VisionFive2 Linux kernel

b24413180f560 (Greg Kroah-Hartman 2017-11-01 15:07:57 +0100   1) // SPDX-License-Identifier: GPL-2.0
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700   2) /*
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700   3)  * A fast, small, non-recursive O(n log n) sort for the Linux kernel
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700   4)  *
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700   5)  * This performs n*log2(n) + 0.37*n + o(n) comparisons on average,
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700   6)  * and 1.5*n*log2(n) + O(n) in the (very contrived) worst case.
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700   7)  *
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700   8)  * Glibc qsort() manages n*log2(n) - 1.26*n for random inputs (1.63*n
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700   9)  * better) at the expense of stack usage and much larger code to avoid
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700  10)  * quicksort's O(n^2) worst case.
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700  11)  */
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700  12) 
c5adae9583ef6 (Kostenzer Felix    2017-02-24 15:01:07 -0800  13) #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
c5adae9583ef6 (Kostenzer Felix    2017-02-24 15:01:07 -0800  14) 
42cf809654e4e (Rasmus Villemoes   2015-02-12 15:02:35 -0800  15) #include <linux/types.h>
42cf809654e4e (Rasmus Villemoes   2015-02-12 15:02:35 -0800  16) #include <linux/export.h>
ecec4cb7a9df5 (Adrian Bunk        2005-09-10 00:26:59 -0700  17) #include <linux/sort.h>
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700  18) 
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  19) /**
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  20)  * is_aligned - is this pointer & size okay for word-wide copying?
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  21)  * @base: pointer to data
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  22)  * @size: size of each element
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700  23)  * @align: required alignment (typically 4 or 8)
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  24)  *
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  25)  * Returns true if elements can be copied using word loads and stores.
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  26)  * The size must be a multiple of the alignment, and the base address must
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  27)  * be if we do not have CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS.
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  28)  *
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  29)  * For some reason, gcc doesn't know to optimize "if (a & mask || b & mask)"
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  30)  * to "if ((a | b) & mask)", so we do that by hand.
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  31)  */
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  32) __attribute_const__ __always_inline
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  33) static bool is_aligned(const void *base, size_t size, unsigned char align)
ca96ab859ab4b (Daniel Wagner      2015-06-25 15:02:14 -0700  34) {
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  35) 	unsigned char lsbits = (unsigned char)size;
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  36) 
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  37) 	(void)base;
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  38) #ifndef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  39) 	lsbits |= (unsigned char)(uintptr_t)base;
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  40) #endif
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  41) 	return (lsbits & (align - 1)) == 0;
ca96ab859ab4b (Daniel Wagner      2015-06-25 15:02:14 -0700  42) }
ca96ab859ab4b (Daniel Wagner      2015-06-25 15:02:14 -0700  43) 
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  44) /**
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  45)  * swap_words_32 - swap two elements in 32-bit chunks
aa52619ccbe05 (Randy Dunlap       2019-05-31 22:30:00 -0700  46)  * @a: pointer to the first element to swap
aa52619ccbe05 (Randy Dunlap       2019-05-31 22:30:00 -0700  47)  * @b: pointer to the second element to swap
aa52619ccbe05 (Randy Dunlap       2019-05-31 22:30:00 -0700  48)  * @n: element size (must be a multiple of 4)
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  49)  *
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  50)  * Exchange the two objects in memory.  This exploits base+index addressing,
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  51)  * which basically all CPUs have, to minimize loop overhead computations.
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  52)  *
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  53)  * For some reason, on x86 gcc 7.3.0 adds a redundant test of n at the
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  54)  * bottom of the loop, even though the zero flag is stil valid from the
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  55)  * subtract (since the intervening mov instructions don't alter the flags).
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  56)  * Gcc 8.1.0 doesn't have that problem.
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  57)  */
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700  58) static void swap_words_32(void *a, void *b, size_t n)
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700  59) {
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  60) 	do {
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  61) 		u32 t = *(u32 *)(a + (n -= 4));
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  62) 		*(u32 *)(a + n) = *(u32 *)(b + n);
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  63) 		*(u32 *)(b + n) = t;
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  64) 	} while (n);
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700  65) }
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700  66) 
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  67) /**
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  68)  * swap_words_64 - swap two elements in 64-bit chunks
aa52619ccbe05 (Randy Dunlap       2019-05-31 22:30:00 -0700  69)  * @a: pointer to the first element to swap
aa52619ccbe05 (Randy Dunlap       2019-05-31 22:30:00 -0700  70)  * @b: pointer to the second element to swap
aa52619ccbe05 (Randy Dunlap       2019-05-31 22:30:00 -0700  71)  * @n: element size (must be a multiple of 8)
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  72)  *
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  73)  * Exchange the two objects in memory.  This exploits base+index
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  74)  * addressing, which basically all CPUs have, to minimize loop overhead
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  75)  * computations.
