfp_double_normaliseround(dd, &vdd, fpscr, 0, "fsqrt");
+}
+
+/*
+ * Equal	:= ZC
+ * Less than	:= N
+ * Greater than	:= C
+ * Unordered	:= CV
+ */
+static u32 vfp_compare(int dd, int signal_on_qnan, int dm, u32 fpscr)
+{
+	s64 d, m;
+	u32 ret = 0;
+
+	m = vfp_get_double(dm);
+	if (vfp_double_packed_exponent(m) == 2047 && vfp_double_packed_mantissa(m)) {
+		ret |= FPSCR_C | FPSCR_V;
+		if (signal_on_qnan || !(vfp_double_packed_mantissa(m) & (1ULL << (VFP_DOUBLE_MANTISSA_BITS - 1))))
+			/*
+			 * Signalling NaN, or signalling on quiet NaN
+			 */
+			ret |= FPSCR_IOC;
+	}
+
+	d = vfp_get_double(dd);
+	if (vfp_double_packed_exponent(d) == 2047 && vfp_double_packed_mantissa(d)) {
+		ret |= FPSCR_C | FPSCR_V;
+		if (signal_on_qnan || !(vfp_double_packed_mantissa(d) & (1ULL << (VFP_DOUBLE_MANTISSA_BITS - 1))))
+			/*
+			 * Signalling NaN, or signalling on quiet NaN
+			 */
+			ret |= FPSCR_IOC;
+	}
+
+	if (ret == 0) {
+		if (d == m || vfp_double_packed_abs(d | m) == 0) {
+			/*
+			 * equal
+			 */
+			ret |= FPSCR_Z | FPSCR_C;
+		} else if (vfp_double_packed_sign(d ^ m)) {
+			/*
+			 * different signs
+			 */
+			if (vfp_double_packed_sign(d))
+				/*
+				 * d is negative, so d < m
+				 */
+				ret |= FPSCR_N;
+			else
+				/*
+				 * d is positive, so d > m
+				 */
+				ret |= FPSCR_C;
+		} else if ((vfp_double_packed_sign(d) != 0) ^ (d < m)) {
+			/*
+			 * d < m
+			 */
+			ret |= FPSCR_N;
+		} else if ((vfp_double_packed_sign(d) != 0) ^ (d > m)) {
+			/*
+			 * d > m
+			 */
+			ret |= FPSCR_C;
+		}
+	}
+
+	return ret;
+}
+
+static u32 vfp_double_fcmp(int dd, int unused, int dm, u32 fpscr)
+{
+	return vfp_compare(dd, 0, dm, fpscr);
+}
+
+static u32 vfp_double_fcmpe(int dd, int unused, int dm, u32 fpscr)
+{
+	return vfp_compare(dd, 1, dm, fpscr);
+}
+
+static u32 vfp_double_fcmpz(int dd, int unused, int dm, u32 fpscr)
+{
+	return vfp_compare(dd, 0, -1, fpscr);
+}
+
+static u32 vfp_double_fcmpez(int dd, int unused, int dm, u32 fpscr)
+{
+	return vfp_compare(dd, 1, -1, fpscr);
+}
+
+static u32 vfp_double_fcvts(int sd, int unused, int dm, u32 fpscr)
+{
+	struct vfp_double vdm;
+	struct vfp_single vsd;
+	int tm;
+	u32 exceptions = 0;
+
+	vfp_double_unpack(&vdm, vfp_get_double(dm));
+
+	tm = vfp_double_type(&vdm);
+
+	/*
+	 * If we have a signalling NaN, signal invalid operation.
+	 */
+	if (tm == VFP_SNAN)
+		exceptions = FPSCR_IOC;
+
+	if (tm & VFP_DENORMAL)
+		vfp_double_normalise_denormal(&vdm);
+
+	vsd.sign = vdm.sign;
+	vsd.significand = vfp_hi64to32jamming(vdm.significand);
+
+	/*
+	 * If we have an infinity or a NaN, the exponent must be 255
+	 */
+	if (tm & (VFP_INFINITY|VFP_NAN)) {
+		vsd.exponent = 255;
+		if (tm & VFP_NAN)
+			vsd.significand |= VFP_SINGLE_SIGNIFICAND_QNAN;
+		goto pack_nan;
+	} else if (tm & VFP_ZERO)
+		vsd.exponent = 0;
+	else
+		vsd.exponent = vdm.exponent - (1023 - 127);
+
+	return vfp_single_normaliseround(sd, &vsd, fpscr, exceptions, "fcvts");
+
+ pack_nan:
+	vfp_put_float(sd, vfp_single_pack(&vsd));
+	return exceptions;
+}
+
+static u32 vfp_double_fuito(int dd, int unused, int dm, u32 fpscr)
+{
+	struct vfp_double vdm;
+	u32 m = vfp_get_float(dm);
+
+	vdm.sign = 0;
+	vdm.exponent = 1023 + 63 - 1;
+	vdm.significand = (u64)m;
+
+	return vfp_double_normaliseround(dd, &vdm, fpscr, 0, "fuito");
+}
+
+static u32 vfp_double_fsito(int dd, int unused, int dm, u32 fpscr)
+{
+	struct vfp_double vdm;
+	u32 m = vfp_get_float(dm);
+
+	vdm.sign = (m & 0x80000000) >> 16;
+	vdm.exponent = 1023 + 63 - 1;
+	vdm.significand = vdm.sign ? -m : m;
+
+	return vfp_double_normaliseround(dd, &vdm, fpscr, 0, "fsito");
+}
+
+static u32 vfp_double_ftoui(int sd, int unused, int dm, u32 fpscr)
+{
+	struct vfp_double vdm;
+	u32 d, exceptions = 0;
+	int rmode = fpscr & FPSCR_RMODE_MASK;
+	int tm;
+
+	vfp_double_unpack(&vdm, vfp_get_double(dm));
+
+	/*
+	 * Do we have a denormalised number?
+	 */
+	tm = vfp_double_type(&vdm);
+	if (tm & VFP_DENORMAL)
+		exceptions |= FPSCR_IDC;
+
+	if (tm & VFP_NAN)
+		vdm.sign = 0;
+
+	if (vdm.exponent >= 1023 + 32) {
+		d = vdm.sign ? 0 : 0xffffffff;
+		exceptions = FPSCR_IOC;
+	} else if (vdm.exponent >= 1023 - 1) {
+		int shift = 1023 + 63 - vdm.exponent;
+		u64 rem, incr = 0;
+
+		/*
+		 * 2^0 <= m < 2^32-2^8
+		 */
+		d = (vdm.significand << 1) >> shift;
+		rem = vdm.significand << (65 - shift);
+
+		if (rmode == FPSCR_ROUND_NEAREST) {
+			incr = 0x8000000000000000ULL;
+			if ((d & 1) == 0)
+				incr -= 1;
+		} else if (rmode == FPSCR_ROUND_TOZERO) {
+			incr = 0;
+		} else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vdm.sign != 0)) {
+			incr = ~0ULL;
+		}
+
+		if ((rem + incr) < rem) {
+			if (d < 0xffffffff)
+				d += 1;
+			else
+				exceptions |= FPSCR_IOC;
+		}
+
+		if (d && vdm.sign) {
+			d = 0;
+			exceptions |= FPSCR_IOC;
+		} else if (rem)
+			exceptions |= FPSCR_IXC;
+	} else {
+		d = 0;
+		if (vdm.exponent | vdm.significand) {
+			exceptions |= FPSCR_IXC;
+			if (rmode == FPSCR_ROUND_PLUSINF && vdm.sign == 0)
+				d = 1;
+			else if (rmode == FPSCR_ROUND_MINUSINF && vdm.sign) {
+				d = 0;
+				exceptions |= FPSCR_IOC;
+			}
+		}
+	}
+
+	pr_debug("VFP: ftoui: d(s%d)=%08x exceptions=%08x\n", sd, d, exceptions);
+
+	vfp_put_float(sd, d);
+
+	return exceptions;
+}
+
+static u32 vfp_double_ftouiz(int sd, int unused, int dm, u32 fpscr)
+{
+	return vfp_double_ftoui(sd, unused, dm, FPSCR_ROUND_TOZERO);
+}
+
+static u32 vfp_double_ftosi(int sd, int unused, int dm, u32 fpscr)
+{
+	struct vfp_double vdm;
+	u32 d, exceptions = 0;
+	int rmode = fpscr & FPSCR_RMODE_MASK;
+
+	vfp_double_unpack(&vdm, vfp_get_double(dm));
+	vfp_double_dump("VDM", &vdm);
+
+	/*
+	 * Do we have denormalised number?
+	 */
+	if (vfp_double_type(&vdm) & VFP_DENORMAL)
+		exceptions |= FPSCR_IDC;
+
+	if (vdm.exponent >= 1023 + 32) {
+		d = 0x7fffffff;
+		if (vdm.sign)
+			d = ~d;
+		exceptions |= FPSCR_IOC;
+	} else if (vdm.exponent >= 1023 - 1) {
+		int shift = 1023 + 63 - vdm.exponent;	/* 58 */
+		u64 rem, incr = 0;
+
+		d = (vdm.significand << 1) >> shift;
+		rem = vdm.significand << (65 - shift);
+
+		if (rmode == FPSCR_ROUND_NEAREST) {
+			incr = 0x8000000000000000ULL;
+			if ((d & 1) == 0)
+				incr -= 1;
+		} else if (rmode == FPSCR_ROUND_TOZERO) {
+			incr = 0;
+		} else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vdm.sign != 0)) {
+			incr = ~0ULL;
+		}
+
+		if ((rem + incr) < rem && d < 0xffffffff)
+			d += 1;
+		if (d > 0x7fffffff + (vdm.sign != 0)) {
+			d = 0x7fffffff + (vdm.sign != 0);
+			exceptions |= FPSCR_IOC;
+		} else if (rem)
+			exceptions |= FPSCR_IXC;
+
+		if (vdm.sign)
+			d = -d;
+	} else {
+		d = 0;
+		if (vdm.exponent | vdm.significand) {
+			exceptions |= FPSCR_IXC;
+			if (rmode == FPSCR_ROUND_PLUSINF && vdm.sign == 0)
+				d = 1;
+			else if (rmode == FPSCR_ROUND_MINUSINF && vdm.sign)
+				d = -1;
+		}
+	}
+
+	pr_debug("VFP: ftosi: d(s%d)=%08x exceptions=%08x\n", sd, d, exceptions);
+
+	vfp_put_float(sd, (s32)d);
+
+	return exceptions;
+}
+
+static u32 vfp_double_ftosiz(int dd, int unused, int dm, u32 fpscr)
+{
+	return vfp_double_ftosi(dd, unused, dm, FPSCR_ROUND_TOZERO);
+}
+
+
+static u32 (* const fop_extfns[32])(int dd, int unused, int dm, u32 fpscr) = {
+	[FEXT_TO_IDX(FEXT_FCPY)]	= vfp_double_fcpy,
+	[FEXT_TO_IDX(FEXT_FABS)]	= vfp_double_fabs,
+	[FEXT_TO_IDX(FEXT_FNEG)]	= vfp_double_fneg,
+	[FEXT_TO_IDX(FEXT_FSQRT)]	= vfp_double_fsqrt,
+	[FEXT_TO_IDX(FEXT_FCMP)]	= vfp_double_fcmp,
+	[FEXT_TO_IDX(FEXT_FCMPE)]	= vfp_double_fcmpe,
+	[FEXT_TO_IDX(FEXT_FCMPZ)]	= vfp_double_fcmpz,
+	[FEXT_TO_IDX(FEXT_FCMPEZ)]	= vfp_double_fcmpez,
+	[FEXT_TO_IDX(FEXT_FCVT)]	= vfp_double_fcvts,
+	[FEXT_TO_IDX(FEXT_FUITO)]	= vfp_double_fuito,
+	[FEXT_TO_IDX(FEXT_FSITO)]	= vfp_double_fsito,
+	[FEXT_TO_IDX(FEXT_FTOUI)]	= vfp_double_ftoui,
+	[FEXT_TO_IDX(FEXT_FTOUIZ)]	= vfp_double_ftouiz,
+	[FEXT_TO_IDX(FEXT_FTOSI)]	= vfp_double_ftosi,
+	[FEXT_TO_IDX(FEXT_FTOSIZ)]	= vfp_double_ftosiz,
+};
+
+
+
+
+static u32
+vfp_double_fadd_nonnumber(struct vfp_double *vdd, struct vfp_double *vdn,
+			  struct vfp_double *vdm, u32 fpscr)
+{
+	struct vfp_double *vdp;
+	u32 exceptions = 0;
+	int tn, tm;
+
+	tn = vfp_double_type(vdn);
+	tm = vfp_double_type(vdm);
+
+	if (tn & tm & VFP_INFINITY) {
+		/*
+		 * Two infinities.  Are they different signs?
+		 */
+		if (vdn->sign ^ vdm->sign) {
+			/*
+			 * different signs -> invalid
+			 */
+			exceptions = FPSCR_IOC;
+			vdp = &vfp_double_default_qnan;
+		} else {
+			/*
+			 * same signs -> valid
+			 */
+			vdp = vdn;
+		}
+	} else if (tn & VFP_INFINITY && tm & VFP_NUMBER) {
+		/*
+		 * One infinity and one number -> infinity
+		 */
+		vdp = vdn;
+	} else {
+		/*
+		 * 'n' is a NaN of some type
+		 */
+		return vfp_propagate_nan(vdd, vdn, vdm, fpscr);
+	}
+	*vdd = *vdp;
+	return exceptions;
+}
+
+static u32
+vfp_double_add(struct vfp_double *vdd, struct vfp_double *vdn,
+	       struct vfp_double *vdm, u32 fpscr)
+{
+	u32 exp_diff;
+	u64 m_sig;
+
+	if (vdn->significand & (1ULL << 63) ||
+	    vdm->significand & (1ULL << 63)) {
+		pr_info("VFP: bad FP values in %s\n", __func__);
+		vfp_double_dump("VDN", vdn);
+		vfp_double_dump("VDM", vdm);
+	}
+
+	/*
+	 * Ensure that 'n' is the largest magnitude number.  Note that
+	 * if 'n' and 'm' have equal exponents, we do not swap them.
+	 * This ensures that NaN propagation works correctly.
+	 */
+	if (vdn->exponent < vdm->exponent) {
+		struct vfp_double *t = vdn;
+		vdn = vdm;
+		vdm = t;
+	}
+
+	/*
+	 * Is 'n' an infinity or a NaN?  Note that 'm' may be a number,
+	 * infinity or a NaN here.
+	 */
+	if (vdn->exponent == 2047)
+		return vfp_double_fadd_nonnumber(vdd, vdn, vdm, fpscr);
+
+	/*
+	 * We have two proper numbers, where 'vdn' is the larger magnitude.
+	 *
+	 * Copy 'n' to 'd' before doing the arithmetic.
+	 */
+	*vdd = *vdn;
+
+	/*
+	 * Align 'm' with the result.
+	 */
+	exp_diff = vdn->exponent - vdm->exponent;
+	m_sig = vfp_shiftright64jamming(vdm->significand, exp_diff);
+
+	/*
+	 * If the signs are different, we are really subtracting.
+	 */
+	if (vdn->sign ^ vdm->sign) {
+		m_sig = vdn->significand - m_sig;
+		if ((s64)m_sig < 0) {
+			vdd->sign = vfp_sign_negate(vdd->sign);
+			m_sig = -m_sig;
+		}
+	} else {
+		m_sig += vdn->significand;
+	}
+	vdd->significand = m_sig;
+
+	return 0;
+}
+
+static u32
+vfp_double_multiply(struct vfp_double *vdd, struct vfp_double *vdn,
+		    struct vfp_double *vdm, u32 fpscr)
+{
