lib: div64: sync with Linux

Sync with Linux commit ad0376eb1483b ("Merge tag 'edac_for_4.11_2'"). Signed-off-by: Peng Fan <peng.fan@nxp.com> Cc: Tom Rini <trini@konsulko.com>

lib: div64: sync with Linux
Sync with Linux commit ad0376eb1483b ("Merge tag 'edac_for_4.11_2'"). Signed-off-by: Peng Fan <peng.fan@nxp.com> Cc: Tom Rini <trini@konsulko.com>
Peng Fan · Tom Rini
1 parent 6823e6fe66
Showing 3 changed files with 508 additions and 10 deletions Side-by-side Diff
include/div64.h
include/linux/math64.h
lib/div64.c
@@ -4,13 +4,16 @@
  * Copyright (C) 2003 Bernardo Innocenti <bernie@develer.com>
  * Based on former asm-ppc/div64.h and asm-m68knommu/div64.h
  *
+ * Optimization for constant divisors on 32-bit machines:
+ * Copyright (C) 2006-2015 Nicolas Pitre
+ *
  * The semantics of do_div() are:
  *
  * uint32_t do_div(uint64_t *n, uint32_t base)
  * {
- *	uint32_t remainder = *n % base;
- *	*n = *n / base;
- *	return remainder;
+ * 	uint32_t remainder = *n % base;
+ * 	*n = *n / base;
+ * 	return remainder;
  * }
  *
  * NOTE: macro parameter n is evaluated multiple times,
  
  
@@ -18,8 +21,182 @@
  */
  
 #include <linux/types.h>
+#include <linux/compiler.h>
  
+#if BITS_PER_LONG == 64
+
+# define do_div(n,base) ({					\
+	uint32_t __base = (base);				\
+	uint32_t __rem;						\
+	__rem = ((uint64_t)(n)) % __base;			\
+	(n) = ((uint64_t)(n)) / __base;				\
+	__rem;							\
+ })
+
+#elif BITS_PER_LONG == 32
+
+#include <linux/log2.h>
+
+/*
+ * If the divisor happens to be constant, we determine the appropriate
+ * inverse at compile time to turn the division into a few inline
+ * multiplications which ought to be much faster. And yet only if compiling
+ * with a sufficiently recent gcc version to perform proper 64-bit constant
+ * propagation.
+ *
+ * (It is unfortunate that gcc doesn't perform all this internally.)
+ */
+
+#ifndef __div64_const32_is_OK
+#define __div64_const32_is_OK (__GNUC__ >= 4)
+#endif
+
+#define __div64_const32(n, ___b)					\
+({									\
+	/*								\
+	 * Multiplication by reciprocal of b: n / b = n * (p / b) / p	\
+	 *								\
+	 * We rely on the fact that most of this code gets optimized	\
+	 * away at compile time due to constant propagation and only	\
+	 * a few multiplication instructions should remain.		\
+	 * Hence this monstrous macro (static inline doesn't always	\
+	 * do the trick here).						\
+	 */								\
+	uint64_t ___res, ___x, ___t, ___m, ___n = (n);			\
+	uint32_t ___p, ___bias;						\
+									\
+	/* determine MSB of b */					\
+	___p = 1 << ilog2(___b);					\
+									\
+	/* compute m = ((p << 64) + b - 1) / b */			\
+	___m = (~0ULL / ___b) * ___p;					\
+	___m += (((~0ULL % ___b + 1) * ___p) + ___b - 1) / ___b;	\
+									\
+	/* one less than the dividend with highest result */		\
+	___x = ~0ULL / ___b * ___b - 1;					\
+									\
+	/* test our ___m with res = m * x / (p << 64) */		\
+	___res = ((___m & 0xffffffff) * (___x & 0xffffffff)) >> 32;	\
+	___t = ___res += (___m & 0xffffffff) * (___x >> 32);		\
+	___res += (___x & 0xffffffff) * (___m >> 32);			\
+	___t = (___res < ___t) ? (1ULL << 32) : 0;			\
+	___res = (___res >> 32) + ___t;					\
+	___res += (___m >> 32) * (___x >> 32);				\
+	___res /= ___p;							\
+									\
+	/* Now sanitize and optimize what we've got. */			\
+	if (~0ULL % (___b / (___b & -___b)) == 0) {			\
+		/* special case, can be simplified to ... */		\
+		___n /= (___b & -___b);					\
+		___m = ~0ULL / (___b / (___b & -___b));			\
+		___p = 1;						\
+		___bias = 1;						\
+	} else if (___res != ___x / ___b) {				\
+		/*							\
+		 * We can't get away without a bias to compensate	\
+		 * for bit truncation errors.  To avoid it we'd need an	\
+		 * additional bit to represent m which would overflow	\
+		 * a 64-bit variable.					\
+		 *							\
+		 * Instead we do m = p / b and n / b = (n * m + m) / p.	\
+		 */							\
+		___bias = 1;						\
+		/* Compute m = (p << 64) / b */				\
+		___m = (~0ULL / ___b) * ___p;				\
+		___m += ((~0ULL % ___b + 1) * ___p) / ___b;		\
+	} else {							\
+		/*							\
+		 * Reduce m / p, and try to clear bit 31 of m when	\
+		 * possible, otherwise that'll need extra overflow	\
+		 * handling later.					\
+		 */							\
+		uint32_t ___bits = -(___m & -___m);			\
+		___bits |= ___m >> 32;					\
+		___bits = (~___bits) << 1;				\
+		/*							\
+		 * If ___bits == 0 then setting bit 31 is  unavoidable.	\
+		 * Simply apply the maximum possible reduction in that	\
+		 * case. Otherwise the MSB of ___bits indicates the	\
+		 * best reduction we should apply.			\
+		 */							\
+		if (!___bits) {						\
+			___p /= (___m & -___m);				\
+			___m /= (___m & -___m);				\
+		} else {						\
+			___p >>= ilog2(___bits);			\
+			___m >>= ilog2(___bits);			\
+		}							\
+		/* No bias needed. */					\
+		___bias = 0;						\
+	}								\
+									\
+	/*								\
+	 * Now we have a combination of 2 conditions:			\
+	 *								\
+	 * 1) whether or not we need to apply a bias, and		\
+	 *								\
+	 * 2) whether or not there might be an overflow in the cross	\
+	 *    product determined by (___m & ((1 << 63) | (1 << 31))).	\
+	 *								\
+	 * Select the best way to do (m_bias + m * n) / (1 << 64).	\
+	 * From now on there will be actual runtime code generated.	\
+	 */								\
+	___res = __arch_xprod_64(___m, ___n, ___bias);			\
+									\
+	___res /= ___p;							\
+})
+
+#ifndef __arch_xprod_64
+/*
+ * Default C implementation for __arch_xprod_64()
+ *
+ * Prototype: uint64_t __arch_xprod_64(const uint64_t m, uint64_t n, bool bias)
+ * Semantic:  retval = ((bias ? m : 0) + m * n) >> 64
+ *
+ * The product is a 128-bit value, scaled down to 64 bits.
+ * Assuming constant propagation to optimize away unused conditional code.
+ * Architectures may provide their own optimized assembly implementation.
+ */
+static inline uint64_t __arch_xprod_64(const uint64_t m, uint64_t n, bool bias)
+{
+	uint32_t m_lo = m;
+	uint32_t m_hi = m >> 32;
+	uint32_t n_lo = n;
+	uint32_t n_hi = n >> 32;
+	uint64_t res, tmp;
+
+	if (!bias) {
+		res = ((uint64_t)m_lo * n_lo) >> 32;
+	} else if (!(m & ((1ULL << 63) | (1ULL << 31)))) {
+		/* there can't be any overflow here */
+		res = (m + (uint64_t)m_lo * n_lo) >> 32;
+	} else {
+		res = m + (uint64_t)m_lo * n_lo;
+		tmp = (res < m) ? (1ULL << 32) : 0;
+		res = (res >> 32) + tmp;
+	}
+
+	if (!(m & ((1ULL << 63) | (1ULL << 31)))) {
+		/* there can't be any overflow here */
+		res += (uint64_t)m_lo * n_hi;
+		res += (uint64_t)m_hi * n_lo;
+		res >>= 32;
+	} else {
+		tmp = res += (uint64_t)m_lo * n_hi;
+		res += (uint64_t)m_hi * n_lo;
+		tmp = (res < tmp) ? (1ULL << 32) : 0;
+		res = (res >> 32) + tmp;
+	}
+
+	res += (uint64_t)m_hi * n_hi;
+
+	return res;
+}
+#endif
+
+#ifndef __div64_32
 extern uint32_t __div64_32(uint64_t *dividend, uint32_t divisor);
+#endif
  
