add wifi6 8852be driver

hc

2024-12-19 9370bb92b2d16684ee45cf24e879c93c509162da

 kernel/lib/sort.c
..
..
@@ -1,8 +1,13 @@
1
1
 // SPDX-License-Identifier: GPL-2.0
2
2
 /*
3
- * A fast, small, non-recursive O(nlog n) sort for the Linux kernel
3
+ * A fast, small, non-recursive O(n log n) sort for the Linux kernel
4
4
  *
5
- * Jan 23 2005  Matt Mackall <mpm@selenic.com>
5
+ * This performs n*log2(n) + 0.37*n + o(n) comparisons on average,
6
+ * and 1.5*n*log2(n) + O(n) in the (very contrived) worst case.
7
+ *
8
+ * Glibc qsort() manages n*log2(n) - 1.26*n for random inputs (1.63*n
9
+ * better) at the expense of stack usage and much larger code to avoid
10
+ * quicksort's O(n^2) worst case.
6
11
  */
7
12
 
8
13
 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
..
..
@@ -11,96 +16,257 @@
11
16
 #include <linux/export.h>
12
17
 #include <linux/sort.h>
13
18
 
14
-static int alignment_ok(const void *base, int align)
19
+/**
20
+ * is_aligned - is this pointer & size okay for word-wide copying?
21
+ * @base: pointer to data
22
+ * @size: size of each element
23
+ * @align: required alignment (typically 4 or 8)
24
+ *
25
+ * Returns true if elements can be copied using word loads and stores.
26
+ * The size must be a multiple of the alignment, and the base address must
27
+ * be if we do not have CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS.
28
+ *
29
+ * For some reason, gcc doesn't know to optimize "if (a & mask || b & mask)"
30
+ * to "if ((a | b) & mask)", so we do that by hand.
31
+ */
32
+__attribute_const__ __always_inline
33
+static bool is_aligned(const void *base, size_t size, unsigned char align)
15
34
 {
16
-	return IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) ||
17
-		((unsigned long)base & (align - 1)) == 0;
18
-}
35
+	unsigned char lsbits = (unsigned char)size;
19
36
 
20
-static void u32_swap(void *a, void *b, int size)
21
-{
22
-	u32 t = *(u32 *)a;
23
-	*(u32 *)a = *(u32 *)b;
24
-	*(u32 *)b = t;
25
-}
26
-
27
-static void u64_swap(void *a, void *b, int size)
28
-{
29
-	u64 t = *(u64 *)a;
30
-	*(u64 *)a = *(u64 *)b;
31
-	*(u64 *)b = t;
32
-}
33
-
34
-static void generic_swap(void *a, void *b, int size)
35
-{
36
-	char t;
37
-
38
-	do {
39
-		t = *(char *)a;
40
-		*(char *)a++ = *(char *)b;
41
-		*(char *)b++ = t;
42
-	} while (--size > 0);
37
+	(void)base;
38
+#ifndef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
39
+	lsbits |= (unsigned char)(uintptr_t)base;
40
+#endif
41
+	return (lsbits & (align - 1)) == 0;
43
42
 }
44
43
 