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  76)  *
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  77)  * We'd like to use 64-bit loads if possible.  If they're not, emulating
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  78)  * one requires base+index+4 addressing which x86 has but most other
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  79)  * processors do not.  If CONFIG_64BIT, we definitely have 64-bit loads,
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  80)  * but it's possible to have 64-bit loads without 64-bit pointers (e.g.
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  81)  * x32 ABI).  Are there any cases the kernel needs to worry about?
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  82)  */
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700  83) static void swap_words_64(void *a, void *b, size_t n)
ca96ab859ab4b (Daniel Wagner      2015-06-25 15:02:14 -0700  84) {
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  85) 	do {
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  86) #ifdef CONFIG_64BIT
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  87) 		u64 t = *(u64 *)(a + (n -= 8));
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  88) 		*(u64 *)(a + n) = *(u64 *)(b + n);
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  89) 		*(u64 *)(b + n) = t;
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  90) #else
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  91) 		/* Use two 32-bit transfers to avoid base+index+4 addressing */
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  92) 		u32 t = *(u32 *)(a + (n -= 4));
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  93) 		*(u32 *)(a + n) = *(u32 *)(b + n);
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  94) 		*(u32 *)(b + n) = t;
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  95) 
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  96) 		t = *(u32 *)(a + (n -= 4));
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  97) 		*(u32 *)(a + n) = *(u32 *)(b + n);
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  98) 		*(u32 *)(b + n) = t;
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700  99) #endif
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700 100) 	} while (n);
ca96ab859ab4b (Daniel Wagner      2015-06-25 15:02:14 -0700 101) }
ca96ab859ab4b (Daniel Wagner      2015-06-25 15:02:14 -0700 102) 
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700 103) /**
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700 104)  * swap_bytes - swap two elements a byte at a time
aa52619ccbe05 (Randy Dunlap       2019-05-31 22:30:00 -0700 105)  * @a: pointer to the first element to swap
aa52619ccbe05 (Randy Dunlap       2019-05-31 22:30:00 -0700 106)  * @b: pointer to the second element to swap
aa52619ccbe05 (Randy Dunlap       2019-05-31 22:30:00 -0700 107)  * @n: element size
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700 108)  *
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700 109)  * This is the fallback if alignment doesn't allow using larger chunks.
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700 110)  */
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 111) static void swap_bytes(void *a, void *b, size_t n)
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 112) {
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 113) 	do {
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700 114) 		char t = ((char *)a)[--n];
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700 115) 		((char *)a)[n] = ((char *)b)[n];
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700 116) 		((char *)b)[n] = t;
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700 117) 	} while (n);
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 118) }
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 119) 
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 120) /*
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 121)  * The values are arbitrary as long as they can't be confused with
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 122)  * a pointer, but small integers make for the smallest compare
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 123)  * instructions.
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 124)  */
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 125) #define SWAP_WORDS_64 (swap_func_t)0
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 126) #define SWAP_WORDS_32 (swap_func_t)1
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 127) #define SWAP_BYTES    (swap_func_t)2
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 128) 
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 129) /*
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 130)  * The function pointer is last to make tail calls most efficient if the
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 131)  * compiler decides not to inline this function.