 /* The unnecessary pointer compare is there
  * to check for type safety (n must be 64bit)
  
  
@@ -28,13 +205,31 @@
 	uint32_t __base = (base);			\
 	uint32_t __rem;					\
 	(void)(((typeof((n)) *)0) == ((uint64_t *)0));	\
-	if (((n) >> 32) == 0) {			\
+	if (__builtin_constant_p(__base) &&		\
+	    is_power_of_2(__base)) {			\
+		__rem = (n) & (__base - 1);		\
+		(n) >>= ilog2(__base);			\
+	} else if (__div64_const32_is_OK &&		\
+		   __builtin_constant_p(__base) &&	\
+		   __base != 0) {			\
+		uint32_t __res_lo, __n_lo = (n);	\
+		(n) = __div64_const32(n, __base);	\
+		/* the remainder can be computed with 32-bit regs */ \
+		__res_lo = (n);				\
+		__rem = __n_lo - __res_lo * __base;	\
+	} else if (likely(((n) >> 32) == 0)) {		\
 		__rem = (uint32_t)(n) % __base;		\
 		(n) = (uint32_t)(n) / __base;		\
-	} else						\
+	} else 						\
 		__rem = __div64_32(&(n), __base);	\
 	__rem;						\
  })
+
+#else /* BITS_PER_LONG == ?? */
+
+# error do_div() does not yet support the C64
+
+#endif /* BITS_PER_LONG */
  
 /* Wrapper for do_div(). Doesn't modify dividend and returns
  * the result, not reminder.
 #ifndef _LINUX_MATH64_H
 #define _LINUX_MATH64_H
  
+#include <div64.h>
+#include <linux/bitops.h>
 #include <linux/types.h>
  
 #if BITS_PER_LONG == 64
  
+#define div64_long(x, y) div64_s64((x), (y))
+#define div64_ul(x, y)   div64_u64((x), (y))
+
 /**
  * div_u64_rem - unsigned 64bit divide with 32bit divisor with remainder
  *
@@ -27,6 +32,15 @@
 }
  
 /**
+ * div64_u64_rem - unsigned 64bit divide with 64bit divisor and remainder
+ */
+static inline u64 div64_u64_rem(u64 dividend, u64 divisor, u64 *remainder)
+{
+	*remainder = dividend % divisor;
+	return dividend / divisor;
+}
+
+/**
  * div64_u64 - unsigned 64bit divide with 64bit divisor
  */
 static inline u64 div64_u64(u64 dividend, u64 divisor)
  
@@ -34,8 +48,19 @@
 	return dividend / divisor;
 }
  
+/**
+ * div64_s64 - signed 64bit divide with 64bit divisor
+ */
+static inline s64 div64_s64(s64 dividend, s64 divisor)
+{
+	return dividend / divisor;
+}
+
 #elif BITS_PER_LONG == 32
  
+#define div64_long(x, y) div_s64((x), (y))
+#define div64_ul(x, y)   div_u64((x), (y))
+
 #ifndef div_u64_rem
 static inline u64 div_u64_rem(u64 dividend, u32 divisor, u32 *remainder)
 {
  
@@ -48,10 +73,18 @@
 extern s64 div_s64_rem(s64 dividend, s32 divisor, s32 *remainder);
 #endif
  