+	vfp_double_dump("VDN", vdn);
+	vfp_double_dump("VDM", vdm);
+
+	/*
+	 * Ensure that 'n' is the largest magnitude number.  Note that
+	 * if 'n' and 'm' have equal exponents, we do not swap them.
+	 * This ensures that NaN propagation works correctly.
+	 */
+	if (vdn->exponent < vdm->exponent) {
+		struct vfp_double *t = vdn;
+		vdn = vdm;
+		vdm = t;
+		pr_debug("VFP: swapping M <-> N\n");
+	}
+
+	vdd->sign = vdn->sign ^ vdm->sign;
+
+	/*
+	 * If 'n' is an infinity or NaN, handle it.  'm' may be anything.
+	 */
+	if (vdn->exponent == 2047) {
+		if (vdn->significand || (vdm->exponent == 2047 && vdm->significand))
+			return vfp_propagate_nan(vdd, vdn, vdm, fpscr);
+		if ((vdm->exponent | vdm->significand) == 0) {
+			*vdd = vfp_double_default_qnan;
+			return FPSCR_IOC;
+		}
+		vdd->exponent = vdn->exponent;
+		vdd->significand = 0;
+		return 0;
+	}
+
+	/*
+	 * If 'm' is zero, the result is always zero.  In this case,
+	 * 'n' may be zero or a number, but it doesn't matter which.
+	 */
+	if ((vdm->exponent | vdm->significand) == 0) {
+		vdd->exponent = 0;
+		vdd->significand = 0;
+		return 0;
+	}
+
+	/*
+	 * We add 2 to the destination exponent for the same reason
+	 * as the addition case - though this time we have +1 from
+	 * each input operand.
+	 */
+	vdd->exponent = vdn->exponent + vdm->exponent - 1023 + 2;
+	vdd->significand = vfp_hi64multiply64(vdn->significand, vdm->significand);
+
+	vfp_double_dump("VDD", vdd);
+	return 0;
+}
+
+#define NEG_MULTIPLY	(1 << 0)
+#define NEG_SUBTRACT	(1 << 1)
+
+static u32
+vfp_double_multiply_accumulate(int dd, int dn, int dm, u32 fpscr, u32 negate, char *func)
+{
+	struct vfp_double vdd, vdp, vdn, vdm;
+	u32 exceptions;
+
+	vfp_double_unpack(&vdn, vfp_get_double(dn));
+	if (vdn.exponent == 0 && vdn.significand)
+		vfp_double_normalise_denormal(&vdn);
+
+	vfp_double_unpack(&vdm, vfp_get_double(dm));
+	if (vdm.exponent == 0 && vdm.significand)
+		vfp_double_normalise_denormal(&vdm);
+
+	exceptions = vfp_double_multiply(&vdp, &vdn, &vdm, fpscr);
+	if (negate & NEG_MULTIPLY)
+		vdp.sign = vfp_sign_negate(vdp.sign);
+
+	vfp_double_unpack(&vdn, vfp_get_double(dd));
+	if (negate & NEG_SUBTRACT)
+		vdn.sign = vfp_sign_negate(vdn.sign);
+
+	exceptions |= vfp_double_add(&vdd, &vdn, &vdp, fpscr);
+
+	return vfp_double_normaliseround(dd, &vdd, fpscr, exceptions, func);
+}
+
+/*
+ * Standard operations
+ */
+
+/*
+ * sd = sd + (sn * sm)
+ */
+static u32 vfp_double_fmac(int dd, int dn, int dm, u32 fpscr)
+{
+	return vfp_double_multiply_accumulate(dd, dn, dm, fpscr, 0, "fmac");
+}
+
+/*
+ * sd = sd - (sn * sm)
+ */
+static u32 vfp_double_fnmac(int dd, int dn, int dm, u32 fpscr)
+{
+	return vfp_double_multiply_accumulate(dd, dn, dm, fpscr, NEG_MULTIPLY, "fnmac");
+}
+
+/*
+ * sd = -sd + (sn * sm)
+ */
+static u32 vfp_double_fmsc(int dd, int dn, int dm, u32 fpscr)
+{
+	return vfp_double_multiply_accumulate(dd, dn, dm, fpscr, NEG_SUBTRACT, "fmsc");
+}
+
+/*
+ * sd = -sd - (sn * sm)
+ */
+static u32 vfp_double_fnmsc(int dd, int dn, int dm, u32 fpscr)
+{
+	return vfp_double_multiply_accumulate(dd, dn, dm, fpscr, NEG_SUBTRACT | NEG_MULTIPLY, "fnmsc");
+}
+
+/*
+ * sd = sn * sm
+ */
+static u32 vfp_double_fmul(int dd, int dn, int dm, u32 fpscr)
+{
+	struct vfp_double vdd, vdn, vdm;
+	u32 exceptions;
+
+	vfp_double_unpack(&vdn, vfp_get_double(dn));
+	if (vdn.exponent == 0 && vdn.significand)
+		vfp_double_normalise_denormal(&vdn);
+
+	vfp_double_unpack(&vdm, vfp_get_double(dm));
+	if (vdm.exponent == 0 && vdm.significand)
+		vfp_double_normalise_denormal(&vdm);
+
+	exceptions = vfp_double_multiply(&vdd, &vdn, &vdm, fpscr);
+	return vfp_double_normaliseround(dd, &vdd, fpscr, exceptions, "fmul");
+}
+
+/*
+ * sd = -(sn * sm)
+ */
+static u32 vfp_double_fnmul(int dd, int dn, int dm, u32 fpscr)
+{
+	struct vfp_double vdd, vdn, vdm;
+	u32 exceptions;
+
+	vfp_double_unpack(&vdn, vfp_get_double(dn));
+	if (vdn.exponent == 0 && vdn.significand)
+		vfp_double_normalise_denormal(&vdn);
+
+	vfp_double_unpack(&vdm, vfp_get_double(dm));
+	if (vdm.exponent == 0 && vdm.significand)
+		vfp_double_normalise_denormal(&vdm);
+
+	exceptions = vfp_double_multiply(&vdd, &vdn, &vdm, fpscr);
+	vdd.sign = vfp_sign_negate(vdd.sign);
+
+	return vfp_double_normaliseround(dd, &vdd, fpscr, exceptions, "fnmul");
+}
+
+/*
+ * sd = sn + sm
+ */
+static u32 vfp_double_fadd(int dd, int dn, int dm, u32 fpscr)
+{
+	struct vfp_double vdd, vdn, vdm;
+	u32 exceptions;
+
+	vfp_double_unpack(&vdn, vfp_get_double(dn));
+	if (vdn.exponent == 0 && vdn.significand)
+		vfp_double_normalise_denormal(&vdn);
+
+	vfp_double_unpack(&vdm, vfp_get_double(dm));
+	if (vdm.exponent == 0 && vdm.significand)
+		vfp_double_normalise_denormal(&vdm);
+
+	exceptions = vfp_double_add(&vdd, &vdn, &vdm, fpscr);
+
+	return vfp_double_normaliseround(dd, &vdd, fpscr, exceptions, "fadd");
+}
+
+/*
+ * sd = sn - sm
+ */
+static u32 vfp_double_fsub(int dd, int dn, int dm, u32 fpscr)
+{
+	struct vfp_double vdd, vdn, vdm;
+	u32 exceptions;
+
+	vfp_double_unpack(&vdn, vfp_get_double(dn));
+	if (vdn.exponent == 0 && vdn.significand)
+		vfp_double_normalise_denormal(&vdn);
+
+	vfp_double_unpack(&vdm, vfp_get_double(dm));
+	if (vdm.exponent == 0 && vdm.significand)
+		vfp_double_normalise_denormal(&vdm);
+
+	/*
+	 * Subtraction is like addition, but with a negated operand.
+	 */
+	vdm.sign = vfp_sign_negate(vdm.sign);
+
+	exceptions = vfp_double_add(&vdd, &vdn, &vdm, fpscr);
+
+	return vfp_double_normaliseround(dd, &vdd, fpscr, exceptions, "fsub");
+}
+
+/*
+ * sd = sn / sm
+ */
+static u32 vfp_double_fdiv(int dd, int dn, int dm, u32 fpscr)
+{
+	struct vfp_double vdd, vdn, vdm;
+	u32 exceptions = 0;
+	int tm, tn;
+
+	vfp_double_unpack(&vdn, vfp_get_double(dn));
+	vfp_double_unpack(&vdm, vfp_get_double(dm));
+
+	vdd.sign = vdn.sign ^ vdm.sign;
+
+	tn = vfp_double_type(&vdn);
+	tm = vfp_double_type(&vdm);
+
+	/*
+	 * Is n a NAN?
+	 */
+	if (tn & VFP_NAN)
+		goto vdn_nan;
+
+	/*
+	 * Is m a NAN?
+	 */
+	if (tm & VFP_NAN)
+		goto vdm_nan;
+
+	/*
+	 * If n and m are infinity, the result is invalid
+	 * If n and m are zero, the result is invalid
+	 */
+	if (tm & tn & (VFP_INFINITY|VFP_ZERO))
+		goto invalid;
+
+	/*
+	 * If n is infinity, the result is infinity
+	 */
+	if (tn & VFP_INFINITY)
+		goto infinity;
+
+	/*
+	 * If m is zero, raise div0 exceptions
+	 */
+	if (tm & VFP_ZERO)
+		goto divzero;
+
+	/*
+	 * If m is infinity, or n is zero, the result is zero
+	 */
+	if (tm & VFP_INFINITY || tn & VFP_ZERO)
+		goto zero;
+
+	if (tn & VFP_DENORMAL)
+		vfp_double_normalise_denormal(&vdn);
+	if (tm & VFP_DENORMAL)
+		vfp_double_normalise_denormal(&vdm);
+
+	/*
+	 * Ok, we have two numbers, we can perform division.
+	 */
+	vdd.exponent = vdn.exponent - vdm.exponent + 1023 - 1;
+	vdm.significand <<= 1;
+	if (vdm.significand <= (2 * vdn.significand)) {
+		vdn.significand >>= 1;
+		vdd.exponent++;
+	}
+	vdd.significand = vfp_estimate_div128to64(vdn.significand, 0, vdm.significand);
+	if ((vdd.significand & 0x1ff) <= 2) {
+		u64 termh, terml, remh, reml;
+		mul64to128(&termh, &terml, vdm.significand, vdd.significand);
+		sub128(&remh, &reml, vdn.significand, 0, termh, terml);
+		while ((s64)remh < 0) {
+			vdd.significand -= 1;
+			add128(&remh, &reml, remh, reml, 0, vdm.significand);
+		}
+		vdd.significand |= (reml != 0);
+	}
+	return vfp_double_normaliseround(dd, &vdd, fpscr, 0, "fdiv");
+
+ vdn_nan:
+	exceptions = vfp_propagate_nan(&vdd, &vdn, &vdm, fpscr);
+ pack:
+	vfp_put_double(dd, vfp_double_pack(&vdd));
+	return exceptions;
+
+ vdm_nan:
+	exceptions = vfp_propagate_nan(&vdd, &vdm, &vdn, fpscr);
+	goto pack;
+
+ zero:
+	vdd.exponent = 0;
+	vdd.significand = 0;
+	goto pack;
+
+ divzero:
+	exceptions = FPSCR_DZC;
+ infinity:
+	vdd.exponent = 2047;
+	vdd.significand = 0;
+	goto pack;
+
+ invalid:
+	vfp_put_double(dd, vfp_double_pack(&vfp_double_default_qnan));
+	return FPSCR_IOC;
+}
+
+static u32 (* const fop_fns[16])(int dd, int dn, int dm, u32 fpscr) = {
+	[FOP_TO_IDX(FOP_FMAC)]	= vfp_double_fmac,
+	[FOP_TO_IDX(FOP_FNMAC)]	= vfp_double_fnmac,
+	[FOP_TO_IDX(FOP_FMSC)]	= vfp_double_fmsc,
+	[FOP_TO_IDX(FOP_FNMSC)]	= vfp_double_fnmsc,
+	[FOP_TO_IDX(FOP_FMUL)]	= vfp_double_fmul,
+	[FOP_TO_IDX(FOP_FNMUL)]	= vfp_double_fnmul,
+	[FOP_TO_IDX(FOP_FADD)]	= vfp_double_fadd,
+	[FOP_TO_IDX(FOP_FSUB)]	= vfp_double_fsub,
+	[FOP_TO_IDX(FOP_FDIV)]	= vfp_double_fdiv,
+};
+
+#define FREG_BANK(x)	((x) & 0x0c)
+#define FREG_IDX(x)	((x) & 3)
+
+u32 vfp_double_cpdo(u32 inst, u32 fpscr)
+{
+	u32 op = inst & FOP_MASK;
+	u32 exceptions = 0;
+	unsigned int dd = vfp_get_sd(inst);
+	unsigned int dn = vfp_get_sn(inst);
+	unsigned int dm = vfp_get_sm(inst);
+	unsigned int vecitr, veclen, vecstride;
+	u32 (*fop)(int, int, s32, u32);
+
+	veclen = fpscr & FPSCR_LENGTH_MASK;
+	vecstride = (1 + ((fpscr & FPSCR_STRIDE_MASK) == FPSCR_STRIDE_MASK)) * 2;
+
+	/*
+	 * If destination bank is zero, vector length is always '1'.
+	 * ARM DDI0100F C5.1.3, C5.3.2.
+	 */
+	if (FREG_BANK(dd) == 0)
+		veclen = 0;
+
+	pr_debug("VFP: vecstride=%u veclen=%u\n", vecstride,
+		 (veclen >> FPSCR_LENGTH_BIT) + 1);
+
+	fop = (op == FOP_EXT) ? fop_extfns[dn] : fop_fns[FOP_TO_IDX(op)];
+	if (!fop)
+		goto invalid;
+
+	for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) {
+		u32 except;
+
+		if (op == FOP_EXT)
+			pr_debug("VFP: itr%d (d%u.%u) = op[%u] (d%u.%u)\n",
+				 vecitr >> FPSCR_LENGTH_BIT,
+				 dd >> 1, dd & 1, dn,
+				 dm >> 1, dm & 1);
+		else
+			pr_debug("VFP: itr%d (d%u.%u) = (d%u.%u) op[%u] (d%u.%u)\n",
+				 vecitr >> FPSCR_LENGTH_BIT,
+				 dd >> 1, dd & 1,
+				 dn >> 1, dn & 1,
+				 FOP_TO_IDX(op),
+				 dm >> 1, dm & 1);
+
+		except = fop(dd, dn, dm, fpscr);
+		pr_debug("VFP: itr%d: exceptions=%08x\n",
+			 vecitr >> FPSCR_LENGTH_BIT, except);
+
+		exceptions |= except;
+
+		/*
+		 * This ensures that comparisons only operate on scalars;
+		 * comparisons always return with one FPSCR status bit set.
+		 */
+		if (except & (FPSCR_N|FPSCR_Z|FPSCR_C|FPSCR_V))
+			break;
+
+		/*
+		 * CHECK: It appears to be undefined whether we stop when
+		 * we encounter an exception.  We continue.
+		 */
+
+		dd = FREG_BANK(dd) + ((FREG_IDX(dd) + vecstride) & 6);
+		dn = FREG_BANK(dn) + ((FREG_IDX(dn) + vecstride) & 6);
+		if (FREG_BANK(dm) != 0)
+			dm = FREG_BANK(dm) + ((FREG_IDX(dm) + vecstride) & 6);
+	}
+	return exceptions;
+
+ invalid:
+	return ~0;
+}

