Linux-2.6.12-rc2v2.6.12-rc2

Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.

Let it rip!

8 files changed, 3402 insertions, 0 deletions

diff --git a/arch/arm/vfp/Makefile b/arch/arm/vfp/Makefile
new file mode 100644
index 00000000000..afabac31dd1
--- /dev/null
+++ b/arch/arm/vfp/Makefile
@@ -0,0 +1,12 @@
+#
+# 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
diff --git a/arch/arm/vfp/entry.S b/arch/arm/vfp/entry.S
new file mode 100644
index 00000000000..e73c8deca59
--- /dev/null
+++ b/arch/arm/vfp/entry.S
@@ -0,0 +1,45 @@
+/*
+ *  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/constants.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
diff --git a/arch/arm/vfp/vfp.h b/arch/arm/vfp/vfp.h
new file mode 100644
index 00000000000..55a02bc994a
--- /dev/null
+++ b/arch/arm/vfp/vfp.h
@@ -0,0 +1,344 @@
+/*
+ *  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;
+};
+
+/*
+ * VFP_REG_ZERO is a special register number for vfp_get_double
+ * which returns (double)0.0.  This is useful for the compare with
+ * zero instructions.
+ */
+#define VFP_REG_ZERO	16
+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);
+
+/*
+ * A special flag to tell the normalisation code not to normalise.
+ */
+#define VFP_NAN_FLAG	0x100
diff --git a/arch/arm/vfp/vfpdouble.c b/arch/arm/vfp/vfpdouble.c
new file mode 100644
index 00000000000..fa3053e84db
--- /dev/null
+++ b/arch/arm/vfp/vfpdouble.c
@@ -0,0 +1,1186 @@
+/*
+ *  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 <linux/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 & ~VFP_NAN_FLAG;
+}
+
+/*
+ * 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 : VFP_NAN_FLAG;
+}
+
+/*
+ * 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, VFP_REG_ZERO, fpscr);
+}
+
+static u32 vfp_double_fcmpez(int dd, int unused, int dm, u32 fpscr)
+{
+	return vfp_compare(dd, 1, VFP_REG_ZERO, 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);
+		vf