+#ifndef div64_u64_rem
+extern u64 div64_u64_rem(u64 dividend, u64 divisor, u64 *remainder);
+#endif
+
 #ifndef div64_u64
 extern u64 div64_u64(u64 dividend, u64 divisor);
 #endif
  
+#ifndef div64_s64
+extern s64 div64_s64(s64 dividend, s64 divisor);
+#endif
+
 #endif /* BITS_PER_LONG */
  
 /**
@@ -81,6 +114,145 @@
 #endif
  
 u32 iter_div_u64_rem(u64 dividend, u32 divisor, u64 *remainder);
+
+static __always_inline u32
+__iter_div_u64_rem(u64 dividend, u32 divisor, u64 *remainder)
+{
+	u32 ret = 0;
+
+	while (dividend >= divisor) {
+		/* The following asm() prevents the compiler from
+		   optimising this loop into a modulo operation.  */
+		asm("" : "+rm"(dividend));
+
+		dividend -= divisor;
+		ret++;
+	}
+
+	*remainder = dividend;
+
+	return ret;
+}
+
+#ifndef mul_u32_u32
+/*
+ * Many a GCC version messes this up and generates a 64x64 mult :-(
+ */
+static inline u64 mul_u32_u32(u32 a, u32 b)
+{
+	return (u64)a * b;
+}
+#endif
+
+#if defined(CONFIG_ARCH_SUPPORTS_INT128) && defined(__SIZEOF_INT128__)
+
+#ifndef mul_u64_u32_shr
+static inline u64 mul_u64_u32_shr(u64 a, u32 mul, unsigned int shift)
+{
+	return (u64)(((unsigned __int128)a * mul) >> shift);
+}
+#endif /* mul_u64_u32_shr */
+
+#ifndef mul_u64_u64_shr
+static inline u64 mul_u64_u64_shr(u64 a, u64 mul, unsigned int shift)
+{
+	return (u64)(((unsigned __int128)a * mul) >> shift);
+}
+#endif /* mul_u64_u64_shr */
+
+#else
+
+#ifndef mul_u64_u32_shr
+static inline u64 mul_u64_u32_shr(u64 a, u32 mul, unsigned int shift)
+{
+	u32 ah, al;
+	u64 ret;
+
+	al = a;
+	ah = a >> 32;
+
+	ret = mul_u32_u32(al, mul) >> shift;
+	if (ah)
+		ret += mul_u32_u32(ah, mul) << (32 - shift);
+
+	return ret;
+}
+#endif /* mul_u64_u32_shr */
+
+#ifndef mul_u64_u64_shr
+static inline u64 mul_u64_u64_shr(u64 a, u64 b, unsigned int shift)
+{
+	union {
+		u64 ll;
+		struct {
+#ifdef __BIG_ENDIAN
+			u32 high, low;
+#else
+			u32 low, high;
+#endif
+		} l;
+	} rl, rm, rn, rh, a0, b0;
+	u64 c;
+
+	a0.ll = a;
+	b0.ll = b;
+
+	rl.ll = mul_u32_u32(a0.l.low, b0.l.low);
+	rm.ll = mul_u32_u32(a0.l.low, b0.l.high);
+	rn.ll = mul_u32_u32(a0.l.high, b0.l.low);
+	rh.ll = mul_u32_u32(a0.l.high, b0.l.high);
+
+	/*
+	 * Each of these lines computes a 64-bit intermediate result into "c",
+	 * starting at bits 32-95.  The low 32-bits go into the result of the
+	 * multiplication, the high 32-bits are carried into the next step.
+	 */
+	rl.l.high = c = (u64)rl.l.high + rm.l.low + rn.l.low;
+	rh.l.low = c = (c >> 32) + rm.l.high + rn.l.high + rh.l.low;
+	rh.l.high = (c >> 32) + rh.l.high;
+
+	/*
+	 * The 128-bit result of the multiplication is in rl.ll and rh.ll,
+	 * shift it right and throw away the high part of the result.
+	 */
+	if (shift == 0)
+		return rl.ll;
+	if (shift < 64)
+		return (rl.ll >> shift) | (rh.ll << (64 - shift));
+	return rh.ll >> (shift & 63);
+}
+#endif /* mul_u64_u64_shr */
+
+#endif
+
+#ifndef mul_u64_u32_div
+static inline u64 mul_u64_u32_div(u64 a, u32 mul, u32 divisor)
+{
+	union {
+		u64 ll;
+		struct {
+#ifdef __BIG_ENDIAN
+			u32 high, low;
+#else
+			u32 low, high;
+#endif
+		} l;
+	} u, rl, rh;
+
+	u.ll = a;
+	rl.ll = mul_u32_u32(u.l.low, mul);
+	rh.ll = mul_u32_u32(u.l.high, mul) + rl.l.high;
+
+	/* Bits 32-63 of the result will be in rh.l.low. */
+	rl.l.high = do_div(rh.ll, divisor);
+
+	/* Bits 0-31 of the result will be in rl.l.low.	*/
+	do_div(rl.ll, divisor);
+
+	rl.l.high = rh.l.low;
+	return rl.ll;
+}
+#endif /* mul_u64_u32_div */
  
 #endif /* _LINUX_MATH64_H */
@@ -13,14 +13,19 @@
  *
  * Code generated for this function might be very inefficient
  * for some CPUs. __div64_32() can be overridden by linking arch-specific
- * assembly versions such as arch/powerpc/lib/div64.S and arch/sh/lib/div64.S.
+ * assembly versions such as arch/ppc/lib/div64.S and arch/sh/lib/div64.S
+ * or by defining a preprocessor macro in arch/include/asm/div64.h.
  */
  
-#include <div64.h>
-#include <linux/types.h>
-#include <linux/compiler.h>
+#include <linux/compat.h>
+#include <linux/kernel.h>
+#include <linux/math64.h>
  