arch/arm/vfp/vfphw.S

+/*
+ *  linux/arch/arm/vfp/vfphw.S
+ *
+ *  Copyright (C) 2004 ARM Limited.
+ *  Written by Deep Blue Solutions Limited.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This code is called from the kernel's undefined instruction trap.
+ * r9 holds the return address for successful handling.
+ * lr holds the return address for unrecognised instructions.
+ * r10 points at the start of the private FP workspace in the thread structure
+ * sp points to a struct pt_regs (as defined in include/asm/proc/ptrace.h)
+ */
+#include <asm/thread_info.h>
+#include <asm/vfpmacros.h>
+#include "../kernel/entry-header.S"
+
+	.macro	DBGSTR, str
+#ifdef DEBUG
+	stmfd	sp!, {r0-r3, ip, lr}
+	add	r0, pc, #4
+	bl	printk
+	b	1f
+	.asciz  "<7>VFP: \str\n"
+	.balign 4
+1:	ldmfd	sp!, {r0-r3, ip, lr}
+#endif
+	.endm
+
+	.macro  DBGSTR1, str, arg
+#ifdef DEBUG
+	stmfd	sp!, {r0-r3, ip, lr}
+	mov	r1, \arg
+	add	r0, pc, #4
+	bl	printk
+	b	1f
+	.asciz  "<7>VFP: \str\n"
+	.balign 4
+1:	ldmfd	sp!, {r0-r3, ip, lr}
+#endif
+	.endm
+
+	.macro  DBGSTR3, str, arg1, arg2, arg3
+#ifdef DEBUG
+	stmfd	sp!, {r0-r3, ip, lr}
+	mov	r3, \arg3
+	mov	r2, \arg2
+	mov	r1, \arg1
+	add	r0, pc, #4
+	bl	printk
+	b	1f
+	.asciz  "<7>VFP: \str\n"
+	.balign 4
+1:	ldmfd	sp!, {r0-r3, ip, lr}
+#endif
+	.endm
+
+
+@ VFP hardware support entry point.
+@
+@  r0  = faulted instruction
+@  r5  = faulted PC+4
+@  r9  = successful return
+@  r10 = vfp_state union
+@  lr  = failure return
+
+	.globl	vfp_support_entry
+vfp_support_entry:
+	DBGSTR3	"instr %08x pc %08x state %p", r0, r5, r10
+
+	VFPFMRX	r1, FPEXC		@ Is the VFP enabled?
+	DBGSTR1	"fpexc %08x", r1
+	tst	r1, #FPEXC_ENABLE
+	bne	look_for_VFP_exceptions	@ VFP is already enabled
+
+	DBGSTR1 "enable %x", r10
+	ldr	r3, last_VFP_context_address
+	orr	r1, r1, #FPEXC_ENABLE	@ user FPEXC has the enable bit set
+	ldr	r4, [r3]		@ last_VFP_context pointer
+	bic	r2, r1, #FPEXC_EXCEPTION @ make sure exceptions are disabled
+	cmp	r4, r10
+	beq	check_for_exception	@ we are returning to the same
+					@ process, so the registers are
+					@ still there.  In this case, we do
+					@ not want to drop a pending exception.
+
+	VFPFMXR	FPEXC, r2		@ enable VFP, disable any pending
+					@ exceptions, so we can get at the
+					@ rest of it
+
+	@ Save out the current registers to the old thread state
+
+	DBGSTR1	"save old state %p", r4
+	cmp	r4, #0
+	beq	no_old_VFP_process
+	VFPFMRX	r2, FPSCR		@ current status
+	VFPFMRX	r6, FPINST		@ FPINST (always there, rev0 onwards)
+	tst	r1, #FPEXC_FPV2		@ is there an FPINST2 to read?
+	VFPFMRX	r8, FPINST2, NE		@ FPINST2 if needed - avoids reading
+					@ nonexistant reg on rev0
+	VFPFSTMIA r4 			@ save the working registers
+	add	r4, r4, #8*16+4
+	stmia	r4, {r1, r2, r6, r8}	@ save FPEXC, FPSCR, FPINST, FPINST2
+					@ and point r4 at the word at the
+					@ start of the register dump
+
+no_old_VFP_process:
+	DBGSTR1	"load state %p", r10
+	str	r10, [r3]		@ update the last_VFP_context pointer
+					@ Load the saved state back into the VFP
+	add	r4, r10, #8*16+4
+	ldmia	r4, {r1, r2, r6, r8}	@ load FPEXC, FPSCR, FPINST, FPINST2
+	VFPFLDMIA r10	 		@ reload the working registers while
+					@ FPEXC is in a safe state
+	tst	r1, #FPEXC_FPV2		@ is there an FPINST2 to write?
+	VFPFMXR	FPINST2, r8, NE		@ FPINST2 if needed - avoids writing
+					@ nonexistant reg on rev0
+	VFPFMXR	FPINST, r6
+	VFPFMXR	FPSCR, r2		@ restore status
+
+check_for_exception:
+	tst	r1, #FPEXC_EXCEPTION
+	bne	process_exception	@ might as well handle the pending
+					@ exception before retrying branch
+					@ out before setting an FPEXC that
+					@ stops us reading stuff
+	VFPFMXR	FPEXC, r1		@ restore FPEXC last
+	sub	r5, r5, #4
+	str	r5, [sp, #S_PC]		@ retry the instruction
+	mov	pc, r9			@ we think we have handled things
+
+
+look_for_VFP_exceptions:
+	tst	r1, #FPEXC_EXCEPTION
+	bne	process_exception
+	VFPFMRX	r2, FPSCR
+	tst	r2, #FPSCR_IXE		@ IXE doesn't set FPEXC_EXCEPTION !
+	bne	process_exception
+
+	@ Fall into hand on to next handler - appropriate coproc instr
+	@ not recognised by VFP
+
+	DBGSTR	"not VFP"
+	mov	pc, lr
+
+process_exception:
+	DBGSTR	"bounce"
+	sub	r5, r5, #4
+	str	r5, [sp, #S_PC]		@ retry the instruction on exit from
+					@ the imprecise exception handling in
+					@ the support code
+	mov	r2, sp			@ nothing stacked - regdump is at TOS
+	mov	lr, r9			@ setup for a return to the user code.
+
+	@ Now call the C code to package up the bounce to the support code
+	@   r0 holds the trigger instruction
+	@   r1 holds the FPEXC value
+	@   r2 pointer to register dump
+	b	VFP9_bounce		@ we have handled this - the support
+					@ code will raise an exception if
+					@ required. If not, the user code will
+					@ retry the faulted instruction
+
+last_VFP_context_address:
+	.word	last_VFP_context
+
+	.globl	vfp_get_float
+vfp_get_float:
+	add	pc, pc, r0, lsl #3
+	mov	r0, r0
+	.irp	dr,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15
+	mrc	p10, 0, r0, c\dr, c0, 0	@ fmrs	r0, s0
+	mov	pc, lr
+	mrc	p10, 0, r0, c\dr, c0, 4	@ fmrs	r0, s1
+	mov	pc, lr
+	.endr
+
+	.globl	vfp_put_float
+vfp_put_float:
+	add	pc, pc, r0, lsl #3
+	mov	r0, r0
+	.irp	dr,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15
+	mcr	p10, 0, r1, c\dr, c0, 0	@ fmsr	r0, s0
+	mov	pc, lr
+	mcr	p10, 0, r1, c\dr, c0, 4	@ fmsr	r0, s1
+	mov	pc, lr
+	.endr
+
+	.globl	vfp_get_double
+vfp_get_double:
+	mov	r0, r0, lsr #1
+	add	pc, pc, r0, lsl #3
+	mov	r0, r0
+	.irp	dr,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15
+	mrrc	p10, 1, r0, r1, c\dr	@ fmrrd	r0, r1, d\dr
+	mov	pc, lr
+	.endr
+
+	.globl	vfp_put_double
+vfp_put_double:
+	mov	r0, r0, lsr #1
+	add	pc, pc, r0, lsl #3
+	mov	r0, r0
+	.irp	dr,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15
+	mcrr	p10, 1, r1, r2, c\dr	@ fmrrd	r1, r2, d\dr
+	mov	pc, lr
+	.endr

arch/arm/vfp/vfpinstr.h

+/*
+ *  linux/arch/arm/vfp/vfpinstr.h
+ *
+ *  Copyright (C) 2004 ARM Limited.
+ *  Written by Deep Blue Solutions Limited.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * VFP instruction masks.
+ */
+#define INST_CPRTDO(inst)	(((inst) & 0x0f000000) == 0x0e000000)
+#define INST_CPRT(inst)		((inst) & (1 << 4))
+#define INST_CPRT_L(inst)	((inst) & (1 << 20))
+#define INST_CPRT_Rd(inst)	(((inst) & (15 << 12)) >> 12)
+#define INST_CPRT_OP(inst)	(((inst) >> 21) & 7)
+#define INST_CPNUM(inst)	((inst) & 0xf00)
+#define CPNUM(cp)		((cp) << 8)
+
+#define FOP_MASK	(0x00b00040)
+#define FOP_FMAC	(0x00000000)
+#define FOP_FNMAC	(0x00000040)
+#define FOP_FMSC	(0x00100000)
+#define FOP_FNMSC	(0x00100040)
+#define FOP_FMUL	(0x00200000)
+#define FOP_FNMUL	(0x00200040)
+#define FOP_FADD	(0x00300000)
+#define FOP_FSUB	(0x00300040)
+#define FOP_FDIV	(0x00800000)
+#define FOP_EXT		(0x00b00040)
+
+#define FOP_TO_IDX(inst)	((inst & 0x00b00000) >> 20 | (inst & (1 << 6)) >> 4)
+
+#define FEXT_MASK	(0x000f0080)
+#define FEXT_FCPY	(0x00000000)
+#define FEXT_FABS	(0x00000080)
+#define FEXT_FNEG	(0x00010000)
+#define FEXT_FSQRT	(0x00010080)
+#define FEXT_FCMP	(0x00040000)
+#define FEXT_FCMPE	(0x00040080)
+#define FEXT_FCMPZ	(0x00050000)
+#define FEXT_FCMPEZ	(0x00050080)
+#define FEXT_FCVT	(0x00070080)
+#define FEXT_FUITO	(0x00080000)
+#define FEXT_FSITO	(0x00080080)
+#define FEXT_FTOUI	(0x000c0000)
+#define FEXT_FTOUIZ	(0x000c0080)
+#define FEXT_FTOSI	(0x000d0000)
+#define FEXT_FTOSIZ	(0x000d0080)
+
+#define FEXT_TO_IDX(inst)	((inst & 0x000f0000) >> 15 | (inst & (1 << 7)) >> 7)
+
+#define vfp_get_sd(inst)	((inst & 0x0000f000) >> 11 | (inst & (1 << 22)) >> 22)
+#define vfp_get_dd(inst)	((inst & 0x0000f000) >> 12)
+#define vfp_get_sm(inst)	((inst & 0x0000000f) << 1 | (inst & (1 << 5)) >> 5)
+#define vfp_get_dm(inst)	((inst & 0x0000000f))
+#define vfp_get_sn(inst)	((inst & 0x000f0000) >> 15 | (inst & (1 << 7)) >> 7)
+#define vfp_get_dn(inst)	((inst & 0x000f0000) >> 16)
+
+#define vfp_single(inst)	(((inst) & 0x0000f00) == 0xa00)
+
+#define FPSCR_N	(1 << 31)
+#define FPSCR_Z	(1 << 30)
+#define FPSCR_C (1 << 29)
+#define FPSCR_V	(1 << 28)
+
+/*
+ * Since we aren't building with -mfpu=vfp, we need to code
+ * these instructions using their MRC/MCR equivalents.
+ */
+#define vfpreg(_vfp_) #_vfp_
+
+#define fmrx(_vfp_) ({			\
+	u32 __v;			\
+	asm("mrc%? p10, 7, %0, " vfpreg(_vfp_) ", cr0, 0 @ fmrx	%0, " #_vfp_	\
+	    : "=r" (__v));		\
+	__v;				\
+ })
+
+#define fmxr(_vfp_,_var_)		\
+	asm("mcr%? p10, 7, %0, " vfpreg(_vfp_) ", cr0, 0 @ fmxr	" #_vfp_ ", %0"	\
+	   : : "r" (_var_))
+
+u32 vfp_single_cpdo(u32 inst, u32 fpscr);
+u32 vfp_single_cprt(u32 inst, u32 fpscr, struct pt_regs *regs);
+
+u32 vfp_double_cpdo(u32 inst, u32 fpscr);

arch/arm/vfp/vfpmodule.c

+/*
+ *  linux/arch/arm/vfp/vfpmodule.c
+ *
+ *  Copyright (C) 2004 ARM Limited.
+ *  Written by Deep Blue Solutions Limited.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+#include <linux/module.h>
+#include <linux/config.h>
+#include <linux/types.h>
+#include <linux/kernel.h>
+#include <linux/signal.h>
+#include <linux/sched.h>
+#include <linux/init.h>
+#include <asm/vfp.h>
+
+#include "vfpinstr.h"
+#include "vfp.h"
+
+/*
+ * Our undef handlers (in entry.S)
+ */
+void vfp_testing_entry(void);
+void vfp_support_entry(void);
+
+void (*vfp_vector)(void) = vfp_testing_entry;
+union vfp_state *last_VFP_context;
+
+/*
+ * Dual-use variable.
+ * Used in startup: set to non-zero if VFP checks fail
+ * After startup, holds VFP architecture
+ */
+unsigned int VFP_arch;
+
+/*
+ * Per-thread VFP initialisation.
+ */
+void vfp_flush_thread(union vfp_state *vfp)
+{
+	memset(vfp, 0, sizeof(union vfp_state));
+
+	vfp->hard.fpexc = FPEXC_ENABLE;
+	vfp->hard.fpscr = FPSCR_ROUND_NEAREST;
+
+	/*
+	 * Disable VFP to ensure we initialise it first.
+	 */
+	fmxr(FPEXC, fmrx(FPEXC) & ~FPEXC_ENABLE);
+
+	/*
+	 * Ensure we don't try to overwrite our newly initialised
+	 * state information on the first fault.
+	 */
+	if (last_VFP_context == vfp)
+		last_VFP_context = NULL;
+}
+
+/*
+ * Per-thread VFP cleanup.
+ */
+void vfp_release_thread(union vfp_state *vfp)
+{
+	if (last_VFP_context == vfp)
+		last_VFP_context = NULL;
+}
+
+/*
+ * Raise a SIGFPE for the current process.
+ * sicode describes the signal being raised.
+ */
+void vfp_raise_sigfpe(unsigned int sicode, struct pt_regs *regs)
+{
+	siginfo_t info;
+
+	memset(&info, 0, sizeof(info));
+
+	info.si_signo = SIGFPE;
+	info.si_code = sicode;
+	info.si_addr = (void *)(instruction_pointer(regs) - 4);
+
+	/*
+	 * This is the same as NWFPE, because it's not clear what
+	 * this is used for
+	 */
+	current->thread.error_code = 0;
+	current->thread.trap_no = 6;
+
+	force_sig_info(SIGFPE, &info, current);
+}
+
+static void vfp_panic(char *reason)
+{
+	int i;
+
+	printk(KERN_ERR "VFP: Error: %s\n", reason);
+	printk(KERN_ERR "VFP: EXC 0x%08x SCR 0x%08x INST 0x%08x\n",
+		fmrx(FPEXC), fmrx(FPSCR), fmrx(FPINST));
+	for (i = 0; i < 32; i += 2)
+		printk(KERN_ERR "VFP: s%2u: 0x%08x s%2u: 0x%08x\n",
+		       i, vfp_get_float(i), i+1, vfp_get_float(i+1));
+}
+
+/*
+ * Process bitmask of exception conditions.
+ */
+static void vfp_raise_exceptions(u32 exceptions, u32 inst, u32 fpscr, struct pt_regs *regs)
+{
+	int si_code = 0;
+
+	pr_debug("VFP: raising exceptions %08x\n", exceptions);
+
+	if (exceptions == (u32)-1) {
+		vfp_panic("unhandled bounce");
+		vfp_raise_sigfpe(0, regs);
+		return;
+	}
+
+	/*
+	 * If any of the status flags are set, update the FPSCR.
+	 * Comparison instructions always return at least one of
+	 * these flags set.
+	 */
+	if (exceptions & (FPSCR_N|FPSCR_Z|FPSCR_C|FPSCR_V))
+		fpscr &= ~(FPSCR_N|FPSCR_Z|FPSCR_C|FPSCR_V);
+
+	fpscr |= exceptions;
+
+	fmxr(FPSCR, fpscr);
+
+#define RAISE(stat,en,sig)				\
+	if (exceptions & stat && fpscr & en)		\
+		si_code = sig;
+
+	/*
+	 * These are arranged in priority order, least to highest.
+	 */
+	RAISE(FPSCR_IXC, FPSCR_IXE, FPE_FLTRES);
+	RAISE(FPSCR_UFC, FPSCR_UFE, FPE_FLTUND);
+	RAISE(FPSCR_OFC, FPSCR_OFE, FPE_FLTOVF);
+	RAISE(FPSCR_IOC, FPSCR_IOE, FPE_FLTINV);
+
+	if (si_code)
+		vfp_raise_sigfpe(si_code, regs);
+}
+
+/*
+ * Emulate a VFP instruction.
+ */
+static u32 vfp_emulate_instruction(u32 inst, u32 fpscr, struct pt_regs *regs)
+{
+	u32 exceptions = (u32)-1;
+
+	pr_debug("VFP: emulate: INST=0x%08x SCR=0x%08x\n", inst, fpscr);
+
+	if (INST_CPRTDO(inst)) {
+		if (!INST_CPRT(inst)) {
+			/*
+			 * CPDO
+			 */
+			if (vfp_single(inst)) {
+				exceptions = vfp_single_cpdo(inst, fpscr);
+			} else {
+				exceptions = vfp_double_cpdo(inst, fpscr);
+			}
+		} else {
+			/*
+			 * A CPRT instruction can not appear in FPINST2, nor
+			 * can it cause an exception.  Therefore, we do not
+			 * have to emulate it.
+			 */
+		}
+	} else {
+		/*
+		 * A CPDT instruction can not appear in FPINST2, nor can
+		 * it cause an exception.  Therefore, we do not have to
+		 * emulate it.
+		 */
+	}
+	return exceptions;
+}
+
+/*
+ * Package up a bounce condition.
+ */
+void VFP9_bounce(u32 trigger, u32 fpexc, struct pt_regs *regs)
+{
+	u32 fpscr, orig_fpscr, exceptions, inst;
+
+	pr_debug("VFP: bounce: trigger %08x fpexc %08x\n", trigger, fpexc);
+
+	/*
+	 * Enable access to the VFP so we can handle the bounce.
+	 */
+	fmxr(FPEXC, fpexc & ~(FPEXC_EXCEPTION|FPEXC_INV|FPEXC_UFC|FPEXC_IOC));
+
+	orig_fpscr = fpscr = fmrx(FPSCR);
+
+	/*
+	 * If we are running with inexact exceptions enabled, we need to
+	 * emulate the trigger instruction.  Note that as we're emulating
+	 * the trigger instruction, we need to increment PC.
+	 */
+	if (fpscr & FPSCR_IXE) {
+		regs->ARM_pc += 4;
+		goto emulate;
+	}
+
+	barrier();
+
+	/*
+	 * Modify fpscr to indicate the number of iterations remaining
+	 */
+	if (fpexc & FPEXC_EXCEPTION) {
+		u32 len;
+
+		len = fpexc + (1 << FPEXC_LENGTH_BIT);
+
+		fpscr &= ~FPSCR_LENGTH_MASK;
+		fpscr |= (len & FPEXC_LENGTH_MASK) << (FPSCR_LENGTH_BIT - FPEXC_LENGTH_BIT);
+	}
+
+	/*
+	 * Handle the first FP instruction.  We used to take note of the
+	 * FPEXC bounce reason, but this appears to be unreliable.
+	 * Emulate the bounced instruction instead.
+	 */
+	inst = fmrx(FPINST);
+	exceptions = vfp_emulate_instruction(inst, fpscr, regs);
+	if (exceptions)
+		vfp_raise_exceptions(exceptions, inst, orig_fpscr, regs);
+
+	/*
+	 * If there isn't a second FP instruction, exit now.
+	 */
+	if (!(fpexc & FPEXC_FPV2))
+		return;
+
+	/*
+	 * The barrier() here prevents fpinst2 being read
+	 * before the condition above.
+	 */
+	barrier();
+	trigger = fmrx(FPINST2);
+	fpscr = fmrx(FPSCR);
+
+ emulate:
+	exceptions = vfp_emulate_instruction(trigger, fpscr, regs);
+	if (exceptions)
+		vfp_raise_exceptions(exceptions, trigger, orig_fpscr, regs);
+}
+ 
+/*
+ * VFP support code initialisation.
+ */
+static int __init vfp_init(void)
+{
+	unsigned int vfpsid;
+
+	/*
+	 * First check that there is a VFP that we can use.
+	 * The handler is already setup to just log calls, so
+	 * we just need to read the VFPSID register.
+	 */
+	vfpsid = fmrx(FPSID);
+
+	printk(KERN_INFO "VFP support v0.3: ");
+	if (VFP_arch) {
+		printk("not present\n");
+	} else if (vfpsid & FPSID_NODOUBLE) {
+		printk("no double precision support\n");
+	} else {
+		VFP_arch = (vfpsid & FPSID_ARCH_MASK) >> FPSID_ARCH_BIT;  /* Extract the architecture version */
+		printk("implementor %02x architecture %d part %02x variant %x rev %x\n",
+			(vfpsid & FPSID_IMPLEMENTER_MASK) >> FPSID_IMPLEMENTER_BIT,
+			(vfpsid & FPSID_ARCH_MASK) >> FPSID_ARCH_BIT,
+			(vfpsid & FPSID_PART_MASK) >> FPSID_PART_BIT,
+			(vfpsid & FPSID_VARIANT_MASK) >> FPSID_VARIANT_BIT,
+			(vfpsid & FPSID_REV_MASK) >> FPSID_REV_BIT);
+		vfp_vector = vfp_support_entry;
+	}
+	return 0;
+}
+
+late_initcall(vfp_init);