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 132)  */
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 133) static void do_swap(void *a, void *b, size_t size, swap_func_t swap_func)
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 134) {
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 135) 	if (swap_func == SWAP_WORDS_64)
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 136) 		swap_words_64(a, b, size);
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 137) 	else if (swap_func == SWAP_WORDS_32)
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 138) 		swap_words_32(a, b, size);
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 139) 	else if (swap_func == SWAP_BYTES)
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 140) 		swap_bytes(a, b, size);
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 141) 	else
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 142) 		swap_func(a, b, (int)size);
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 143) }
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 144) 
4333fb96ca108 (Rasmus Villemoes   2019-08-16 13:01:22 -0300 145) #define _CMP_WRAPPER ((cmp_r_func_t)0L)
4333fb96ca108 (Rasmus Villemoes   2019-08-16 13:01:22 -0300 146) 
52ae533b8a18e (Andy Shevchenko    2019-10-07 16:56:54 +0300 147) static int do_cmp(const void *a, const void *b, cmp_r_func_t cmp, const void *priv)
4333fb96ca108 (Rasmus Villemoes   2019-08-16 13:01:22 -0300 148) {
4333fb96ca108 (Rasmus Villemoes   2019-08-16 13:01:22 -0300 149) 	if (cmp == _CMP_WRAPPER)
4333fb96ca108 (Rasmus Villemoes   2019-08-16 13:01:22 -0300 150) 		return ((cmp_func_t)(priv))(a, b);
4333fb96ca108 (Rasmus Villemoes   2019-08-16 13:01:22 -0300 151) 	return cmp(a, b, priv);
4333fb96ca108 (Rasmus Villemoes   2019-08-16 13:01:22 -0300 152) }
4333fb96ca108 (Rasmus Villemoes   2019-08-16 13:01:22 -0300 153) 
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 154) /**
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 155)  * parent - given the offset of the child, find the offset of the parent.
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 156)  * @i: the offset of the heap element whose parent is sought.  Non-zero.
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 157)  * @lsbit: a precomputed 1-bit mask, equal to "size & -size"
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 158)  * @size: size of each element
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 159)  *
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 160)  * In terms of array indexes, the parent of element j = @i/@size is simply
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 161)  * (j-1)/2.  But when working in byte offsets, we can't use implicit
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 162)  * truncation of integer divides.
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 163)  *
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 164)  * Fortunately, we only need one bit of the quotient, not the full divide.
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 165)  * @size has a least significant bit.  That bit will be clear if @i is
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 166)  * an even multiple of @size, and set if it's an odd multiple.
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 167)  *
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 168)  * Logically, we're doing "if (i & lsbit) i -= size;", but since the
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 169)  * branch is unpredictable, it's done with a bit of clever branch-free
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 170)  * code instead.
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 171)  */
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 172) __attribute_const__ __always_inline
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 173) static size_t parent(size_t i, unsigned int lsbit, size_t size)
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 174) {
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 175) 	i -= size;
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 176) 	i -= size & -(i & lsbit);
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 177) 	return i / 2;
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 178) }
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 179) 
72fd4a35a8243 (Robert P. J. Day   2007-02-10 01:45:59 -0800 180) /**
4333fb96ca108 (Rasmus Villemoes   2019-08-16 13:01:22 -0300 181)  * sort_r - sort an array of elements
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 182)  * @base: pointer to data to sort
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 183)  * @num: number of elements
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 184)  * @size: size of each element
b53907c0100a3 (Wu Fengguang       2009-01-07 18:09:11 -0800 185)  * @cmp_func: pointer to comparison function
b53907c0100a3 (Wu Fengguang       2009-01-07 18:09:11 -0800 186)  * @swap_func: pointer to swap function or NULL
4333fb96ca108 (Rasmus Villemoes   2019-08-16 13:01:22 -0300 187)  * @priv: third argument passed to comparison function
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 188)  *
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700 189)  * This function does a heapsort on the given array.  You may provide
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700 190)  * a swap_func function if you need to do something more than a memory
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700 191)  * copy (e.g. fix up pointers or auxiliary data), but the built-in swap
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 192)  * avoids a slow retpoline and so is significantly faster.
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 193)  *
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 194)  * Sorting time is O(n log n) both on average and worst-case. While
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 195)  * quicksort is slightly faster on average, it suffers from exploitable
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 196)  * O(n*n) worst-case behavior and extra memory requirements that make
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 197)  * it less suitable for kernel use.