-uint32_t notrace __div64_32(uint64_t *n, uint32_t base)
+/* Not needed on 64bit architectures */
+#if BITS_PER_LONG == 32
+
+#ifndef __div64_32
+uint32_t __attribute__((weak)) __div64_32(uint64_t *n, uint32_t base)
 {
 	uint64_t rem = *n;
 	uint64_t b = base;
@@ -52,4 +57,130 @@
 	*n = res;
 	return rem;
 }
+EXPORT_SYMBOL(__div64_32);
+#endif
+
+#ifndef div_s64_rem
+s64 div_s64_rem(s64 dividend, s32 divisor, s32 *remainder)
+{
+	u64 quotient;
+
+	if (dividend < 0) {
+		quotient = div_u64_rem(-dividend, abs(divisor), (u32 *)remainder);
+		*remainder = -*remainder;
+		if (divisor > 0)
+			quotient = -quotient;
+	} else {
+		quotient = div_u64_rem(dividend, abs(divisor), (u32 *)remainder);
+		if (divisor < 0)
+			quotient = -quotient;
+	}
+	return quotient;
+}
+EXPORT_SYMBOL(div_s64_rem);
+#endif
+
+/**
+ * div64_u64_rem - unsigned 64bit divide with 64bit divisor and remainder
+ * @dividend:	64bit dividend
+ * @divisor:	64bit divisor
+ * @remainder:  64bit remainder
+ *
+ * This implementation is a comparable to algorithm used by div64_u64.
+ * But this operation, which includes math for calculating the remainder,
+ * is kept distinct to avoid slowing down the div64_u64 operation on 32bit
+ * systems.
+ */
+#ifndef div64_u64_rem
+u64 div64_u64_rem(u64 dividend, u64 divisor, u64 *remainder)
+{
+	u32 high = divisor >> 32;
+	u64 quot;
+
+	if (high == 0) {
+		u32 rem32;
+		quot = div_u64_rem(dividend, divisor, &rem32);
+		*remainder = rem32;
+	} else {
+		int n = 1 + fls(high);
+		quot = div_u64(dividend >> n, divisor >> n);
+
+		if (quot != 0)
+			quot--;
+
+		*remainder = dividend - quot * divisor;
+		if (*remainder >= divisor) {
+			quot++;
+			*remainder -= divisor;
+		}
+	}
+
+	return quot;
+}
+EXPORT_SYMBOL(div64_u64_rem);
+#endif
+
+/**
+ * div64_u64 - unsigned 64bit divide with 64bit divisor
+ * @dividend:	64bit dividend
+ * @divisor:	64bit divisor
+ *
+ * This implementation is a modified version of the algorithm proposed
+ * by the book 'Hacker's Delight'.  The original source and full proof
+ * can be found here and is available for use without restriction.
+ *
+ * 'http://www.hackersdelight.org/hdcodetxt/divDouble.c.txt'
+ */
+#ifndef div64_u64
+u64 div64_u64(u64 dividend, u64 divisor)
+{
+	u32 high = divisor >> 32;
+	u64 quot;
+
+	if (high == 0) {
+		quot = div_u64(dividend, divisor);
+	} else {
+		int n = 1 + fls(high);
+		quot = div_u64(dividend >> n, divisor >> n);
+
+		if (quot != 0)
+			quot--;
+		if ((dividend - quot * divisor) >= divisor)
+			quot++;
+	}
+
+	return quot;
+}
+EXPORT_SYMBOL(div64_u64);
+#endif
+
+/**
+ * div64_s64 - signed 64bit divide with 64bit divisor
+ * @dividend:	64bit dividend
+ * @divisor:	64bit divisor
+ */
+#ifndef div64_s64
+s64 div64_s64(s64 dividend, s64 divisor)
+{
+	s64 quot, t;
+
+	quot = div64_u64(abs(dividend), abs(divisor));
+	t = (dividend ^ divisor) >> 63;
+
+	return (quot ^ t) - t;
+}
+EXPORT_SYMBOL(div64_s64);
+#endif
+
+#endif /* BITS_PER_LONG == 32 */
+
+/*
+ * Iterative div/mod for use when dividend is not expected to be much
+ * bigger than divisor.
+ */
+u32 iter_div_u64_rem(u64 dividend, u32 divisor, u64 *remainder)
+{
+	return __iter_div_u64_rem(dividend, divisor, remainder);
+}
+EXPORT_SYMBOL(iter_div_u64_rem);
...	...	@@ -4,13 +4,16 @@
4	4	* Copyright (C) 2003 Bernardo Innocenti <bernie@develer.com>
5	5	* Based on former asm-ppc/div64.h and asm-m68knommu/div64.h
6	6	*
	7	+ * Optimization for constant divisors on 32-bit machines:
	8	+ * Copyright (C) 2006-2015 Nicolas Pitre
	9	+ *
7	10	* The semantics of do_div() are:
8	11	*
9	12	* uint32_t do_div(uint64_t *n, uint32_t base)
10	13	* {
11		- * uint32_t remainder = *n % base;
12		- * n = n / base;
13		- * return remainder;
	14	+ * uint32_t remainder = *n % base;
	15	+ * n = n / base;
	16	+ * return remainder;
14	17	* }
15	18	*
16	19	* NOTE: macro parameter n is evaluated multiple times,
17	20
18	21
...	...	@@ -18,8 +21,182 @@
18	21	*/
19	22
20	23	#include <linux/types.h>
	24	+#include <linux/compiler.h>
21	25
	26	+#if BITS_PER_LONG == 64
	27	+
	28	+# define do_div(n,base) ({ \
	29	+ uint32_t __base = (base); \
	30	+ uint32_t __rem; \
	31	+ __rem = ((uint64_t)(n)) % __base; \
	32	+ (n) = ((uint64_t)(n)) / __base; \
	33	+ __rem; \
	34	+ })
	35	+
	36	+#elif BITS_PER_LONG == 32
	37	+