45
44
 /**
46
- * sort - sort an array of elements
45
+ * swap_words_32 - swap two elements in 32-bit chunks
46
+ * @a: pointer to the first element to swap
47
+ * @b: pointer to the second element to swap
48
+ * @n: element size (must be a multiple of 4)
49
+ *
50
+ * Exchange the two objects in memory.  This exploits base+index addressing,
51
+ * which basically all CPUs have, to minimize loop overhead computations.
52
+ *
53
+ * For some reason, on x86 gcc 7.3.0 adds a redundant test of n at the
54
+ * bottom of the loop, even though the zero flag is stil valid from the
55
+ * subtract (since the intervening mov instructions don't alter the flags).
56
+ * Gcc 8.1.0 doesn't have that problem.
57
+ */
58
+static void swap_words_32(void *a, void *b, size_t n)
59
+{
60
+	do {
61
+		u32 t = *(u32 *)(a + (n -= 4));
62
+		*(u32 *)(a + n) = *(u32 *)(b + n);
63
+		*(u32 *)(b + n) = t;
64
+	} while (n);
65
+}
66
+
67
+/**
68
+ * swap_words_64 - swap two elements in 64-bit chunks
69
+ * @a: pointer to the first element to swap
70
+ * @b: pointer to the second element to swap
71
+ * @n: element size (must be a multiple of 8)
72
+ *
73
+ * Exchange the two objects in memory.  This exploits base+index
74
+ * addressing, which basically all CPUs have, to minimize loop overhead
75
+ * computations.
76
+ *
77
+ * We'd like to use 64-bit loads if possible.  If they're not, emulating
78
+ * one requires base+index+4 addressing which x86 has but most other
79
+ * processors do not.  If CONFIG_64BIT, we definitely have 64-bit loads,
80
+ * but it's possible to have 64-bit loads without 64-bit pointers (e.g.
81
+ * x32 ABI).  Are there any cases the kernel needs to worry about?
82
+ */
83
+static void swap_words_64(void *a, void *b, size_t n)
84
+{
85
+	do {
86
+#ifdef CONFIG_64BIT
87
+		u64 t = *(u64 *)(a + (n -= 8));
88
+		*(u64 *)(a + n) = *(u64 *)(b + n);
89
+		*(u64 *)(b + n) = t;
90
+#else
91
+		/* Use two 32-bit transfers to avoid base+index+4 addressing */
92
+		u32 t = *(u32 *)(a + (n -= 4));
93
+		*(u32 *)(a + n) = *(u32 *)(b + n);
94
+		*(u32 *)(b + n) = t;
95
+
96
+		t = *(u32 *)(a + (n -= 4));
97
+		*(u32 *)(a + n) = *(u32 *)(b + n);
98
+		*(u32 *)(b + n) = t;
99
+#endif
100
+	} while (n);
101
+}
102
+
103
+/**
104
+ * swap_bytes - swap two elements a byte at a time
105
+ * @a: pointer to the first element to swap
106
+ * @b: pointer to the second element to swap
107
+ * @n: element size
108
+ *
109
+ * This is the fallback if alignment doesn't allow using larger chunks.
110
+ */
111
+static void swap_bytes(void *a, void *b, size_t n)
112
+{
113
+	do {
114
+		char t = ((char *)a)[--n];
115
+		((char *)a)[n] = ((char *)b)[n];
116
+		((char *)b)[n] = t;
117
+	} while (n);
118
+}
119
+
120
+/*
121
+ * The values are arbitrary as long as they can't be confused with
122
+ * a pointer, but small integers make for the smallest compare
123
+ * instructions.
124
+ */
125
+#define SWAP_WORDS_64 (swap_func_t)0
126
+#define SWAP_WORDS_32 (swap_func_t)1
127
+#define SWAP_BYTES    (swap_func_t)2
128
+
129
+/*
130
+ * The function pointer is last to make tail calls most efficient if the
131
+ * compiler decides not to inline this function.
132
+ */
133
+static void do_swap(void *a, void *b, size_t size, swap_func_t swap_func)
134
+{
135
+	if (swap_func == SWAP_WORDS_64)
136
+		swap_words_64(a, b, size);
137
+	else if (swap_func == SWAP_WORDS_32)
138
+		swap_words_32(a, b, size);
139
+	else if (swap_func == SWAP_BYTES)
140
+		swap_bytes(a, b, size);
141
+	else
142
+		swap_func(a, b, (int)size);
143
+}
144
+
145
+#define _CMP_WRAPPER ((cmp_r_func_t)0L)
146
+
147
+static int do_cmp(const void *a, const void *b, cmp_r_func_t cmp, const void *priv)