arch/arm/vfp/vfpsingle.c

+/*
+ *  linux/arch/arm/vfp/vfpsingle.c
+ *
+ * This code is derived in part from John R. Housers softfloat library, which
+ * carries the following notice:
+ *
+ * ===========================================================================
+ * This C source file is part of the SoftFloat IEC/IEEE Floating-point
+ * Arithmetic Package, Release 2.
+ *
+ * Written by John R. Hauser.  This work was made possible in part by the
+ * International Computer Science Institute, located at Suite 600, 1947 Center
+ * Street, Berkeley, California 94704.  Funding was partially provided by the
+ * National Science Foundation under grant MIP-9311980.  The original version
+ * of this code was written as part of a project to build a fixed-point vector
+ * processor in collaboration with the University of California at Berkeley,
+ * overseen by Profs. Nelson Morgan and John Wawrzynek.  More information
+ * is available through the web page `http://HTTP.CS.Berkeley.EDU/~jhauser/
+ * arithmetic/softfloat.html'.
+ *
+ * THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE.  Although reasonable effort
+ * has been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT
+ * TIMES RESULT IN INCORRECT BEHAVIOR.  USE OF THIS SOFTWARE IS RESTRICTED TO
+ * PERSONS AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ANY
+ * AND ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM ITS USE.
+ *
+ * Derivative works are acceptable, even for commercial purposes, so long as
+ * (1) they include prominent notice that the work is derivative, and (2) they
+ * include prominent notice akin to these three paragraphs for those parts of
+ * this code that are retained.
+ * ===========================================================================
+ */
+#include <linux/kernel.h>
+#include <asm/bitops.h>
+#include <asm/ptrace.h>
+#include <asm/vfp.h>
+
+#include "vfpinstr.h"
+#include "vfp.h"
+
+static struct vfp_single vfp_single_default_qnan = {
+	.exponent	= 255,
+	.sign		= 0,
+	.significand	= VFP_SINGLE_SIGNIFICAND_QNAN,
+};
+
+static void vfp_single_dump(const char *str, struct vfp_single *s)
+{
+	pr_debug("VFP: %s: sign=%d exponent=%d significand=%08x\n",
+		 str, s->sign != 0, s->exponent, s->significand);
+}
+
+static void vfp_single_normalise_denormal(struct vfp_single *vs)
+{
+	int bits = 31 - fls(vs->significand);
+
+	vfp_single_dump("normalise_denormal: in", vs);
+
+	if (bits) {
+		vs->exponent -= bits - 1;
+		vs->significand <<= bits;
+	}
+
+	vfp_single_dump("normalise_denormal: out", vs);
+}
+
+#ifndef DEBUG
+#define vfp_single_normaliseround(sd,vsd,fpscr,except,func) __vfp_single_normaliseround(sd,vsd,fpscr,except)
+u32 __vfp_single_normaliseround(int sd, struct vfp_single *vs, u32 fpscr, u32 exceptions)
+#else
+u32 vfp_single_normaliseround(int sd, struct vfp_single *vs, u32 fpscr, u32 exceptions, const char *func)
+#endif
+{
+	u32 significand, incr, rmode;
+	int exponent, shift, underflow;
+
+	vfp_single_dump("pack: in", vs);
+
+	/*
+	 * Infinities and NaNs are a special case.
+	 */
+	if (vs->exponent == 255 && (vs->significand == 0 || exceptions))
+		goto pack;
+
+	/*
+	 * Special-case zero.
+	 */
+	if (vs->significand == 0) {
+		vs->exponent = 0;
+		goto pack;
+	}
+
+	exponent = vs->exponent;
+	significand = vs->significand;
+
+	/*
+	 * Normalise first.  Note that we shift the significand up to
+	 * bit 31, so we have VFP_SINGLE_LOW_BITS + 1 below the least
+	 * significant bit.
+	 */
+	shift = 32 - fls(significand);
+	if (shift < 32 && shift) {
+		exponent -= shift;
+		significand <<= shift;
+	}
+
+#ifdef DEBUG
+	vs->exponent = exponent;
+	vs->significand = significand;
+	vfp_single_dump("pack: normalised", vs);
+#endif
+
+	/*
+	 * Tiny number?
+	 */
+	underflow = exponent < 0;
+	if (underflow) {
+		significand = vfp_shiftright32jamming(significand, -exponent);
+		exponent = 0;
+#ifdef DEBUG
+		vs->exponent = exponent;
+		vs->significand = significand;
+		vfp_single_dump("pack: tiny number", vs);
+#endif
+		if (!(significand & ((1 << (VFP_SINGLE_LOW_BITS + 1)) - 1)))
+			underflow = 0;
+	}
+
+	/*
+	 * Select rounding increment.
+	 */
+	incr = 0;
+	rmode = fpscr & FPSCR_RMODE_MASK;
+
+	if (rmode == FPSCR_ROUND_NEAREST) {
+		incr = 1 << VFP_SINGLE_LOW_BITS;
+		if ((significand & (1 << (VFP_SINGLE_LOW_BITS + 1))) == 0)
+			incr -= 1;
+	} else if (rmode == FPSCR_ROUND_TOZERO) {
+		incr = 0;
+	} else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vs->sign != 0))
+		incr = (1 << (VFP_SINGLE_LOW_BITS + 1)) - 1;
+
+	pr_debug("VFP: rounding increment = 0x%08x\n", incr);
+
+	/*
+	 * Is our rounding going to overflow?
+	 */
+	if ((significand + incr) < significand) {
+		exponent += 1;
+		significand = (significand >> 1) | (significand & 1);
+		incr >>= 1;
+#ifdef DEBUG
+		vs->exponent = exponent;
+		vs->significand = significand;
+		vfp_single_dump("pack: overflow", vs);
+#endif
+	}
+
+	/*
+	 * If any of the low bits (which will be shifted out of the
+	 * number) are non-zero, the result is inexact.
+	 */
+	if (significand & ((1 << (VFP_SINGLE_LOW_BITS + 1)) - 1))
+		exceptions |= FPSCR_IXC;
+
+	/*
+	 * Do our rounding.
+	 */
+	significand += incr;
+
+	/*
+	 * Infinity?
+	 */
+	if (exponent >= 254) {
+		exceptions |= FPSCR_OFC | FPSCR_IXC;
+		if (incr == 0) {
+			vs->exponent = 253;
+			vs->significand = 0x7fffffff;
+		} else {
+			vs->exponent = 255;		/* infinity */
+			vs->significand = 0;
+		}
+	} else {
+		if (significand >> (VFP_SINGLE_LOW_BITS + 1) == 0)
+			exponent = 0;
+		if (exponent || significand > 0x80000000)
+			underflow = 0;
+		if (underflow)
+			exceptions |= FPSCR_UFC;
+		vs->exponent = exponent;
+		vs->significand = significand >> 1;
+	}
+
+ pack:
+	vfp_single_dump("pack: final", vs);
+	{
+		s32 d = vfp_single_pack(vs);
+		pr_debug("VFP: %s: d(s%d)=%08x exceptions=%08x\n", func,
+			 sd, d, exceptions);
+		vfp_put_float(sd, d);
+	}
+
+	return exceptions;
+}
+
+/*
+ * Propagate the NaN, setting exceptions if it is signalling.
+ * 'n' is always a NaN.  'm' may be a number, NaN or infinity.
+ */
+static u32
+vfp_propagate_nan(struct vfp_single *vsd, struct vfp_single *vsn,
+		  struct vfp_single *vsm, u32 fpscr)
+{
+	struct vfp_single *nan;
+	int tn, tm = 0;
+
+	tn = vfp_single_type(vsn);
+
+	if (vsm)
+		tm = vfp_single_type(vsm);
+
+	if (fpscr & FPSCR_DEFAULT_NAN)
+		/*
+		 * Default NaN mode - always returns a quiet NaN
+		 */
+		nan = &vfp_single_default_qnan;
+	else {
+		/*
+		 * Contemporary mode - select the first signalling
+		 * NAN, or if neither are signalling, the first
+		 * quiet NAN.
+		 */
+		if (tn == VFP_SNAN || (tm != VFP_SNAN && tn == VFP_QNAN))
+			nan = vsn;
+		else
+			nan = vsm;
+		/*
+		 * Make the NaN quiet.
+		 */
+		nan->significand |= VFP_SINGLE_SIGNIFICAND_QNAN;
+	}
+
+	*vsd = *nan;
+
+	/*
+	 * If one was a signalling NAN, raise invalid operation.
+	 */
+	return tn == VFP_SNAN || tm == VFP_SNAN ? FPSCR_IOC : 0x100;
+}
+
+
+/*
+ * Extended operations
+ */
+static u32 vfp_single_fabs(int sd, int unused, s32 m, u32 fpscr)
+{
+	vfp_put_float(sd, vfp_single_packed_abs(m));
+	return 0;
+}
+
+static u32 vfp_single_fcpy(int sd, int unused, s32 m, u32 fpscr)
+{
+	vfp_put_float(sd, m);
+	return 0;
+}
+
+static u32 vfp_single_fneg(int sd, int unused, s32 m, u32 fpscr)
+{
+	vfp_put_float(sd, vfp_single_packed_negate(m));
+	return 0;
+}
+
+static const u16 sqrt_oddadjust[] = {
+	0x0004, 0x0022, 0x005d, 0x00b1, 0x011d, 0x019f, 0x0236, 0x02e0,
+	0x039c, 0x0468, 0x0545, 0x0631, 0x072b, 0x0832, 0x0946, 0x0a67
+};
+
+static const u16 sqrt_evenadjust[] = {
+	0x0a2d, 0x08af, 0x075a, 0x0629, 0x051a, 0x0429, 0x0356, 0x029e,
+	0x0200, 0x0179, 0x0109, 0x00af, 0x0068, 0x0034, 0x0012, 0x0002
+};
+
+u32 vfp_estimate_sqrt_significand(u32 exponent, u32 significand)
+{
+	int index;
+	u32 z, a;
+
+	if ((significand & 0xc0000000) != 0x40000000) {
+		printk(KERN_WARNING "VFP: estimate_sqrt: invalid significand\n");
+	}
+
+	a = significand << 1;
+	index = (a >> 27) & 15;
+	if (exponent & 1) {
+		z = 0x4000 + (a >> 17) - sqrt_oddadjust[index];
+		z = ((a / z) << 14) + (z << 15);
+		a >>= 1;
+	} else {
+		z = 0x8000 + (a >> 17) - sqrt_evenadjust[index];
+		z = a / z + z;
+		z = (z >= 0x20000) ? 0xffff8000 : (z << 15);
+		if (z <= a)
+			return (s32)a >> 1;
+	}
+	return (u32)(((u64)a << 31) / z) + (z >> 1);
+}
+
+static u32 vfp_single_fsqrt(int sd, int unused, s32 m, u32 fpscr)
+{
+	struct vfp_single vsm, vsd;
+	int ret, tm;
+
+	vfp_single_unpack(&vsm, m);
+	tm = vfp_single_type(&vsm);
+	if (tm & (VFP_NAN|VFP_INFINITY)) {
+		struct vfp_single *vsp = &vsd;
+
+		if (tm & VFP_NAN)
+			ret = vfp_propagate_nan(vsp, &vsm, NULL, fpscr);
+		else if (vsm.sign == 0) {
+ sqrt_copy:
+			vsp = &vsm;
+			ret = 0;
+		} else {
+ sqrt_invalid:
+			vsp = &vfp_single_default_qnan;
+			ret = FPSCR_IOC;
+		}
+		vfp_put_float(sd, vfp_single_pack(vsp));
+		return ret;
+	}
+
+	/*
+	 * sqrt(+/- 0) == +/- 0
+	 */
+	if (tm & VFP_ZERO)
+		goto sqrt_copy;
+
+	/*
+	 * Normalise a denormalised number
+	 */
+	if (tm & VFP_DENORMAL)
+		vfp_single_normalise_denormal(&vsm);
+
+	/*
+	 * sqrt(<0) = invalid
+	 */
+	if (vsm.sign)
+		goto sqrt_invalid;
+
+	vfp_single_dump("sqrt", &vsm);
+
+	/*
+	 * Estimate the square root.
+	 */
+	vsd.sign = 0;
+	vsd.exponent = ((vsm.exponent - 127) >> 1) + 127;
+	vsd.significand = vfp_estimate_sqrt_significand(vsm.exponent, vsm.significand) + 2;
+
+	vfp_single_dump("sqrt estimate", &vsd);
+
+	/*
+	 * And now adjust.
+	 */
+	if ((vsd.significand & VFP_SINGLE_LOW_BITS_MASK) <= 5) {
+		if (vsd.significand < 2) {
+			vsd.significand = 0xffffffff;
+		} else {
+			u64 term;
+			s64 rem;
+			vsm.significand <<= !(vsm.exponent & 1);
+			term = (u64)vsd.significand * vsd.significand;
+			rem = ((u64)vsm.significand << 32) - term;
+
+			pr_debug("VFP: term=%016llx rem=%016llx\n", term, rem);
+
+			while (rem < 0) {
+				vsd.significand -= 1;
+				rem += ((u64)vsd.significand << 1) | 1;
+			}
+			vsd.significand |= rem != 0;
+		}
+	}
+	vsd.significand = vfp_shiftright32jamming(vsd.significand, 1);
+
+	return vfp_single_normaliseround(sd, &vsd, fpscr, 0, "fsqrt");
+}
+
+/*
+ * Equal	:= ZC
+ * Less than	:= N
+ * Greater than	:= C
+ * Unordered	:= CV
+ */
+static u32 vfp_compare(int sd, int signal_on_qnan, s32 m, u32 fpscr)
+{
+	s32 d;
+	u32 ret = 0;
+
+	d = vfp_get_float(sd);
+	if (vfp_single_packed_exponent(m) == 255 && vfp_single_packed_mantissa(m)) {
+		ret |= FPSCR_C | FPSCR_V;
+		if (signal_on_qnan || !(vfp_single_packed_mantissa(m) & (1 << (VFP_SINGLE_MANTISSA_BITS - 1))))
+			/*
+			 * Signalling NaN, or signalling on quiet NaN
+			 */
+			ret |= FPSCR_IOC;
+	}
+
+	if (vfp_single_packed_exponent(d) == 255 && vfp_single_packed_mantissa(d)) {
+		ret |= FPSCR_C | FPSCR_V;
+		if (signal_on_qnan || !(vfp_single_packed_mantissa(d) & (1 << (VFP_SINGLE_MANTISSA_BITS - 1))))
+			/*
+			 * Signalling NaN, or signalling on quiet NaN
+			 */
+			ret |= FPSCR_IOC;
+	}
+
+	if (ret == 0) {
+		if (d == m || vfp_single_packed_abs(d | m) == 0) {
+			/*
+			 * equal
+			 */
+			ret |= FPSCR_Z | FPSCR_C;
+		} else if (vfp_single_packed_sign(d ^ m)) {
+			/*
+			 * different signs
+			 */
+			if (vfp_single_packed_sign(d))
+				/*
+				 * d is negative, so d < m
+				 */
+				ret |= FPSCR_N;
+			else
+				/*
+				 * d is positive, so d > m
+				 */
+				ret |= FPSCR_C;
+		} else if ((vfp_single_packed_sign(d) != 0) ^ (d < m)) {
+			/*
+			 * d < m
+			 */
+			ret |= FPSCR_N;
+		} else if ((vfp_single_packed_sign(d) != 0) ^ (d > m)) {
+			/*
+			 * d > m
+			 */
+			ret |= FPSCR_C;
+		}
+	}
+	return ret;
+}
+
+static u32 vfp_single_fcmp(int sd, int unused, s32 m, u32 fpscr)
+{
+	return vfp_compare(sd, 0, m, fpscr);
+}
+
+static u32 vfp_single_fcmpe(int sd, int unused, s32 m, u32 fpscr)
+{
+	return vfp_compare(sd, 1, m, fpscr);
+}
+
+static u32 vfp_single_fcmpz(int sd, int unused, s32 m, u32 fpscr)
+{
+	return vfp_compare(sd, 0, 0, fpscr);
+}
+
+static u32 vfp_single_fcmpez(int sd, int unused, s32 m, u32 fpscr)
+{
+	return vfp_compare(sd, 1, 0, fpscr);
+}
+
+static u32 vfp_single_fcvtd(int dd, int unused, s32 m, u32 fpscr)
+{
+	struct vfp_single vsm;
+	struct vfp_double vdd;
+	int tm;
+	u32 exceptions = 0;
+
+	vfp_single_unpack(&vsm, m);
+
+	tm = vfp_single_type(&vsm);
+
+	/*
+	 * If we have a signalling NaN, signal invalid operation.
+	 */
+	if (tm == VFP_SNAN)
+		exceptions = FPSCR_IOC;
+
+	if (tm & VFP_DENORMAL)
+		vfp_single_normalise_denormal(&vsm);
+
+	vdd.sign = vsm.sign;
+	vdd.significand = (u64)vsm.significand << 32;
+
+	/*
+	 * If we have an infinity or NaN, the exponent must be 2047.
+	 */
+	if (tm & (VFP_INFINITY|VFP_NAN)) {
+		vdd.exponent = 2047;
+		if (tm & VFP_NAN)
+			vdd.significand |= VFP_DOUBLE_SIGNIFICAND_QNAN;
+		goto pack_nan;
+	} else if (tm & VFP_ZERO)
+		vdd.exponent = 0;
+	else
+		vdd.exponent = vsm.exponent + (1023 - 127);
+
+	/*
+	 * Technically, if bit 0 of dd is set, this is an invalid
+	 * instruction.  However, we ignore this for efficiency.
+	 */
+	return vfp_double_normaliseround(dd, &vdd, fpscr, exceptions, "fcvtd");
+
+ pack_nan:
+	vfp_put_double(dd, vfp_double_pack(&vdd));
+	return exceptions;
+}
+
+static u32 vfp_single_fuito(int sd, int unused, s32 m, u32 fpscr)
+{
+	struct vfp_single vs;
+
+	vs.sign = 0;
+	vs.exponent = 127 + 31 - 1;
+	vs.significand = (u32)m;
+
+	return vfp_single_normaliseround(sd, &vs, fpscr, 0, "fuito");
+}
+
+static u32 vfp_single_fsito(int sd, int unused, s32 m, u32 fpscr)
+{
+	struct vfp_single vs;
+
+	vs.sign = (m & 0x80000000) >> 16;
+	vs.exponent = 127 + 31 - 1;
+	vs.significand = vs.sign ? -m : m;
+
+	return vfp_single_normaliseround(sd, &vs, fpscr, 0, "fsito");
+}
+
+static u32 vfp_single_ftoui(int sd, int unused, s32 m, u32 fpscr)
+{
+	struct vfp_single vsm;
+	u32 d, exceptions = 0;
+	int rmode = fpscr & FPSCR_RMODE_MASK;
+	int tm;
+
+	vfp_single_unpack(&vsm, m);
+	vfp_single_dump("VSM", &vsm);
+
+	/*
+	 * Do we have a denormalised number?
+	 */
+	tm = vfp_single_type(&vsm);
+	if (tm & VFP_DENORMAL)
+		exceptions |= FPSCR_IDC;
+
+	if (tm & VFP_NAN)
+		vsm.sign = 0;
+
+	if (vsm.exponent >= 127 + 32) {
+		d = vsm.sign ? 0 : 0xffffffff;
+		exceptions = FPSCR_IOC;
+	} else if (vsm.exponent >= 127 - 1) {
+		int shift = 127 + 31 - vsm.exponent;
+		u32 rem, incr = 0;
+
+		/*
+		 * 2^0 <= m < 2^32-2^8
+		 */
+		d = (vsm.significand << 1) >> shift;
+		rem = vsm.significand << (33 - shift);
+
+		if (rmode == FPSCR_ROUND_NEAREST) {
+			incr = 0x80000000;
+			if ((d & 1) == 0)
+				incr -= 1;
+		} else if (rmode == FPSCR_ROUND_TOZERO) {
+			incr = 0;
+		} else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vsm.sign != 0)) {
+			incr = ~0;
+		}
+
+		if ((rem + incr) < rem) {
+			if (d < 0xffffffff)
+				d += 1;
+			else
+				exceptions |= FPSCR_IOC;
+		}
+
+		if (d && vsm.sign) {
+			d = 0;
+			exceptions |= FPSCR_IOC;
+		} else if (rem)
+			exceptions |= FPSCR_IXC;
+	} else {
+		d = 0;
+		if (vsm.exponent | vsm.significand) {
+			exceptions |= FPSCR_IXC;
+			if (rmode == FPSCR_ROUND_PLUSINF && vsm.sign == 0)
+				d = 1;
+			else if (rmode == FPSCR_ROUND_MINUSINF && vsm.sign) {
+				d = 0;
+				exceptions |= FPSCR_IOC;
+			}
+		}
+	}
+
+	pr_debug("VFP: ftoui: d(s%d)=%08x exceptions=%08x\n", sd, d, exceptions);
+