	38	+#include <linux/log2.h>
	39	+
	40	+/*
	41	+ * If the divisor happens to be constant, we determine the appropriate
	42	+ * inverse at compile time to turn the division into a few inline
	43	+ * multiplications which ought to be much faster. And yet only if compiling
	44	+ * with a sufficiently recent gcc version to perform proper 64-bit constant
	45	+ * propagation.
	46	+ *
	47	+ * (It is unfortunate that gcc doesn't perform all this internally.)
	48	+ */
	49	+
	50	+#ifndef __div64_const32_is_OK
	51	+#define __div64_const32_is_OK (__GNUC__ >= 4)
	52	+#endif
	53	+
	54	+#define __div64_const32(n, ___b) \
	55	+({ \
	56	+ /* \
	57	+ * Multiplication by reciprocal of b: n / b = n * (p / b) / p \
	58	+ * \
	59	+ * We rely on the fact that most of this code gets optimized \
	60	+ * away at compile time due to constant propagation and only \
	61	+ * a few multiplication instructions should remain. \
	62	+ * Hence this monstrous macro (static inline doesn't always \
	63	+ * do the trick here). \
	64	+ */ \
	65	+ uint64_t ___res, ___x, ___t, ___m, ___n = (n); \
	66	+ uint32_t ___p, ___bias; \
	67	+ \
	68	+ /* determine MSB of b */ \
	69	+ ___p = 1 << ilog2(___b); \
	70	+ \
	71	+ /* compute m = ((p << 64) + b - 1) / b */ \
	72	+ ___m = (~0ULL / ___b) * ___p; \
	73	+ ___m += (((~0ULL % ___b + 1) * ___p) + ___b - 1) / ___b; \
	74	+ \
	75	+ /* one less than the dividend with highest result */ \
	76	+ ___x = ~0ULL / ___b * ___b - 1; \
	77	+ \
	78	+ /* test our ___m with res = m * x / (p << 64) */ \
	79	+ ___res = ((___m & 0xffffffff) * (___x & 0xffffffff)) >> 32; \
	80	+ ___t = ___res += (___m & 0xffffffff) * (___x >> 32); \
	81	+ ___res += (___x & 0xffffffff) * (___m >> 32); \
	82	+ ___t = (___res < ___t) ? (1ULL << 32) : 0; \
	83	+ ___res = (___res >> 32) + ___t; \
	84	+ ___res += (___m >> 32) * (___x >> 32); \
	85	+ ___res /= ___p; \
	86	+ \
	87	+ /* Now sanitize and optimize what we've got. */ \
	88	+ if (~0ULL % (___b / (___b & -___b)) == 0) { \
	89	+ /* special case, can be simplified to ... */ \
	90	+ ___n /= (___b & -___b); \
	91	+ ___m = ~0ULL / (___b / (___b & -___b)); \
	92	+ ___p = 1; \
	93	+ ___bias = 1; \
	94	+ } else if (___res != ___x / ___b) { \
	95	+ /* \
	96	+ * We can't get away without a bias to compensate \
	97	+ * for bit truncation errors. To avoid it we'd need an \
	98	+ * additional bit to represent m which would overflow \
	99	+ * a 64-bit variable. \
	100	+ * \
	101	+ * Instead we do m = p / b and n / b = (n * m + m) / p. \
	102	+ */ \
	103	+ ___bias = 1; \
	104	+ /* Compute m = (p << 64) / b */ \
	105	+ ___m = (~0ULL / ___b) * ___p; \
	106	+ ___m += ((~0ULL % ___b + 1) * ___p) / ___b; \
	107	+ } else { \
	108	+ /* \
	109	+ * Reduce m / p, and try to clear bit 31 of m when \
	110	+ * possible, otherwise that'll need extra overflow \
	111	+ * handling later. \
	112	+ */ \
	113	+ uint32_t ___bits = -(___m & -___m); \
	114	+ ___bits \|= ___m >> 32; \
	115	+ ___bits = (~___bits) << 1; \
	116	+ /* \
	117	+ * If ___bits == 0 then setting bit 31 is unavoidable. \
	118	+ * Simply apply the maximum possible reduction in that \
	119	+ * case. Otherwise the MSB of ___bits indicates the \
	120	+ * best reduction we should apply. \
	121	+ */ \
	122	+ if (!___bits) { \
	123	+ ___p /= (___m & -___m); \
	124	+ ___m /= (___m & -___m); \
	125	+ } else { \
	126	+ ___p >>= ilog2(___bits); \
	127	+ ___m >>= ilog2(___bits); \
	128	+ } \
	129	+ /* No bias needed. */ \
	130	+ ___bias = 0; \
	131	+ } \
	132	+ \
	133	+ /* \
	134	+ * Now we have a combination of 2 conditions: \
	135	+ * \
	136	+ * 1) whether or not we need to apply a bias, and \
	137	+ * \
	138	+ * 2) whether or not there might be an overflow in the cross \
	139	+ * product determined by (___m & ((1 << 63) \| (1 << 31))). \
	140	+ * \
	141	+ * Select the best way to do (m_bias + m * n) / (1 << 64). \
	142	+ * From now on there will be actual runtime code generated. \
	143	+ */ \
	144	+ ___res = __arch_xprod_64(___m, ___n, ___bias); \
	145	+ \
	146	+ ___res /= ___p; \
	147	+})
	148	+
	149	+#ifndef __arch_xprod_64
	150	+/*
	151	+ * Default C implementation for __arch_xprod_64()
	152	+ *
	153	+ * Prototype: uint64_t __arch_xprod_64(const uint64_t m, uint64_t n, bool bias)
	154	+ * Semantic: retval = ((bias ? m : 0) + m * n) >> 64
	155	+ *