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 198)  */
4333fb96ca108 (Rasmus Villemoes   2019-08-16 13:01:22 -0300 199) void sort_r(void *base, size_t num, size_t size,
52ae533b8a18e (Andy Shevchenko    2019-10-07 16:56:54 +0300 200) 	    cmp_r_func_t cmp_func,
52ae533b8a18e (Andy Shevchenko    2019-10-07 16:56:54 +0300 201) 	    swap_func_t swap_func,
4333fb96ca108 (Rasmus Villemoes   2019-08-16 13:01:22 -0300 202) 	    const void *priv)
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 203) {
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 204) 	/* pre-scale counters for performance */
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 205) 	size_t n = num * size, a = (num/2) * size;
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 206) 	const unsigned int lsbit = size & -size;  /* Used to find parent */
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 207) 
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 208) 	if (!a)		/* num < 2 || size == 0 */
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 209) 		return;
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 210) 
ca96ab859ab4b (Daniel Wagner      2015-06-25 15:02:14 -0700 211) 	if (!swap_func) {
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700 212) 		if (is_aligned(base, size, 8))
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 213) 			swap_func = SWAP_WORDS_64;
37d0ec34d111a (George Spelvin     2019-05-14 15:42:49 -0700 214) 		else if (is_aligned(base, size, 4))
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 215) 			swap_func = SWAP_WORDS_32;
ca96ab859ab4b (Daniel Wagner      2015-06-25 15:02:14 -0700 216) 		else
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 217) 			swap_func = SWAP_BYTES;
ca96ab859ab4b (Daniel Wagner      2015-06-25 15:02:14 -0700 218) 	}
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 219) 
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 220) 	/*
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 221) 	 * Loop invariants:
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 222) 	 * 1. elements [a,n) satisfy the heap property (compare greater than
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 223) 	 *    all of their children),
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 224) 	 * 2. elements [n,num*size) are sorted, and
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 225) 	 * 3. a <= b <= c <= d <= n (whenever they are valid).
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 226) 	 */
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 227) 	for (;;) {
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 228) 		size_t b, c, d;
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 229) 
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 230) 		if (a)			/* Building heap: sift down --a */
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 231) 			a -= size;
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 232) 		else if (n -= size)	/* Sorting: Extract root to --n */
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 233) 			do_swap(base, base + n, size, swap_func);
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 234) 		else			/* Sort complete */
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 235) 			break;
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 236) 
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 237) 		/*
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 238) 		 * Sift element at "a" down into heap.  This is the
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 239) 		 * "bottom-up" variant, which significantly reduces
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 240) 		 * calls to cmp_func(): we find the sift-down path all
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 241) 		 * the way to the leaves (one compare per level), then
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 242) 		 * backtrack to find where to insert the target element.
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 243) 		 *
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 244) 		 * Because elements tend to sift down close to the leaves,
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 245) 		 * this uses fewer compares than doing two per level
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 246) 		 * on the way down.  (A bit more than half as many on
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 247) 		 * average, 3/4 worst-case.)
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 248) 		 */
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 249) 		for (b = a; c = 2*b + size, (d = c + size) < n;)
4333fb96ca108 (Rasmus Villemoes   2019-08-16 13:01:22 -0300 250) 			b = do_cmp(base + c, base + d, cmp_func, priv) >= 0 ? c : d;
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 251) 		if (d == n)	/* Special case last leaf with no sibling */
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 252) 			b = c;
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 253) 
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 254) 		/* Now backtrack from "b" to the correct location for "a" */
4333fb96ca108 (Rasmus Villemoes   2019-08-16 13:01:22 -0300 255) 		while (b != a && do_cmp(base + a, base + b, cmp_func, priv) >= 0)
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 256) 			b = parent(b, lsbit, size);
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 257) 		c = b;			/* Where "a" belongs */
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 258) 		while (b != a) {	/* Shift it into place */
22a241ccb2c19 (George Spelvin     2019-05-14 15:42:52 -0700 259) 			b = parent(b, lsbit, size);
8fb583c4258d0 (George Spelvin     2019-05-14 15:42:55 -0700 260) 			do_swap(base + b, base + c, size, swap_func);
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 261) 		}
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 262) 	}
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 263) }
4333fb96ca108 (Rasmus Villemoes   2019-08-16 13:01:22 -0300 264) EXPORT_SYMBOL(sort_r);
4333fb96ca108 (Rasmus Villemoes   2019-08-16 13:01:22 -0300 265) 
4333fb96ca108 (Rasmus Villemoes   2019-08-16 13:01:22 -0300 266) void sort(void *base, size_t num, size_t size,
52ae533b8a18e (Andy Shevchenko    2019-10-07 16:56:54 +0300 267) 	  cmp_func_t cmp_func,
52ae533b8a18e (Andy Shevchenko    2019-10-07 16:56:54 +0300 268) 	  swap_func_t swap_func)
4333fb96ca108 (Rasmus Villemoes   2019-08-16 13:01:22 -0300 269) {
4333fb96ca108 (Rasmus Villemoes   2019-08-16 13:01:22 -0300 270) 	return sort_r(base, num, size, _CMP_WRAPPER, swap_func, cmp_func);
4333fb96ca108 (Rasmus Villemoes   2019-08-16 13:01:22 -0300 271) }
^1da177e4c3f4 (Linus Torvalds     2005-04-16 15:20:36 -0700 272) EXPORT_SYMBOL(sort);