+	vfp_put_float(sd, d);
+
+	return exceptions;
+}
+
+static u32 vfp_single_ftouiz(int sd, int unused, s32 m, u32 fpscr)
+{
+	return vfp_single_ftoui(sd, unused, m, FPSCR_ROUND_TOZERO);
+}
+
+static u32 vfp_single_ftosi(int sd, int unused, s32 m, u32 fpscr)
+{
+	struct vfp_single vsm;
+	u32 d, exceptions = 0;
+	int rmode = fpscr & FPSCR_RMODE_MASK;
+
+	vfp_single_unpack(&vsm, m);
+	vfp_single_dump("VSM", &vsm);
+
+	/*
+	 * Do we have a denormalised number?
+	 */
+	if (vfp_single_type(&vsm) & VFP_DENORMAL)
+		exceptions |= FPSCR_IDC;
+
+	if (vsm.exponent >= 127 + 32) {
+		/*
+		 * m >= 2^31-2^7: invalid
+		 */
+		d = 0x7fffffff;
+		if (vsm.sign)
+			d = ~d;
+		exceptions |= FPSCR_IOC;
+	} else if (vsm.exponent >= 127 - 1) {
+		int shift = 127 + 31 - vsm.exponent;
+		u32 rem, incr = 0;
+
+		/* 2^0 <= m <= 2^31-2^7 */
+		d = (vsm.significand << 1) >> shift;
+		rem = vsm.significand << (33 - shift);
+
+		if (rmode == FPSCR_ROUND_NEAREST) {
+			incr = 0x80000000;
+			if ((d & 1) == 0)
+				incr -= 1;
+		} else if (rmode == FPSCR_ROUND_TOZERO) {
+			incr = 0;
+		} else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vsm.sign != 0)) {
+			incr = ~0;
+		}
+
+		if ((rem + incr) < rem && d < 0xffffffff)
+			d += 1;
+		if (d > 0x7fffffff + (vsm.sign != 0)) {
+			d = 0x7fffffff + (vsm.sign != 0);
+			exceptions |= FPSCR_IOC;
+		} else if (rem)
+			exceptions |= FPSCR_IXC;
+
+		if (vsm.sign)
+			d = -d;
+	} else {
+		d = 0;
+		if (vsm.exponent | vsm.significand) {
+			exceptions |= FPSCR_IXC;
+			if (rmode == FPSCR_ROUND_PLUSINF && vsm.sign == 0)
+				d = 1;
+			else if (rmode == FPSCR_ROUND_MINUSINF && vsm.sign)
+				d = -1;
+		}
+	}
+
+	pr_debug("VFP: ftosi: d(s%d)=%08x exceptions=%08x\n", sd, d, exceptions);
+
+	vfp_put_float(sd, (s32)d);
+
+	return exceptions;
+}
+
+static u32 vfp_single_ftosiz(int sd, int unused, s32 m, u32 fpscr)
+{
+	return vfp_single_ftosi(sd, unused, m, FPSCR_ROUND_TOZERO);
+}
+
+static u32 (* const fop_extfns[32])(int sd, int unused, s32 m, u32 fpscr) = {
+	[FEXT_TO_IDX(FEXT_FCPY)]	= vfp_single_fcpy,
+	[FEXT_TO_IDX(FEXT_FABS)]	= vfp_single_fabs,
+	[FEXT_TO_IDX(FEXT_FNEG)]	= vfp_single_fneg,
+	[FEXT_TO_IDX(FEXT_FSQRT)]	= vfp_single_fsqrt,
+	[FEXT_TO_IDX(FEXT_FCMP)]	= vfp_single_fcmp,
+	[FEXT_TO_IDX(FEXT_FCMPE)]	= vfp_single_fcmpe,
+	[FEXT_TO_IDX(FEXT_FCMPZ)]	= vfp_single_fcmpz,
+	[FEXT_TO_IDX(FEXT_FCMPEZ)]	= vfp_single_fcmpez,
+	[FEXT_TO_IDX(FEXT_FCVT)]	= vfp_single_fcvtd,
+	[FEXT_TO_IDX(FEXT_FUITO)]	= vfp_single_fuito,
+	[FEXT_TO_IDX(FEXT_FSITO)]	= vfp_single_fsito,
+	[FEXT_TO_IDX(FEXT_FTOUI)]	= vfp_single_ftoui,
+	[FEXT_TO_IDX(FEXT_FTOUIZ)]	= vfp_single_ftouiz,
+	[FEXT_TO_IDX(FEXT_FTOSI)]	= vfp_single_ftosi,
+	[FEXT_TO_IDX(FEXT_FTOSIZ)]	= vfp_single_ftosiz,
+};
+
+
+
+
+
+static u32
+vfp_single_fadd_nonnumber(struct vfp_single *vsd, struct vfp_single *vsn,
+			  struct vfp_single *vsm, u32 fpscr)
+{
+	struct vfp_single *vsp;
+	u32 exceptions = 0;
+	int tn, tm;
+
+	tn = vfp_single_type(vsn);
+	tm = vfp_single_type(vsm);
+
+	if (tn & tm & VFP_INFINITY) {
+		/*
+		 * Two infinities.  Are they different signs?
+		 */
+		if (vsn->sign ^ vsm->sign) {
+			/*
+			 * different signs -> invalid
+			 */
+			exceptions = FPSCR_IOC;
+			vsp = &vfp_single_default_qnan;
+		} else {
+			/*
+			 * same signs -> valid
+			 */
+			vsp = vsn;
+		}
+	} else if (tn & VFP_INFINITY && tm & VFP_NUMBER) {
+		/*
+		 * One infinity and one number -> infinity
+		 */
+		vsp = vsn;
+	} else {
+		/*
+		 * 'n' is a NaN of some type
+		 */
+		return vfp_propagate_nan(vsd, vsn, vsm, fpscr);
+	}
+	*vsd = *vsp;
+	return exceptions;
+}
+
+static u32
+vfp_single_add(struct vfp_single *vsd, struct vfp_single *vsn,
+	       struct vfp_single *vsm, u32 fpscr)
+{
+	u32 exp_diff, m_sig;
+
+	if (vsn->significand & 0x80000000 ||
+	    vsm->significand & 0x80000000) {
+		pr_info("VFP: bad FP values in %s\n", __func__);
+		vfp_single_dump("VSN", vsn);
+		vfp_single_dump("VSM", vsm);
+	}
+
+	/*
+	 * Ensure that 'n' is the largest magnitude number.  Note that
+	 * if 'n' and 'm' have equal exponents, we do not swap them.
+	 * This ensures that NaN propagation works correctly.
+	 */
+	if (vsn->exponent < vsm->exponent) {
+		struct vfp_single *t = vsn;
+		vsn = vsm;
+		vsm = t;
+	}
+
+	/*
+	 * Is 'n' an infinity or a NaN?  Note that 'm' may be a number,
+	 * infinity or a NaN here.
+	 */
+	if (vsn->exponent == 255)
+		return vfp_single_fadd_nonnumber(vsd, vsn, vsm, fpscr);
+
+	/*
+	 * We have two proper numbers, where 'vsn' is the larger magnitude.
+	 *
+	 * Copy 'n' to 'd' before doing the arithmetic.
+	 */
+	*vsd = *vsn;
+
+	/*
+	 * Align both numbers.
+	 */
+	exp_diff = vsn->exponent - vsm->exponent;
+	m_sig = vfp_shiftright32jamming(vsm->significand, exp_diff);
+
+	/*
+	 * If the signs are different, we are really subtracting.
+	 */
+	if (vsn->sign ^ vsm->sign) {
+		m_sig = vsn->significand - m_sig;
+		if ((s32)m_sig < 0) {
+			vsd->sign = vfp_sign_negate(vsd->sign);
+			m_sig = -m_sig;
+		} else if (m_sig == 0) {
+			vsd->sign = (fpscr & FPSCR_RMODE_MASK) ==
+				      FPSCR_ROUND_MINUSINF ? 0x8000 : 0;
+		}
+	} else {
+		m_sig = vsn->significand + m_sig;
+	}
+	vsd->significand = m_sig;
+
+	return 0;
+}
+
+static u32
+vfp_single_multiply(struct vfp_single *vsd, struct vfp_single *vsn, struct vfp_single *vsm, u32 fpscr)
+{
+	vfp_single_dump("VSN", vsn);
+	vfp_single_dump("VSM", vsm);
+
+	/*
+	 * Ensure that 'n' is the largest magnitude number.  Note that
+	 * if 'n' and 'm' have equal exponents, we do not swap them.
+	 * This ensures that NaN propagation works correctly.
+	 */
+	if (vsn->exponent < vsm->exponent) {
+		struct vfp_single *t = vsn;
+		vsn = vsm;
+		vsm = t;
+		pr_debug("VFP: swapping M <-> N\n");
+	}
+
+	vsd->sign = vsn->sign ^ vsm->sign;
+
+	/*
+	 * If 'n' is an infinity or NaN, handle it.  'm' may be anything.
+	 */
+	if (vsn->exponent == 255) {
+		if (vsn->significand || (vsm->exponent == 255 && vsm->significand))
+			return vfp_propagate_nan(vsd, vsn, vsm, fpscr);
+		if ((vsm->exponent | vsm->significand) == 0) {
+			*vsd = vfp_single_default_qnan;
+			return FPSCR_IOC;
+		}
+		vsd->exponent = vsn->exponent;
+		vsd->significand = 0;
+		return 0;
+	}
+
+	/*
+	 * If 'm' is zero, the result is always zero.  In this case,
+	 * 'n' may be zero or a number, but it doesn't matter which.
+	 */
+	if ((vsm->exponent | vsm->significand) == 0) {
+		vsd->exponent = 0;
+		vsd->significand = 0;
+		return 0;
+	}
+
+	/*
+	 * We add 2 to the destination exponent for the same reason as
+	 * the addition case - though this time we have +1 from each
+	 * input operand.
+	 */
+	vsd->exponent = vsn->exponent + vsm->exponent - 127 + 2;
+	vsd->significand = vfp_hi64to32jamming((u64)vsn->significand * vsm->significand);
+
+	vfp_single_dump("VSD", vsd);
+	return 0;
+}
+
+#define NEG_MULTIPLY	(1 << 0)
+#define NEG_SUBTRACT	(1 << 1)
+
+static u32
+vfp_single_multiply_accumulate(int sd, int sn, s32 m, u32 fpscr, u32 negate, char *func)
+{
+	struct vfp_single vsd, vsp, vsn, vsm;
+	u32 exceptions;
+	s32 v;
+
+	v = vfp_get_float(sn);
+	pr_debug("VFP: s%u = %08x\n", sn, v);
+	vfp_single_unpack(&vsn, v);
+	if (vsn.exponent == 0 && vsn.significand)
+		vfp_single_normalise_denormal(&vsn);
+
+	vfp_single_unpack(&vsm, m);
+	if (vsm.exponent == 0 && vsm.significand)
+		vfp_single_normalise_denormal(&vsm);
+
+	exceptions = vfp_single_multiply(&vsp, &vsn, &vsm, fpscr);
+	if (negate & NEG_MULTIPLY)
+		vsp.sign = vfp_sign_negate(vsp.sign);
+
+	v = vfp_get_float(sd);
+	pr_debug("VFP: s%u = %08x\n", sd, v);
+	vfp_single_unpack(&vsn, v);
+	if (negate & NEG_SUBTRACT)
+		vsn.sign = vfp_sign_negate(vsn.sign);
+
+	exceptions |= vfp_single_add(&vsd, &vsn, &vsp, fpscr);
+
+	return vfp_single_normaliseround(sd, &vsd, fpscr, exceptions, func);
+}
+
+/*
+ * Standard operations
+ */
+
+/*
+ * sd = sd + (sn * sm)
+ */
+static u32 vfp_single_fmac(int sd, int sn, s32 m, u32 fpscr)
+{
+	return vfp_single_multiply_accumulate(sd, sn, m, fpscr, 0, "fmac");
+}
+
+/*
+ * sd = sd - (sn * sm)
+ */
+static u32 vfp_single_fnmac(int sd, int sn, s32 m, u32 fpscr)
+{
+	return vfp_single_multiply_accumulate(sd, sn, m, fpscr, NEG_MULTIPLY, "fnmac");
+}
+
+/*
+ * sd = -sd + (sn * sm)
+ */
+static u32 vfp_single_fmsc(int sd, int sn, s32 m, u32 fpscr)
+{
+	return vfp_single_multiply_accumulate(sd, sn, m, fpscr, NEG_SUBTRACT, "fmsc");
+}
+
+/*
+ * sd = -sd - (sn * sm)
+ */
+static u32 vfp_single_fnmsc(int sd, int sn, s32 m, u32 fpscr)
+{
+	return vfp_single_multiply_accumulate(sd, sn, m, fpscr, NEG_SUBTRACT | NEG_MULTIPLY, "fnmsc");
+}
+
+/*
+ * sd = sn * sm
+ */
+static u32 vfp_single_fmul(int sd, int sn, s32 m, u32 fpscr)
+{
+	struct vfp_single vsd, vsn, vsm;
+	u32 exceptions;
+	s32 n = vfp_get_float(sn);
+
+	pr_debug("VFP: s%u = %08x\n", sn, n);
+
+	vfp_single_unpack(&vsn, n);
+	if (vsn.exponent == 0 && vsn.significand)
+		vfp_single_normalise_denormal(&vsn);
+
+	vfp_single_unpack(&vsm, m);
+	if (vsm.exponent == 0 && vsm.significand)
+		vfp_single_normalise_denormal(&vsm);
+
+	exceptions = vfp_single_multiply(&vsd, &vsn, &vsm, fpscr);
+	return vfp_single_normaliseround(sd, &vsd, fpscr, exceptions, "fmul");
+}
+
+/*
+ * sd = -(sn * sm)
+ */
+static u32 vfp_single_fnmul(int sd, int sn, s32 m, u32 fpscr)
+{
+	struct vfp_single vsd, vsn, vsm;
+	u32 exceptions;
+	s32 n = vfp_get_float(sn);
+
+	pr_debug("VFP: s%u = %08x\n", sn, n);
+
+	vfp_single_unpack(&vsn, n);
+	if (vsn.exponent == 0 && vsn.significand)
+		vfp_single_normalise_denormal(&vsn);
+
+	vfp_single_unpack(&vsm, m);
+	if (vsm.exponent == 0 && vsm.significand)
+		vfp_single_normalise_denormal(&vsm);
+
+	exceptions = vfp_single_multiply(&vsd, &vsn, &vsm, fpscr);
+	vsd.sign = vfp_sign_negate(vsd.sign);
+	return vfp_single_normaliseround(sd, &vsd, fpscr, exceptions, "fnmul");
+}
+
+/*
+ * sd = sn + sm
+ */
+static u32 vfp_single_fadd(int sd, int sn, s32 m, u32 fpscr)
+{
+	struct vfp_single vsd, vsn, vsm;
+	u32 exceptions;
+	s32 n = vfp_get_float(sn);
+
+	pr_debug("VFP: s%u = %08x\n", sn, n);
+
+	/*
+	 * Unpack and normalise denormals.
+	 */
+	vfp_single_unpack(&vsn, n);
+	if (vsn.exponent == 0 && vsn.significand)
+		vfp_single_normalise_denormal(&vsn);
+
+	vfp_single_unpack(&vsm, m);
+	if (vsm.exponent == 0 && vsm.significand)
+		vfp_single_normalise_denormal(&vsm);
+
+	exceptions = vfp_single_add(&vsd, &vsn, &vsm, fpscr);
+
+	return vfp_single_normaliseround(sd, &vsd, fpscr, exceptions, "fadd");
+}
+
+/*
+ * sd = sn - sm
+ */
+static u32 vfp_single_fsub(int sd, int sn, s32 m, u32 fpscr)
+{
+	/*
+	 * Subtraction is addition with one sign inverted.
+	 */
+	return vfp_single_fadd(sd, sn, vfp_single_packed_negate(m), fpscr);
+}
+
+/*
+ * sd = sn / sm
+ */
+static u32 vfp_single_fdiv(int sd, int sn, s32 m, u32 fpscr)
+{
+	struct vfp_single vsd, vsn, vsm;
+	u32 exceptions = 0;
+	s32 n = vfp_get_float(sn);
+	int tm, tn;
+
+	pr_debug("VFP: s%u = %08x\n", sn, n);
+
+	vfp_single_unpack(&vsn, n);
+	vfp_single_unpack(&vsm, m);
+
+	vsd.sign = vsn.sign ^ vsm.sign;
+
+	tn = vfp_single_type(&vsn);
+	tm = vfp_single_type(&vsm);
+
+	/*
+	 * Is n a NAN?
+	 */
+	if (tn & VFP_NAN)
+		goto vsn_nan;
+
+	/*
+	 * Is m a NAN?
+	 */
+	if (tm & VFP_NAN)
+		goto vsm_nan;
+
+	/*
+	 * If n and m are infinity, the result is invalid
+	 * If n and m are zero, the result is invalid
+	 */
+	if (tm & tn & (VFP_INFINITY|VFP_ZERO))
+		goto invalid;
+
+	/*
+	 * If n is infinity, the result is infinity
+	 */
+	if (tn & VFP_INFINITY)
+		goto infinity;
+
+	/*
+	 * If m is zero, raise div0 exception
+	 */
+	if (tm & VFP_ZERO)
+		goto divzero;
+
+	/*
+	 * If m is infinity, or n is zero, the result is zero
+	 */
+	if (tm & VFP_INFINITY || tn & VFP_ZERO)
+		goto zero;
+
+	if (tn & VFP_DENORMAL)
+		vfp_single_normalise_denormal(&vsn);
+	if (tm & VFP_DENORMAL)
+		vfp_single_normalise_denormal(&vsm);
+
+	/*
+	 * Ok, we have two numbers, we can perform division.
+	 */
+	vsd.exponent = vsn.exponent - vsm.exponent + 127 - 1;
+	vsm.significand <<= 1;
+	if (vsm.significand <= (2 * vsn.significand)) {
+		vsn.significand >>= 1;
+		vsd.exponent++;
+	}
+	vsd.significand = ((u64)vsn.significand << 32) / vsm.significand;
+	if ((vsd.significand & 0x3f) == 0)
+		vsd.significand |= ((u64)vsm.significand * vsd.significand != (u64)vsn.significand << 32);
+
+	return vfp_single_normaliseround(sd, &vsd, fpscr, 0, "fdiv");
+
+ vsn_nan:
+	exceptions = vfp_propagate_nan(&vsd, &vsn, &vsm, fpscr);
+ pack:
+	vfp_put_float(sd, vfp_single_pack(&vsd));
+	return exceptions;
+
+ vsm_nan:
+	exceptions = vfp_propagate_nan(&vsd, &vsm, &vsn, fpscr);
+	goto pack;
+
+ zero:
+	vsd.exponent = 0;
+	vsd.significand = 0;
+	goto pack;
+
+ divzero:
+	exceptions = FPSCR_DZC;
+ infinity:
+	vsd.exponent = 255;
+	vsd.significand = 0;
+	goto pack;
+
+ invalid:
+	vfp_put_float(sd, vfp_single_pack(&vfp_single_default_qnan));
+	return FPSCR_IOC;
+}
+
+static u32 (* const fop_fns[16])(int sd, int sn, s32 m, u32 fpscr) = {
+	[FOP_TO_IDX(FOP_FMAC)]	= vfp_single_fmac,
+	[FOP_TO_IDX(FOP_FNMAC)]	= vfp_single_fnmac,
+	[FOP_TO_IDX(FOP_FMSC)]	= vfp_single_fmsc,
+	[FOP_TO_IDX(FOP_FNMSC)]	= vfp_single_fnmsc,
+	[FOP_TO_IDX(FOP_FMUL)]	= vfp_single_fmul,
+	[FOP_TO_IDX(FOP_FNMUL)]	= vfp_single_fnmul,
+	[FOP_TO_IDX(FOP_FADD)]	= vfp_single_fadd,
+	[FOP_TO_IDX(FOP_FSUB)]	= vfp_single_fsub,
+	[FOP_TO_IDX(FOP_FDIV)]	= vfp_single_fdiv,
+};
+
+#define FREG_BANK(x)	((x) & 0x18)
+#define FREG_IDX(x)	((x) & 7)
+
+u32 vfp_single_cpdo(u32 inst, u32 fpscr)
+{
+	u32 op = inst & FOP_MASK;
+	u32 exceptions = 0;
+	unsigned int sd = vfp_get_sd(inst);
+	unsigned int sn = vfp_get_sn(inst);
+	unsigned int sm = vfp_get_sm(inst);
+	unsigned int vecitr, veclen, vecstride;
+	u32 (*fop)(int, int, s32, u32);
+
+	veclen = fpscr & FPSCR_LENGTH_MASK;
+	vecstride = 1 + ((fpscr & FPSCR_STRIDE_MASK) == FPSCR_STRIDE_MASK);
+
+	/*
+	 * If destination bank is zero, vector length is always '1'.
+	 * ARM DDI0100F C5.1.3, C5.3.2.
+	 */
+	if (FREG_BANK(sd) == 0)
+		veclen = 0;
+
+	pr_debug("VFP: vecstride=%u veclen=%u\n", vecstride,
+		 (veclen >> FPSCR_LENGTH_BIT) + 1);
+
+	fop = (op == FOP_EXT) ? fop_extfns[sn] : fop_fns[FOP_TO_IDX(op)];
+	if (!fop)
+		goto invalid;
+
+	for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) {
+		s32 m = vfp_get_float(sm);
+		u32 except;
+
+		if (op == FOP_EXT)
+			pr_debug("VFP: itr%d (s%u) = op[%u] (s%u=%08x)\n",
+				 vecitr >> FPSCR_LENGTH_BIT, sd, sn, sm, m);
+		else
+			pr_debug("VFP: itr%d (s%u) = (s%u) op[%u] (s%u=%08x)\n",
+				 vecitr >> FPSCR_LENGTH_BIT, sd, sn,
+				 FOP_TO_IDX(op), sm, m);
+
+		except = fop(sd, sn, m, fpscr);
+		pr_debug("VFP: itr%d: exceptions=%08x\n",
+			 vecitr >> FPSCR_LENGTH_BIT, except);
+
+		exceptions |= except;
+
+		/*
+		 * This ensures that comparisons only operate on scalars;
+		 * comparisons always return with one FPSCR status bit set.
+		 */
+		if (except & (FPSCR_N|FPSCR_Z|FPSCR_C|FPSCR_V))
+			break;
+
+		/*
+		 * CHECK: It appears to be undefined whether we stop when
+		 * we encounter an exception.  We continue.
+		 */
+
+		sd = FREG_BANK(sd) + ((FREG_IDX(sd) + vecstride) & 7);
+		sn = FREG_BANK(sn) + ((FREG_IDX(sn) + vecstride) & 7);
+		if (FREG_BANK(sm) != 0)
+			sm = FREG_BANK(sm) + ((FREG_IDX(sm) + vecstride) & 7);
+	}
+	return exceptions;
+
+ invalid:
+	return (u32)-1;
+}