	156	+ * The product is a 128-bit value, scaled down to 64 bits.
	157	+ * Assuming constant propagation to optimize away unused conditional code.
	158	+ * Architectures may provide their own optimized assembly implementation.
	159	+ */
	160	+static inline uint64_t __arch_xprod_64(const uint64_t m, uint64_t n, bool bias)
	161	+{
	162	+ uint32_t m_lo = m;
	163	+ uint32_t m_hi = m >> 32;
	164	+ uint32_t n_lo = n;
	165	+ uint32_t n_hi = n >> 32;
	166	+ uint64_t res, tmp;
	167	+
	168	+ if (!bias) {
	169	+ res = ((uint64_t)m_lo * n_lo) >> 32;
	170	+ } else if (!(m & ((1ULL << 63) \| (1ULL << 31)))) {
	171	+ /* there can't be any overflow here */
	172	+ res = (m + (uint64_t)m_lo * n_lo) >> 32;
	173	+ } else {
	174	+ res = m + (uint64_t)m_lo * n_lo;
	175	+ tmp = (res < m) ? (1ULL << 32) : 0;
	176	+ res = (res >> 32) + tmp;
	177	+ }
	178	+
	179	+ if (!(m & ((1ULL << 63) \| (1ULL << 31)))) {
	180	+ /* there can't be any overflow here */
	181	+ res += (uint64_t)m_lo * n_hi;
	182	+ res += (uint64_t)m_hi * n_lo;
	183	+ res >>= 32;
	184	+ } else {
	185	+ tmp = res += (uint64_t)m_lo * n_hi;
	186	+ res += (uint64_t)m_hi * n_lo;
	187	+ tmp = (res < tmp) ? (1ULL << 32) : 0;
	188	+ res = (res >> 32) + tmp;
	189	+ }
	190	+
	191	+ res += (uint64_t)m_hi * n_hi;
	192	+
	193	+ return res;
	194	+}
	195	+#endif
	196	+
	197	+#ifndef __div64_32
22	198	extern uint32_t __div64_32(uint64_t *dividend, uint32_t divisor);
	199	+#endif
23	200
24	201	/* The unnecessary pointer compare is there
25	202	* to check for type safety (n must be 64bit)
26	203
27	204
...	...	@@ -28,13 +205,31 @@
28	205	uint32_t __base = (base); \
29	206	uint32_t __rem; \
30	207	(void)(((typeof((n)) )0) == ((uint64_t )0)); \
31		- if (((n) >> 32) == 0) { \
	208	+ if (__builtin_constant_p(__base) && \
	209	+ is_power_of_2(__base)) { \
	210	+ __rem = (n) & (__base - 1); \
	211	+ (n) >>= ilog2(__base); \
	212	+ } else if (__div64_const32_is_OK && \
	213	+ __builtin_constant_p(__base) && \
	214	+ __base != 0) { \
	215	+ uint32_t __res_lo, __n_lo = (n); \
	216	+ (n) = __div64_const32(n, __base); \
	217	+ /* the remainder can be computed with 32-bit regs */ \
	218	+ __res_lo = (n); \
	219	+ __rem = __n_lo - __res_lo * __base; \
	220	+ } else if (likely(((n) >> 32) == 0)) { \
32	221	__rem = (uint32_t)(n) % __base; \
33	222	(n) = (uint32_t)(n) / __base; \
34		- } else \
	223	+ } else \
35	224	__rem = __div64_32(&(n), __base); \
36	225	__rem; \
37	226	})
	227	+
	228	+#else /* BITS_PER_LONG == ?? */
	229	+
	230	+# error do_div() does not yet support the C64
	231	+
	232	+#endif /* BITS_PER_LONG */
38	233
39	234	/* Wrapper for do_div(). Doesn't modify dividend and returns
40	235	* the result, not reminder.
1	1	#ifndef _LINUX_MATH64_H
2	2	#define _LINUX_MATH64_H
3	3
	4	+#include <div64.h>
	5	+#include <linux/bitops.h>
4	6	#include <linux/types.h>
5	7
6	8	#if BITS_PER_LONG == 64
7	9
	10	+#define div64_long(x, y) div64_s64((x), (y))
	11	+#define div64_ul(x, y) div64_u64((x), (y))
	12	+
8	13	/**
9	14	* div_u64_rem - unsigned 64bit divide with 32bit divisor with remainder
10	15	*
...	...	@@ -27,6 +32,15 @@
27	32	}
28	33
29	34	/**
	35	+ * div64_u64_rem - unsigned 64bit divide with 64bit divisor and remainder
	36	+ */
	37	+static inline u64 div64_u64_rem(u64 dividend, u64 divisor, u64 *remainder)
	38	+{
	39	+ *remainder = dividend % divisor;
	40	+ return dividend / divisor;
	41	+}
	42	+
	43	+/**
30	44	* div64_u64 - unsigned 64bit divide with 64bit divisor
31	45	*/
32	46	static inline u64 div64_u64(u64 dividend, u64 divisor)
33	47
...	...	@@ -34,8 +48,19 @@
34	48	return dividend / divisor;
35	49	}
36	50
	51	+/**
	52	+ * div64_s64 - signed 64bit divide with 64bit divisor
	53	+ */
	54	+static inline s64 div64_s64(s64 dividend, s64 divisor)
	55	+{
	56	+ return dividend / divisor;
	57	+}
	58	+
37	59	#elif BITS_PER_LONG == 32
38	60
	61	+#define div64_long(x, y) div_s64((x), (y))
	62	+#define div64_ul(x, y) div_u64((x), (y))
	63	+
39	64	#ifndef div_u64_rem
40	65	static inline u64 div_u64_rem(u64 dividend, u32 divisor, u32 *remainder)
41	66	{
42	67
...	...	@@ -48,10 +73,18 @@
48	73	extern s64 div_s64_rem(s64 dividend, s32 divisor, s32 *remainder);
49	74	#endif
50	75
	76	+#ifndef div64_u64_rem
	77	+extern u64 div64_u64_rem(u64 dividend, u64 divisor, u64 *remainder);
	78	+#endif
	79	+
51	80	#ifndef div64_u64
52	81	extern u64 div64_u64(u64 dividend, u64 divisor);
53	82	#endif
54	83
	84	+#ifndef div64_s64
	85	+extern s64 div64_s64(s64 dividend, s64 divisor);
	86	+#endif
	87	+
55	88	#endif /* BITS_PER_LONG */
56	89
57	90	/**
...	...	@@ -81,6 +114,145 @@
81	114	#endif
82	115
83	116	u32 iter_div_u64_rem(u64 dividend, u32 divisor, u64 *remainder);
	117	+
	118	+static __always_inline u32
	119	+__iter_div_u64_rem(u64 dividend, u32 divisor, u64 *remainder)
	120	+{
	121	+ u32 ret = 0;
	122	+
	123	+ while (dividend >= divisor) {
	124	+ /* The following asm() prevents the compiler from
	125	+ optimising this loop into a modulo operation. */
	126	+ asm("" : "+rm"(dividend));
	127	+
	128	+ dividend -= divisor;
	129	+ ret++;
	130	+ }
	131	+
	132	+ *remainder = dividend;
	133	+
	134	+ return ret;
	135	+}
	136	+
	137	+#ifndef mul_u32_u32
	138	+/*
	139	+ * Many a GCC version messes this up and generates a 64x64 mult :-(
	140	+ */
	141	+static inline u64 mul_u32_u32(u32 a, u32 b)
	142	+{
	143	+ return (u64)a * b;
	144	+}
	145	+#endif
	146	+
	147	+#if defined(CONFIG_ARCH_SUPPORTS_INT128) && defined(__SIZEOF_INT128__)
	148	+
	149	+#ifndef mul_u64_u32_shr
	150	+static inline u64 mul_u64_u32_shr(u64 a, u32 mul, unsigned int shift)
	151	+{
	152	+ return (u64)(((unsigned __int128)a * mul) >> shift);
	153	+}
	154	+#endif /* mul_u64_u32_shr */
	155	+
	156	+#ifndef mul_u64_u64_shr
	157	+static inline u64 mul_u64_u64_shr(u64 a, u64 mul, unsigned int shift)
	158	+{
	159	+ return (u64)(((unsigned __int128)a * mul) >> shift);
	160	+}
	161	+#endif /* mul_u64_u64_shr */
	162	+
	163	+#else
	164	+
	165	+#ifndef mul_u64_u32_shr
	166	+static inline u64 mul_u64_u32_shr(u64 a, u32 mul, unsigned int shift)
	167	+{
	168	+ u32 ah, al;
	169	+ u64 ret;
	170	+
	171	+ al = a;
	172	+ ah = a >> 32;
	173	+
	174	+ ret = mul_u32_u32(al, mul) >> shift;
	175	+ if (ah)
	176	+ ret += mul_u32_u32(ah, mul) << (32 - shift);
	177	+
	178	+ return ret;
	179	+}
	180	+#endif /* mul_u64_u32_shr */
	181	+
	182	+#ifndef mul_u64_u64_shr
	183	+static inline u64 mul_u64_u64_shr(u64 a, u64 b, unsigned int shift)
	184	+{
	185	+ union {
	186	+ u64 ll;
	187	+ struct {
	188	+#ifdef __BIG_ENDIAN
	189	+ u32 high, low;
	190	+#else
	191	+ u32 low, high;
	192	+#endif
	193	+ } l;
	194	+ } rl, rm, rn, rh, a0, b0;
	195	+ u64 c;
	196	+
	197	+ a0.ll = a;
	198	+ b0.ll = b;
	199	+
	200	+ rl.ll = mul_u32_u32(a0.l.low, b0.l.low);
	201	+ rm.ll = mul_u32_u32(a0.l.low, b0.l.high);
	202	+ rn.ll = mul_u32_u32(a0.l.high, b0.l.low);
	203	+ rh.ll = mul_u32_u32(a0.l.high, b0.l.high);
	204	+
	205	+ /*
	206	+ * Each of these lines computes a 64-bit intermediate result into "c",
	207	+ * starting at bits 32-95. The low 32-bits go into the result of the
	208	+ * multiplication, the high 32-bits are carried into the next step.
	209	+ */
	210	+ rl.l.high = c = (u64)rl.l.high + rm.l.low + rn.l.low;
	211	+ rh.l.low = c = (c >> 32) + rm.l.high + rn.l.high + rh.l.low;
	212	+ rh.l.high = (c >> 32) + rh.l.high;
	213	+
	214	+ /*
	215	+ * The 128-bit result of the multiplication is in rl.ll and rh.ll,
	216	+ * shift it right and throw away the high part of the result.
	217	+ */
	218	+ if (shift == 0)
	219	+ return rl.ll;
	220	+ if (shift < 64)
	221	+ return (rl.ll >> shift) \| (rh.ll << (64 - shift));
	222	+ return rh.ll >> (shift & 63);
	223	+}
	224	+#endif /* mul_u64_u64_shr */
	225	+
	226	+#endif
	227	+
	228	+#ifndef mul_u64_u32_div
	229	+static inline u64 mul_u64_u32_div(u64 a, u32 mul, u32 divisor)
	230	+{
	231	+ union {
	232	+ u64 ll;
	233	+ struct {
	234	+#ifdef __BIG_ENDIAN
	235	+ u32 high, low;
	236	+#else
	237	+ u32 low, high;
	238	+#endif
	239	+ } l;
	240	+ } u, rl, rh;
	241	+
	242	+ u.ll = a;
	243	+ rl.ll = mul_u32_u32(u.l.low, mul);
	244	+ rh.ll = mul_u32_u32(u.l.high, mul) + rl.l.high;
	245	+
	246	+ /* Bits 32-63 of the result will be in rh.l.low. */
	247	+ rl.l.high = do_div(rh.ll, divisor);
	248	+
	249	+ /* Bits 0-31 of the result will be in rl.l.low. */
	250	+ do_div(rl.ll, divisor);
	251	+
	252	+ rl.l.high = rh.l.low;
	253	+ return rl.ll;
	254	+}
	255	+#endif /* mul_u64_u32_div */
84	256
85	257	#endif /* _LINUX_MATH64_H */
...	...	@@ -13,14 +13,19 @@
13	13	*
14	14	* Code generated for this function might be very inefficient
15	15	* for some CPUs. __div64_32() can be overridden by linking arch-specific