drivers/input/serio/ambakmi.c

@@ -18,17 +18,19 @@
 #include <linux/device.h>
 #include <linux/delay.h>
 #include <linux/slab.h>
+#include <linux/err.h>
 
 #include <asm/io.h>
 #include <asm/irq.h>
 #include <asm/hardware/amba.h>
 #include <asm/hardware/amba_kmi.h>
+#include <asm/hardware/clock.h>
 
 #define KMI_BASE	(kmi->base)
 
 struct amba_kmi_port {
 	struct serio		io;
-	struct amba_kmi_port	*next;
+	struct clk		*clk;
 	unsigned char		*base;
 	unsigned int		irq;
 	unsigned int		divisor;
@@ -67,21 +69,38 @@ static int amba_kmi_write(struct serio *io, unsigned char val)
 static int amba_kmi_open(struct serio *io)
 {
 	struct amba_kmi_port *kmi = io->driver;
+	unsigned int divisor;
 	int ret;
 
-	writeb(kmi->divisor, KMICLKDIV);
+	ret = clk_use(kmi->clk);
+	if (ret)
+		goto out;
+
+	ret = clk_enable(kmi->clk);
+	if (ret)
+		goto clk_unuse;
+
+	divisor = clk_get_rate(kmi->clk) / 8000000 - 1;
+	writeb(divisor, KMICLKDIV);
 	writeb(KMICR_EN, KMICR);
 
 	ret = request_irq(kmi->irq, amba_kmi_int, 0, "kmi-pl050", kmi);
 	if (ret) {
 		printk(KERN_ERR "kmi: failed to claim IRQ%d\n", kmi->irq);
 		writeb(0, KMICR);
-		return ret;
+		goto clk_disable;
 	}
 
 	writeb(KMICR_EN | KMICR_RXINTREN, KMICR);
 
 	return 0;
+
+ clk_disable:
+	clk_disable(kmi->clk);
+ clk_unuse:
+	clk_unuse(kmi->clk);
+ out:
+	return ret;
 }
 
 static void amba_kmi_close(struct serio *io)
@@ -91,6 +110,8 @@ static void amba_kmi_close(struct serio *io)
 	writeb(0, KMICR);
 
 	free_irq(kmi->irq, kmi);
+	clk_disable(kmi->clk);
+	clk_unuse(kmi->clk);
 }
 
 static int amba_kmi_probe(struct amba_device *dev, void *id)
@@ -124,14 +145,20 @@ static int amba_kmi_probe(struct amba_device *dev, void *id)
 		goto out;
 	}
 
-	kmi->irq	= dev->irq[0];
-	kmi->divisor	= 24 / 8 - 1;
+	kmi->clk = clk_get(&dev->dev, "KMIREFCLK");
+	if (IS_ERR(kmi->clk)) {
+		ret = PTR_ERR(kmi->clk);
+		goto unmap;
+	}
 
+	kmi->irq = dev->irq[0];
 	amba_set_drvdata(dev, kmi);
 
 	serio_register_port(&kmi->io);
 	return 0;
 
+ unmap:
+	iounmap(kmi->base);
  out:
 	kfree(kmi);
 	amba_release_regions(dev);
@@ -145,6 +172,7 @@ static int amba_kmi_remove(struct amba_device *dev)
 	amba_set_drvdata(dev, NULL);
 
 	serio_unregister_port(&kmi->io);
+	clk_put(kmi->clk);
 	iounmap(kmi->base);
 	kfree(kmi);
 	amba_release_regions(dev);

drivers/serial/amba-pl011.c

@@ -44,6 +44,7 @@
 #include <asm/io.h>
 #include <asm/irq.h>
 #include <asm/hardware/amba.h>
+#include <asm/hardware/clock.h>
 
 #if defined(CONFIG_SERIAL_AMBA_PL011_CONSOLE) && defined(CONFIG_MAGIC_SYSRQ)
 #define SUPPORT_SYSRQ
@@ -68,6 +69,7 @@
  */
 struct uart_amba_port {
 	struct uart_port	port;
+	struct clk		*clk;
 	unsigned int		im;	/* interrupt mask */
 	unsigned int		old_status;
 };
@@ -351,12 +353,21 @@ static int pl011_startup(struct uart_port *port)
 	unsigned int cr;
 	int retval;
 
+	/*
+	 * Try to enable the clock producer.
+	 */
+	retval = clk_enable(uap->clk);
+	if (retval)
+		goto out;
+
+	uap->port.uartclk = clk_get_rate(uap->clk);
+
 	/*
 	 * Allocate the IRQ
 	 */
 	retval = request_irq(uap->port.irq, pl011_int, 0, "uart-pl011", uap);
 	if (retval)
-		goto out;
+		goto clk_dis;
 
 	writew(UART011_IFLS_RX4_8|UART011_IFLS_TX4_8,
 	       uap->port.membase + UART011_IFLS);
@@ -391,6 +402,8 @@ static int pl011_startup(struct uart_port *port)
 
 	return 0;
 
+ clk_dis:
+	clk_disable(uap->clk);
  out:
 	return retval;
 }
@@ -425,6 +438,11 @@ static void pl011_shutdown(struct uart_port *port)
 	val = readw(uap->port.membase + UART011_LCRH);
 	val &= ~(UART01x_LCRH_BRK | UART01x_LCRH_FEN);
 	writew(val, uap->port.membase + UART011_LCRH);
+
+	/*
+	 * Shut down the clock producer
+	 */
+	clk_disable(uap->clk);
 }
 
 static void
@@ -594,38 +612,40 @@ static struct uart_amba_port *amba_ports[UART_NR];
 #ifdef CONFIG_SERIAL_AMBA_PL011_CONSOLE
 
 static inline void
-pl011_console_write_char(struct uart_port *port, char ch)
+pl011_console_write_char(struct uart_amba_port *uap, char ch)
 {
 	unsigned int status;
 
 	do {
-		status = readw(port->membase + UART01x_FR);
+		status = readw(uap->port.membase + UART01x_FR);
 	} while (status & UART01x_FR_TXFF);
-	writew(ch, port->membase + UART01x_DR);
+	writew(ch, uap->port.membase + UART01x_DR);
 }
 
 static void
 pl011_console_write(struct console *co, const char *s, unsigned int count)
 {
-	struct uart_port *port = &amba_ports[co->index]->port;
+	struct uart_amba_port *uap = amba_ports[co->index];
 	unsigned int status, old_cr, new_cr;
 	int i;
 