16		- * assembly versions such as arch/powerpc/lib/div64.S and arch/sh/lib/div64.S.
	16	+ * assembly versions such as arch/ppc/lib/div64.S and arch/sh/lib/div64.S
	17	+ * or by defining a preprocessor macro in arch/include/asm/div64.h.
17	18	*/
18	19
19		-#include <div64.h>
20		-#include <linux/types.h>
21		-#include <linux/compiler.h>
	20	+#include <linux/compat.h>
	21	+#include <linux/kernel.h>
	22	+#include <linux/math64.h>
22	23
23		-uint32_t notrace __div64_32(uint64_t *n, uint32_t base)
	24	+/* Not needed on 64bit architectures */
	25	+#if BITS_PER_LONG == 32
	26	+
	27	+#ifndef __div64_32
	28	+uint32_t __attribute__((weak)) __div64_32(uint64_t *n, uint32_t base)
24	29	{
25	30	uint64_t rem = *n;
26	31	uint64_t b = base;
...	...	@@ -52,4 +57,130 @@
52	57	*n = res;
53	58	return rem;
54	59	}
	60	+EXPORT_SYMBOL(__div64_32);
	61	+#endif
	62	+
	63	+#ifndef div_s64_rem
	64	+s64 div_s64_rem(s64 dividend, s32 divisor, s32 *remainder)
	65	+{
	66	+ u64 quotient;
	67	+
	68	+ if (dividend < 0) {
	69	+ quotient = div_u64_rem(-dividend, abs(divisor), (u32 *)remainder);
	70	+ remainder = -remainder;
	71	+ if (divisor > 0)
	72	+ quotient = -quotient;
	73	+ } else {
	74	+ quotient = div_u64_rem(dividend, abs(divisor), (u32 *)remainder);
	75	+ if (divisor < 0)
	76	+ quotient = -quotient;
	77	+ }
	78	+ return quotient;
	79	+}
	80	+EXPORT_SYMBOL(div_s64_rem);
	81	+#endif
	82	+
	83	+/**
	84	+ * div64_u64_rem - unsigned 64bit divide with 64bit divisor and remainder
	85	+ * @dividend: 64bit dividend
	86	+ * @divisor: 64bit divisor
	87	+ * @remainder: 64bit remainder
	88	+ *
	89	+ * This implementation is a comparable to algorithm used by div64_u64.
	90	+ * But this operation, which includes math for calculating the remainder,
	91	+ * is kept distinct to avoid slowing down the div64_u64 operation on 32bit
	92	+ * systems.
	93	+ */
	94	+#ifndef div64_u64_rem
	95	+u64 div64_u64_rem(u64 dividend, u64 divisor, u64 *remainder)
	96	+{
	97	+ u32 high = divisor >> 32;
	98	+ u64 quot;
	99	+
	100	+ if (high == 0) {
	101	+ u32 rem32;
	102	+ quot = div_u64_rem(dividend, divisor, &rem32);
	103	+ *remainder = rem32;
	104	+ } else {
	105	+ int n = 1 + fls(high);
	106	+ quot = div_u64(dividend >> n, divisor >> n);
	107	+
	108	+ if (quot != 0)
	109	+ quot--;
	110	+
	111	+ remainder = dividend - quot divisor;
	112	+ if (*remainder >= divisor) {
	113	+ quot++;
	114	+ *remainder -= divisor;
	115	+ }
	116	+ }
	117	+
	118	+ return quot;
	119	+}
	120	+EXPORT_SYMBOL(div64_u64_rem);
	121	+#endif
	122	+
	123	+/**
	124	+ * div64_u64 - unsigned 64bit divide with 64bit divisor
	125	+ * @dividend: 64bit dividend
	126	+ * @divisor: 64bit divisor
	127	+ *
	128	+ * This implementation is a modified version of the algorithm proposed
	129	+ * by the book 'Hacker's Delight'. The original source and full proof
	130	+ * can be found here and is available for use without restriction.
	131	+ *
	132	+ * 'http://www.hackersdelight.org/hdcodetxt/divDouble.c.txt'
	133	+ */
	134	+#ifndef div64_u64
	135	+u64 div64_u64(u64 dividend, u64 divisor)
	136	+{
	137	+ u32 high = divisor >> 32;
	138	+ u64 quot;
	139	+
	140	+ if (high == 0) {
	141	+ quot = div_u64(dividend, divisor);
	142	+ } else {
	143	+ int n = 1 + fls(high);
	144	+ quot = div_u64(dividend >> n, divisor >> n);
	145	+
	146	+ if (quot != 0)
	147	+ quot--;
	148	+ if ((dividend - quot * divisor) >= divisor)
	149	+ quot++;
	150	+ }
	151	+
	152	+ return quot;
	153	+}
	154	+EXPORT_SYMBOL(div64_u64);
	155	+#endif
	156	+
	157	+/**
	158	+ * div64_s64 - signed 64bit divide with 64bit divisor
	159	+ * @dividend: 64bit dividend
	160	+ * @divisor: 64bit divisor
	161	+ */
	162	+#ifndef div64_s64
	163	+s64 div64_s64(s64 dividend, s64 divisor)
	164	+{
	165	+ s64 quot, t;
	166	+
	167	+ quot = div64_u64(abs(dividend), abs(divisor));
	168	+ t = (dividend ^ divisor) >> 63;
	169	+
	170	+ return (quot ^ t) - t;
	171	+}
	172	+EXPORT_SYMBOL(div64_s64);
	173	+#endif
	174	+
	175	+#endif /* BITS_PER_LONG == 32 */
	176	+
	177	+/*
	178	+ * Iterative div/mod for use when dividend is not expected to be much
	179	+ * bigger than divisor.
	180	+ */
	181	+u32 iter_div_u64_rem(u64 dividend, u32 divisor, u64 *remainder)
	182	+{
	183	+ return __iter_div_u64_rem(dividend, divisor, remainder);
	184	+}
	185	+EXPORT_SYMBOL(iter_div_u64_rem);