+	clk_enable(uap->clk);
+
 	/*
 	 *	First save the CR then disable the interrupts
 	 */
-	old_cr = readw(port->membase + UART011_CR);
+	old_cr = readw(uap->port.membase + UART011_CR);
 	new_cr = old_cr & ~UART011_CR_CTSEN;
 	new_cr |= UART01x_CR_UARTEN | UART011_CR_TXE;
-	writew(new_cr, port->membase + UART011_CR);
+	writew(new_cr, uap->port.membase + UART011_CR);
 
 	/*
 	 *	Now, do each character
 	 */
 	for (i = 0; i < count; i++) {
-		pl011_console_write_char(port, s[i]);
+		pl011_console_write_char(uap, s[i]);
 		if (s[i] == '\n')
-			pl011_console_write_char(port, '\r');
+			pl011_console_write_char(uap, '\r');
 	}
 
 	/*
@@ -633,19 +653,21 @@ pl011_console_write(struct console *co, const char *s, unsigned int count)
 	 *	and restore the TCR
 	 */
 	do {
-		status = readw(port->membase + UART01x_FR);
+		status = readw(uap->port.membase + UART01x_FR);
 	} while (status & UART01x_FR_BUSY);
-	writew(old_cr, port->membase + UART011_CR);
+	writew(old_cr, uap->port.membase + UART011_CR);
+
+	clk_disable(uap->clk);
 }
 
 static void __init
-pl011_console_get_options(struct uart_port *port, int *baud,
+pl011_console_get_options(struct uart_amba_port *uap, int *baud,
 			     int *parity, int *bits)
 {
-	if (readw(port->membase + UART011_CR) & UART01x_CR_UARTEN) {
+	if (readw(uap->port.membase + UART011_CR) & UART01x_CR_UARTEN) {
 		unsigned int lcr_h, ibrd, fbrd;
 
-		lcr_h = readw(port->membase + UART011_LCRH);
+		lcr_h = readw(uap->port.membase + UART011_LCRH);
 
 		*parity = 'n';
 		if (lcr_h & UART01x_LCRH_PEN) {
@@ -660,10 +682,10 @@ pl011_console_get_options(struct uart_port *port, int *baud,
 		else
 			*bits = 8;
 
-		ibrd = readw(port->membase + UART011_IBRD);
-		fbrd = readw(port->membase + UART011_FBRD);
+		ibrd = readw(uap->port.membase + UART011_IBRD);
+		fbrd = readw(uap->port.membase + UART011_FBRD);
 
-		*baud = port->uartclk * 4 / (64 * ibrd + fbrd);
+		*baud = uap->port.uartclk * 4 / (64 * ibrd + fbrd);
 	}
 }
 
@@ -685,10 +707,12 @@ static int __init pl011_console_setup(struct console *co, char *options)
 		co->index = 0;
 	uap = amba_ports[co->index];
 
+	uap->port.uartclk = clk_get_rate(uap->clk);
+
 	if (options)
 		uart_parse_options(options, &baud, &parity, &bits, &flow);
 	else
-		pl011_console_get_options(&uap->port, &baud, &parity, &bits);
+		pl011_console_get_options(uap, &baud, &parity, &bits);
 
 	return uart_set_options(&uap->port, co, baud, parity, bits, flow);
 }
@@ -747,16 +771,21 @@ static int pl011_probe(struct amba_device *dev, void *id)
 	}
 
 	memset(uap, 0, sizeof(struct uart_amba_port));
+	uap->clk = clk_get(&dev->dev, "UARTCLK");
+	if (IS_ERR(uap->clk)) {
+		ret = PTR_ERR(uap->clk);
+		goto unmap;
+	}
+
+	ret = clk_use(uap->clk);
+	if (ret)
+		goto putclk;
+
 	uap->port.dev = &dev->dev;
 	uap->port.mapbase = dev->res.start;
 	uap->port.membase = base;
 	uap->port.iotype = UPIO_MEM;
 	uap->port.irq = dev->irq[0];
-#if 0 /* FIXME */
-	uap->port.uartclk = 14745600;
-#else
-	uap->port.uartclk = 24000000;
-#endif
 	uap->port.fifosize = 16;
 	uap->port.ops = &amba_pl011_pops;
 	uap->port.flags = UPF_BOOT_AUTOCONF;
@@ -769,6 +798,10 @@ static int pl011_probe(struct amba_device *dev, void *id)
 	if (ret) {
 		amba_set_drvdata(dev, NULL);
 		amba_ports[i] = NULL;
+		clk_unuse(uap->clk);
+ putclk:
+		clk_put(uap->clk);
+ unmap:
 		iounmap(base);
  free:
 		kfree(uap);
@@ -791,6 +824,8 @@ static int pl011_remove(struct amba_device *dev)
 			amba_ports[i] = NULL;
 
 	iounmap(uap->port.membase);
+	clk_unuse(uap->clk);
+	clk_put(uap->clk);
 	kfree(uap);
 	return 0;
 }

drivers/video/pxafb.c

@@ -42,7 +42,6 @@
 #include <asm/hardware.h>
 #include <asm/io.h>
 #include <asm/irq.h>
-#include <asm/mach-types.h>
 #include <asm/uaccess.h>
 #include <asm/arch/bitfield.h>
 #include <asm/arch/pxafb.h>
@@ -109,10 +108,13 @@ pxafb_setpalettereg(u_int regno, u_int red, u_int green, u_int blue,
 	u_int val, ret = 1;
 
 	if (regno < fbi->palette_size) {
+		if (fbi->fb.var.grayscale) {
+			val = ((blue >> 8) & 0x00ff);
+		} else {
 			val  = ((red   >>  0) & 0xf800);
 			val |= ((green >>  5) & 0x07e0);
 			val |= ((blue  >> 11) & 0x001f);
-
+		}
 		fbi->palette_cpu[regno] = val;
 		ret = 0;
 	}
@@ -150,7 +152,7 @@ pxafb_setcolreg(u_int regno, u_int red, u_int green, u_int blue,
 	switch (fbi->fb.fix.visual) {
 	case FB_VISUAL_TRUECOLOR:
 		/*
-		 * 12 or 16-bit True Colour.  We encode the RGB value
+		 * 16-bit True Colour.  We encode the RGB value
 		 * according to the RGB bitfield information.
 		 */
 		if (regno < 16) {
@@ -242,7 +244,7 @@ static int pxafb_check_var(struct fb_var_screeninfo *var, struct fb_info *info)
 	 * The pixel packing format is described on page 7-11 of the
 	 * PXA2XX Developer's Manual.
          */
-	if ( var->bits_per_pixel == 16 ) {
+	if (var->bits_per_pixel == 16) {
 		var->red.offset   = 11; var->red.length   = 5;
 		var->green.offset = 5;  var->green.length = 6;
 		var->blue.offset  = 0;  var->blue.length  = 5;
@@ -297,7 +299,10 @@ static int pxafb_set_par(struct fb_info *info)
 
 	fbi->fb.fix.line_length = var->xres_virtual *
 				  var->bits_per_pixel / 8;
-	fbi->palette_size = var->bits_per_pixel == 8 ? 256 : 16;
+	if (var->bits_per_pixel == 16)
+		fbi->palette_size = 0;
+	else
+		fbi->palette_size = var->bits_per_pixel == 1 ? 4 : 1 << var->bits_per_pixel;
 
 	palette_mem_size = fbi->palette_size * sizeof(u16);
 
@@ -311,6 +316,11 @@ static int pxafb_set_par(struct fb_info *info)
 	 */
 	pxafb_set_truecolor(fbi->fb.fix.visual == FB_VISUAL_TRUECOLOR);
 
+	if (fbi->fb.var.bits_per_pixel == 16)
+		fb_dealloc_cmap(&fbi->fb.cmap);
+	else
+		fb_alloc_cmap(&fbi->fb.cmap, 1<<fbi->fb.var.bits_per_pixel, 0);
+
 	pxafb_activate_var(var, fbi);
 
 	return 0;
@@ -349,7 +359,7 @@ static int pxafb_set_par(struct fb_info *info)
 /*
  * pxafb_blank():
  *	Blank the display by setting all palette values to zero.  Note, the
- * 	12 and 16 bpp modes don't really use the palette, so this will not
+ * 	16 bpp mode does not really use the palette, so this will not
  *      blank the display in all modes.
  */
 static int pxafb_blank(int blank, struct fb_info *info)
@@ -514,7 +524,7 @@ static int pxafb_activate_var(struct fb_var_screeninfo *var, struct pxafb_info *
 	 * the YRES parameter.
 	 */
 	lines_per_panel = var->yres;
-	if (fbi->lccr0 & LCCR0_SDS)
+	if ((fbi->lccr0 & LCCR0_SDS) == LCCR0_Dual)
 		lines_per_panel /= 2;
 
 	new_regs.lccr2 =
@@ -566,20 +576,16 @@ static int pxafb_activate_var(struct fb_var_screeninfo *var, struct pxafb_info *
 	fbi->dmadesc_palette_cpu->fidr  = 0;
 	fbi->dmadesc_palette_cpu->ldcmd = (fbi->palette_size * 2) | LDCMD_PAL;
 
-	if( var->bits_per_pixel < 12)
-	{
-		/* assume any mode with <12 bpp is palette driven */
-		fbi->dmadesc_palette_cpu->fdadr = fbi->dmadesc_fbhigh_dma;
-		fbi->dmadesc_fbhigh_cpu->fdadr = fbi->dmadesc_palette_dma;
-		fbi->fdadr0 = fbi->dmadesc_palette_dma; /* flips back and forth between pal and fbhigh */
-	}
-	else
-	{
+	if (var->bits_per_pixel == 16) {
 		/* palette shouldn't be loaded in true-color mode */
 		fbi->dmadesc_fbhigh_cpu->fdadr = fbi->dmadesc_fbhigh_dma;
 		fbi->fdadr0 = fbi->dmadesc_fbhigh_dma; /* no pal just fbhigh */
 		/* init it to something, even though we won't be using it */
 		fbi->dmadesc_palette_cpu->fdadr = fbi->dmadesc_palette_dma;
+	} else {
+		fbi->dmadesc_palette_cpu->fdadr = fbi->dmadesc_fbhigh_dma;
+		fbi->dmadesc_fbhigh_cpu->fdadr = fbi->dmadesc_palette_dma;
+		fbi->fdadr0 = fbi->dmadesc_palette_dma; /* flips back and forth between pal and fbhigh */
 	}
 
 #if 0
@@ -696,7 +702,7 @@ static void pxafb_setup_gpio(struct pxafb_info *fbi)
 	}
 
 	else {
-	        printk( KERN_ERR "pxafb_setup_gpio: unable to determine bits per pixel\n");
+	        printk(KERN_ERR "pxafb_setup_gpio: unable to determine bits per pixel\n");
         }
 }
 
@@ -1123,11 +1129,11 @@ static int __init pxafb_parse_options(struct device *dev, char *options)
 				res_specified = 1;
 			}
 		done:
-			if ( res_specified ) {
-				dev_info(dev, "overriding resolution: %dx%x\n", xres, yres);
+			if (res_specified) {
+				dev_info(dev, "overriding resolution: %dx%d\n", xres, yres);
 				inf->xres = xres; inf->yres = yres;
 			}
-			if ( bpp_specified )
+			if (bpp_specified)
 				switch (bpp) {
 				case 1:
 				case 2:
@@ -1142,7 +1148,7 @@ static int __init pxafb_parse_options(struct device *dev, char *options)
 				}
                 } else if (!strncmp(this_opt, "pixclock:", 9)) {
                         inf->pixclock = simple_strtoul(this_opt+9, NULL, 0);
-			dev_info(dev, "override pixclock: %u\n", inf->pixclock);
+			dev_info(dev, "override pixclock: %ld\n", inf->pixclock);
                 } else if (!strncmp(this_opt, "left:", 5)) {
                         inf->left_margin = simple_strtoul(this_opt+5, NULL, 0);
 			dev_info(dev, "override left: %u\n", inf->left_margin);
@@ -1162,7 +1168,7 @@ static int __init pxafb_parse_options(struct device *dev, char *options)
                         inf->vsync_len = simple_strtoul(this_opt+9, NULL, 0);
 			dev_info(dev, "override vsynclen: %u\n", inf->vsync_len);
                 } else if (!strncmp(this_opt, "hsync:", 6)) {
-                        if ( simple_strtoul(this_opt+6, NULL, 0) == 0 ) {
+                        if (simple_strtoul(this_opt+6, NULL, 0) == 0) {
 				dev_info(dev, "override hsync: Active Low\n");
 				inf->sync &= ~FB_SYNC_HOR_HIGH_ACT;
 			} else {
@@ -1170,7 +1176,7 @@ static int __init pxafb_parse_options(struct device *dev, char *options)
 				inf->sync |= FB_SYNC_HOR_HIGH_ACT;
 			}
                 } else if (!strncmp(this_opt, "vsync:", 6)) {
-                        if ( simple_strtoul(this_opt+6, NULL, 0) == 0 ) {
+                        if (simple_strtoul(this_opt+6, NULL, 0) == 0) {
 				dev_info(dev, "override vsync: Active Low\n");
 				inf->sync &= ~FB_SYNC_VERT_HIGH_ACT;
 			} else {
@@ -1178,7 +1184,7 @@ static int __init pxafb_parse_options(struct device *dev, char *options)
 				inf->sync |= FB_SYNC_VERT_HIGH_ACT;
 			}
                 } else if (!strncmp(this_opt, "dpc:", 4)) {
-                        if ( simple_strtoul(this_opt+4, NULL, 0) == 0 ) {
+                        if (simple_strtoul(this_opt+4, NULL, 0) == 0) {
 				dev_info(dev, "override double pixel clock: false\n");
 				inf->lccr3 &= ~LCCR3_DPC;
 			} else {
@@ -1186,20 +1192,20 @@ static int __init pxafb_parse_options(struct device *dev, char *options)
 				inf->lccr3 |= LCCR3_DPC;
 			}
                 } else if (!strncmp(this_opt, "outputen:", 9)) {
-                        if ( simple_strtoul(this_opt+9, NULL, 0) == 0 ) {
+                        if (simple_strtoul(this_opt+9, NULL, 0) == 0) {
 				dev_info(dev, "override output enable: active low\n");
-				inf->lccr3 = ( inf->lccr3 & ~LCCR3_OEP ) | LCCR3_OutEnL;
+				inf->lccr3 = (inf->lccr3 & ~LCCR3_OEP) | LCCR3_OutEnL;
 			} else {
 				dev_info(dev, "override output enable: active high\n");
-				inf->lccr3 = ( inf->lccr3 & ~LCCR3_OEP ) | LCCR3_OutEnH;
+				inf->lccr3 = (inf->lccr3 & ~LCCR3_OEP) | LCCR3_OutEnH;
 			}
                 } else if (!strncmp(this_opt, "pixclockpol:", 12)) {
-                        if ( simple_strtoul(this_opt+12, NULL, 0) == 0 ) {
+                        if (simple_strtoul(this_opt+12, NULL, 0) == 0) {
 				dev_info(dev, "override pixel clock polarity: falling edge\n");
-				inf->lccr3 = ( inf->lccr3 & ~LCCR3_PCP ) | LCCR3_PixFlEdg;
+				inf->lccr3 = (inf->lccr3 & ~LCCR3_PCP) | LCCR3_PixFlEdg;
 			} else {
 				dev_info(dev, "override pixel clock polarity: rising edge\n");
-				inf->lccr3 = ( inf->lccr3 & ~LCCR3_PCP ) | LCCR3_PixRsEdg;
+				inf->lccr3 = (inf->lccr3 & ~LCCR3_PCP) | LCCR3_PixRsEdg;
 			}
                 } else if (!strncmp(this_opt, "color", 5)) {
 			inf->lccr0 = (inf->lccr0 & ~LCCR0_CMS) | LCCR0_Color;
@@ -1244,7 +1250,7 @@ int __init pxafb_probe(struct device *dev)
 
 #ifdef CONFIG_FB_PXA_PARAMETERS
 	ret = pxafb_parse_options(dev, g_options);
-	if ( ret < 0 )
+	if (ret < 0)
 		goto failed;
 #endif
 
@@ -1252,23 +1258,23 @@ int __init pxafb_probe(struct device *dev)
         /* Check for various illegal bit-combinations. Currently only
 	 * a warning is given. */
 
-        if ( inf->lccr0 & LCCR0_INVALID_CONFIG_MASK )
+        if (inf->lccr0 & LCCR0_INVALID_CONFIG_MASK)
                 dev_warn(dev, "machine LCCR0 setting contains illegal bits: %08x\n",
                         inf->lccr0 & LCCR0_INVALID_CONFIG_MASK);
-        if ( inf->lccr3 & LCCR3_INVALID_CONFIG_MASK )
+        if (inf->lccr3 & LCCR3_INVALID_CONFIG_MASK)
                 dev_warn(dev, "machine LCCR3 setting contains illegal bits: %08x\n",
                         inf->lccr3 & LCCR3_INVALID_CONFIG_MASK);
-        if ( inf->lccr0 & LCCR0_DPD &&
-             ( ( inf->lccr0 & LCCR0_PAS ) != LCCR0_Pas ||
-               ( inf->lccr0 & LCCR0_SDS ) != LCCR0_Sngl ||
-               ( inf->lccr0 & LCCR0_CMS ) != LCCR0_Mono ) )
+        if (inf->lccr0 & LCCR0_DPD &&
+	    ((inf->lccr0 & LCCR0_PAS) != LCCR0_Pas ||
+	     (inf->lccr0 & LCCR0_SDS) != LCCR0_Sngl ||
+	     (inf->lccr0 & LCCR0_CMS) != LCCR0_Mono))
                 dev_warn(dev, "Double Pixel Data (DPD) mode is only valid in passive mono"
 			 " single panel mode\n");
-        if ( (inf->lccr0 & LCCR0_PAS) == LCCR0_Act &&
-             ( inf->lccr0 & LCCR0_SDS ) == LCCR0_Dual )
+        if ((inf->lccr0 & LCCR0_PAS) == LCCR0_Act &&
+	    (inf->lccr0 & LCCR0_SDS) == LCCR0_Dual)
                 dev_warn(dev, "Dual panel only valid in passive mode\n");
-        if ( (inf->lccr0 & LCCR0_PAS ) == LCCR0_Pas &&
-             (inf->upper_margin || inf->lower_margin) )
+        if ((inf->lccr0 & LCCR0_PAS) == LCCR0_Pas &&
+             (inf->upper_margin || inf->lower_margin))
                 dev_warn(dev, "Upper and lower margins must be 0 in passive mode\n");
 #endif
 

include/asm-arm/arch-integrator/impd1.h

@@ -14,6 +14,5 @@
 
 struct device;
 
-void impd1_set_vco(struct device *dev, int vconr, unsigned long period);
 void impd1_tweak_control(struct device *dev, u32 mask, u32 val);
 

include/asm-arm/arch-lh7a40x/memory.h

@@ -6,14 +6,14 @@
  *  modify it under the terms of the GNU General Public License
  *  version 2 as published by the Free Software Foundation.
  *
+ *
+ *  Refer to <file:Documentation/arm/Sharp-LH/SDRAM> for more information.
+ *
  */
 
 #ifndef __ASM_ARCH_MEMORY_H
 #define __ASM_ARCH_MEMORY_H
 
-
-#define BANKS_PER_NODE 1	/* Define as either 1 or 2 */
-
 /*
  * Physical DRAM offset.
  */
@@ -30,57 +30,30 @@
 #define __bus_to_virt(x)	 __phys_to_virt(x)
 
 #ifdef CONFIG_DISCONTIGMEM
-/*
- * Because of the wide memory address space between physical RAM
- * banks, it's convenient to use Linux's NUMA support to represent our
- * memory map.  Assuming all memory nodes have equal access
- * characteristics, we then have a generic discontiguous memory setup.
- *
- * Of course, all this isn't mandatory for implementations with only
- * one used memory bank.  For those, simply undefine
- * CONFIG_DISCONTIGMEM.  However, keep in mind that a featurefull
- * system will need more than 4MiB of RAM.
- *
- * The contiguous memory blocks are small enough that it pays to
- * aggregate two banks into one node.  Setting BANKS_PER_NODE to 2
- * puts pairs of banks into a node.
- *
- * A typical layout would start like this:
- *
- *  node 0: 0xc0000000
- *          0xc1000000
- *  node 1: 0xc4000000
- *          0xc5000000
- *  node 2: 0xc8000000
- *          0xc9000000
- *
- * The proximity of the pairs of blocks makes it feasible to combine them.
- *
- */
 
 /*
  * Given a kernel address, find the home node of the underlying memory.
  */
 
-#if BANKS_PER_NODE==1
-#define KVADDR_TO_NID(addr) \
+# ifdef CONFIG_LH7A40X_ONE_BANK_PER_NODE
+#  define KVADDR_TO_NID(addr) \
   (  ((((unsigned long) (addr) - PAGE_OFFSET) >> 24) &  1)\
    | ((((unsigned long) (addr) - PAGE_OFFSET) >> 25) & ~1))
-#else  /* 2 banks per node */
-#define KVADDR_TO_NID(addr) \
-       ((unsigned long) (addr) - PAGE_OFFSET) >> 26)
-#endif
+# else  /* 2 banks per node */
+#  define KVADDR_TO_NID(addr) \
+      (((unsigned long) (addr) - PAGE_OFFSET) >> 26)
+# endif
 
 /*
  * Given a page frame number, convert it to a node id.
  */
 
-#if BANKS_PER_NODE==1
-#define PFN_TO_NID(pfn) \
+# ifdef CONFIG_LH7A40X_ONE_BANK_PER_NODE
+#  define PFN_TO_NID(pfn) \
   (((((pfn) - PHYS_PFN_OFFSET) >> (24 - PAGE_SHIFT)) &  1)\
  | ((((pfn) - PHYS_PFN_OFFSET) >> (25 - PAGE_SHIFT)) & ~1))
-#else  /* 2 banks per node */
-#define PFN_TO_NID(addr) \
+# else  /* 2 banks per node */
+#  define PFN_TO_NID(pfn) \
     (((pfn) - PHYS_PFN_OFFSET) >> (26 - PAGE_SHIFT))
 #endif
 
@@ -88,13 +61,13 @@
  * Given a kaddr, ADDR_TO_MAPBASE finds the owning node of the memory
  * and return the mem_map of that node.
  */
-#define ADDR_TO_MAPBASE(kaddr)	NODE_MEM_MAP(KVADDR_TO_NID(kaddr))
+# define ADDR_TO_MAPBASE(kaddr)	NODE_MEM_MAP(KVADDR_TO_NID(kaddr))
 
 /*
  * Given a page frame number, find the owning node of the memory
  * and return the mem_map of that node.
  */
-#define PFN_TO_MAPBASE(pfn)	NODE_MEM_MAP(PFN_TO_NID(pfn))
+# define PFN_TO_MAPBASE(pfn)	NODE_MEM_MAP(PFN_TO_NID(pfn))
 
 /*
  * Given a kaddr, LOCAL_MEM_MAP finds the owning node of the memory
@@ -102,17 +75,17 @@
  * node's mem_map.
  */
 
-#if BANKS_PER_NODE==1
-#define LOCAL_MAP_NR(addr) \
+# ifdef CONFIG_LH7A40X_ONE_BANK_PER_NODE
+#  define LOCAL_MAP_NR(addr) \
        (((unsigned long)(addr) & 0x003fffff) >> PAGE_SHIFT)
-#else  /* 2 banks per node */
-#define LOCAL_MAP_NR(addr) \
+# else  /* 2 banks per node */
+#  define LOCAL_MAP_NR(addr) \
        (((unsigned long)(addr) & 0x01ffffff) >> PAGE_SHIFT)
-#endif
+# endif
 
 #else
 
-#define PFN_TO_NID(addr)	(0)
+# define PFN_TO_NID(addr)	(0)
 
 #endif
 

include/asm-arm/fpstate.h

@@ -13,6 +13,36 @@
 
 #define FP_SIZE 35
 
+#ifndef __ASSEMBLY__
+
+/*
+ * VFP storage area has:
+ *  - FPEXC, FPSCR, FPINST and FPINST2.
+ *  - 16 double precision data registers
+ *  - an implementation-dependant word of state for FLDMX/FSTMX
+ * 
+ *  FPEXC will always be non-zero once the VFP has been used in this process.
+ */
+
+struct vfp_hard_struct {
+	__u64 fpregs[16];
+	__u32 fpmx_state;
+	__u32 fpexc;
+	__u32 fpscr;
+	/*
+	 * VFP implementation specific state
+	 */
+	__u32 fpinst;
+	__u32 fpinst2;
+};
+
+union vfp_state {
+	struct vfp_hard_struct	hard;
+};
+
+extern void vfp_flush_thread(union vfp_state *);
+extern void vfp_release_thread(union vfp_state *);
+
 struct fp_hard_struct {
 	unsigned int save[FP_SIZE];		/* as yet undefined */
 };
@@ -27,3 +57,5 @@ union fp_state {
 };
 
 #endif
+
+#endif

include/asm-arm/hardware/clock.h

+/*
+ *  linux/include/asm-arm/hardware/clock.h
+ *
+ *  Copyright (C) 2004 ARM Limited.
+ *  Written by Deep Blue Solutions Limited.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+#ifndef ASMARM_CLOCK_H
+#define ASMARM_CLOCK_H
+
+struct device;
+
+/*
+ * The base API.
+ */
+
+
+/*
+ * struct clk - an machine class defined object / cookie.
+ */
+struct clk;
+
+/**
+ * clk_get - lookup and obtain a reference to a clock producer.
+ * @dev: device for clock "consumer"
+ * @id: device ID
+ *
+ * Returns a struct clk corresponding to the clock producer, or
+ * valid IS_ERR() condition containing errno.
+ */
+struct clk *clk_get(struct device *dev, const char *id);
+
+/**
+ * clk_enable - inform the system when the clock source should be running.
+ * @clk: clock source
+ *
+ * If the clock can not be enabled/disabled, this should return success.
+ *
+ * Returns success (0) or negative errno.
+ */
+int clk_enable(struct clk *clk);
+
+/**
+ * clk_disable - inform the system when the clock source is no longer required.
+ * @clk: clock source
+ */
+void clk_disable(struct clk *clk);
+
+/**
+ * clk_use - increment the use count
+ * @clk: clock source
+ *
+ * Returns success (0) or negative errno.
+ */
+int clk_use(struct clk *clk);
+
+/**
+ * clk_unuse - decrement the use count
+ * @clk: clock source
+ */
+void clk_unuse(struct clk *clk);
+
+/**
+ * clk_get_rate - obtain the current clock rate for a clock source.
+ *		  This is only valid once the clock source has been enabled.
+ * @clk: clock source
+ */
+unsigned long clk_get_rate(struct clk *clk);
+
+/**
+ * clk_put	- "free" the clock source
+ * @clk: clock source
+ */
+void clk_put(struct clk *clk);
+
+
+/*
+ * The remaining APIs are optional for machine class support.
+ */
+
+
+/**
+ * clk_round_rate - adjust a rate to the exact rate a clock can provide
+ * @clk: clock source
+ * @rate: desired clock rate in kHz
+ *
+ * Returns rounded clock rate, or negative errno.
+ */
+long clk_round_rate(struct clk *clk, unsigned long rate);
+ 
+/**
+ * clk_set_rate - set the clock rate for a clock source
+ * @clk: clock source
+ * @rate: desired clock rate in kHz
+ *
+ * Returns success (0) or negative errno.
+ */
+int clk_set_rate(struct clk *clk, unsigned long rate);
+ 
+/**
+ * clk_set_parent - set the parent clock source for this clock
+ * @clk: clock source
+ * @parent: parent clock source
+ *
+ * Returns success (0) or negative errno.
+ */
+int clk_set_parent(struct clk *clk, struct clk *parent);
+                                                                                
+/**
+ * clk_get_parent - get the parent clock source for this clock
+ * @clk: clock source
+ *
+ * Returns struct clk corresponding to parent clock source, or
+ * valid IS_ERR() condition containing errno.
+ */
+struct clk *clk_get_parent(struct clk *clk);
+
+#endif

include/asm-arm/thread_info.h

@@ -12,12 +12,13 @@
 
 #ifdef __KERNEL__
 
+#include <asm/fpstate.h>
+
 #ifndef __ASSEMBLY__
 
 struct task_struct;
 struct exec_domain;
 
-#include <asm/fpstate.h>
 #include <asm/ptrace.h>
 #include <asm/types.h>
 #include <asm/domain.h>
@@ -53,6 +54,7 @@ struct thread_info {
 	struct cpu_context_save	cpu_context;	/* cpu context */
 	__u8			used_cp[16];	/* thread used copro */
 	union fp_state		fpstate;
+	union vfp_state		vfpstate;
 	struct restart_block	restart_block;
 };
 
@@ -110,6 +112,7 @@ extern void free_thread_info(struct thread_info *);
 #define TI_CPU_SAVE	28
 #define TI_USED_CP	76
 #define TI_FPSTATE	(TI_USED_CP+16)
+#define TI_VFPSTATE	(TI_FPSTATE+FP_SIZE*4)
 
 #endif
 

include/asm-arm/vfp.h

+/*
+ * linux/include/asm-arm/vfp.h
+ *
+ * VFP register definitions.
+ * First, the standard VFP set.
+ */
+
+#define FPSID			cr0
+#define FPSCR			cr1
+#define FPEXC			cr8
+
+/* FPSID bits */
+#define FPSID_IMPLEMENTER_BIT	(24)
+#define FPSID_IMPLEMENTER_MASK	(0xff << FPSID_IMPLEMENTER_BIT)
+#define FPSID_SOFTWARE		(1<<23)
+#define FPSID_FORMAT_BIT	(21)
+#define FPSID_FORMAT_MASK	(0x3  << FPSID_FORMAT_BIT)
+#define FPSID_NODOUBLE		(1<<20)
+#define FPSID_ARCH_BIT		(16)
+#define FPSID_ARCH_MASK		(0xF  << FPSID_ARCH_BIT)
+#define FPSID_PART_BIT		(8)
+#define FPSID_PART_MASK		(0xFF << FPSID_PART_BIT)
+#define FPSID_VARIANT_BIT	(4)
+#define FPSID_VARIANT_MASK	(0xF  << FPSID_VARIANT_BIT)
+#define FPSID_REV_BIT		(0)
+#define FPSID_REV_MASK		(0xF  << FPSID_REV_BIT)
+
+/* FPEXC bits */
+#define FPEXC_EXCEPTION		(1<<31)
+#define FPEXC_ENABLE		(1<<30)
+
+/* FPSCR bits */
+#define FPSCR_DEFAULT_NAN	(1<<25)
+#define FPSCR_FLUSHTOZERO	(1<<24)
+#define FPSCR_ROUND_NEAREST	(0<<22)
+#define FPSCR_ROUND_PLUSINF	(1<<22)
+#define FPSCR_ROUND_MINUSINF	(2<<22)
+#define FPSCR_ROUND_TOZERO	(3<<22)
+#define FPSCR_RMODE_BIT		(22)
+#define FPSCR_RMODE_MASK	(3 << FPSCR_RMODE_BIT)
+#define FPSCR_STRIDE_BIT	(20)
+#define FPSCR_STRIDE_MASK	(3 << FPSCR_STRIDE_BIT)
+#define FPSCR_LENGTH_BIT	(16)
+#define FPSCR_LENGTH_MASK	(7 << FPSCR_LENGTH_BIT)
+#define FPSCR_IOE		(1<<8)
+#define FPSCR_DZE		(1<<9)
+#define FPSCR_OFE		(1<<10)
+#define FPSCR_UFE		(1<<11)
+#define FPSCR_IXE		(1<<12)
+#define FPSCR_IDE		(1<<15)
+#define FPSCR_IOC		(1<<0)
+#define FPSCR_DZC		(1<<1)
+#define FPSCR_OFC		(1<<2)
+#define FPSCR_UFC		(1<<3)
+#define FPSCR_IXC		(1<<4)
+#define FPSCR_IDC		(1<<7)
+
+/*
+ * VFP9-S specific.
+ */
+#define FPINST			cr9
+#define FPINST2			cr10
+
+/* FPEXC bits */
+#define FPEXC_FPV2		(1<<28)
+#define FPEXC_LENGTH_BIT	(8)
+#define FPEXC_LENGTH_MASK	(7 << FPEXC_LENGTH_BIT)
+#define FPEXC_INV		(1 << 7)
+#define FPEXC_UFC		(1 << 3)
+#define FPEXC_OFC		(1 << 2)
+#define FPEXC_IOC		(1 << 0)
+
+/* Bit patterns for decoding the packaged operation descriptors */
+#define VFPOPDESC_LENGTH_BIT	(9)
+#define VFPOPDESC_LENGTH_MASK	(0x07 << VFPOPDESC_LENGTH_BIT)
+#define VFPOPDESC_UNUSED_BIT	(24)
+#define VFPOPDESC_UNUSED_MASK	(0xFF << VFPOPDESC_UNUSED_BIT)
+#define VFPOPDESC_OPDESC_MASK	(~(VFPOPDESC_LENGTH_MASK | VFPOPDESC_UNUSED_MASK))

include/asm-arm/vfpmacros.h

+/*
+ * linux/include/asm-arm/vfpmacros.h
+ *
+ * Assembler-only file containing VFP macros and register definitions.
+ */
+#include "vfp.h"
+
+@ Macros to allow building with old toolkits (with no VFP support)
+	.macro	VFPFMRX, rd, sysreg, cond
+	MRC\cond	p10, 7, \rd, \sysreg, cr0, 0	@ FMRX	\rd, \sysreg
+	.endm
+
+	.macro	VFPFMXR, sysreg, rd, cond
+	MCR\cond	p10, 7, \rd, \sysreg, cr0, 0	@ FMXR	\sysreg, \rd
+	.endm
+
+	@ read all the working registers back into the VFP
+	.macro	VFPFLDMIA, base
+	LDC	p11, cr0, [\base],#33*4		    @ FLDMIAX \base!, {d0-d15}
+	.endm
+
+	@ write all the working registers out of the VFP
+	.macro	VFPFSTMIA, base
+	STC	p11, cr0, [\base],#33*4		    @ FSTMIAX \base!, {d0-d15}
+	.endm