148
+{
149
+	if (cmp == _CMP_WRAPPER)
150
+		return ((cmp_func_t)(priv))(a, b);
151
+	return cmp(a, b, priv);
152
+}
153
+
154
+/**
155
+ * parent - given the offset of the child, find the offset of the parent.
156
+ * @i: the offset of the heap element whose parent is sought.  Non-zero.
157
+ * @lsbit: a precomputed 1-bit mask, equal to "size & -size"
158
+ * @size: size of each element
159
+ *
160
+ * In terms of array indexes, the parent of element j = @i/@size is simply
161
+ * (j-1)/2.  But when working in byte offsets, we can't use implicit
162
+ * truncation of integer divides.
163
+ *
164
+ * Fortunately, we only need one bit of the quotient, not the full divide.
165
+ * @size has a least significant bit.  That bit will be clear if @i is
166
+ * an even multiple of @size, and set if it's an odd multiple.
167
+ *
168
+ * Logically, we're doing "if (i & lsbit) i -= size;", but since the
169
+ * branch is unpredictable, it's done with a bit of clever branch-free
170
+ * code instead.
171
+ */
172
+__attribute_const__ __always_inline
173
+static size_t parent(size_t i, unsigned int lsbit, size_t size)
174
+{
175
+	i -= size;
176
+	i -= size & -(i & lsbit);
177
+	return i / 2;
178
+}
179
+
180
+/**
181
+ * sort_r - sort an array of elements
47
182
  * @base: pointer to data to sort
48
183
  * @num: number of elements
49
184
  * @size: size of each element
50
185
  * @cmp_func: pointer to comparison function
51
186
  * @swap_func: pointer to swap function or NULL
187
+ * @priv: third argument passed to comparison function
52
188
  *
53
- * This function does a heapsort on the given array. You may provide a
54
- * swap_func function optimized to your element type.
189
+ * This function does a heapsort on the given array.  You may provide
190
+ * a swap_func function if you need to do something more than a memory
191
+ * copy (e.g. fix up pointers or auxiliary data), but the built-in swap
192
+ * avoids a slow retpoline and so is significantly faster.
55
193
  *
56
194
  * Sorting time is O(n log n) both on average and worst-case. While
57
- * qsort is about 20% faster on average, it suffers from exploitable
195
+ * quicksort is slightly faster on average, it suffers from exploitable
58
196
  * O(n*n) worst-case behavior and extra memory requirements that make
59
197
  * it less suitable for kernel use.
60
198
  */
61
-
62
-void sort(void *base, size_t num, size_t size,
63
-	  int (*cmp_func)(const void *, const void *),
64
-	  void (*swap_func)(void *, void *, int size))
199
+void sort_r(void *base, size_t num, size_t size,
200
+	    cmp_r_func_t cmp_func,
201
+	    swap_func_t swap_func,
202
+	    const void *priv)
65
203
 {
66
204
 	/* pre-scale counters for performance */
67
-	int i = (num/2 - 1) * size, n = num * size, c, r;
205
+	size_t n = num * size, a = (num/2) * size;
206
+	const unsigned int lsbit = size & -size;  /* Used to find parent */
207
+
208
+	if (!a)		/* num < 2 || size == 0 */
209
+		return;
68
210
 
69
211
 	if (!swap_func) {
70
-		if (size == 4 && alignment_ok(base, 4))
71
-			swap_func = u32_swap;
72
-		else if (size == 8 && alignment_ok(base, 8))
73
-			swap_func = u64_swap;
212
+		if (is_aligned(base, size, 8))
213
+			swap_func = SWAP_WORDS_64;
214
+		else if (is_aligned(base, size, 4))
215
+			swap_func = SWAP_WORDS_32;
74
216
 		else
75
-			swap_func = generic_swap;
217
+			swap_func = SWAP_BYTES;
76
218
 	}
77
219
 
78
-	/* heapify */
79
-	for ( ; i >= 0; i -= size) {
80
-		for (r = i; r * 2 + size < n; r  = c) {
81
-			c = r * 2 + size;
82
-			if (c < n - size &&
83
-					cmp_func(base + c, base + c + size) < 0)
84
-				c += size;
85
-			if (cmp_func(base + r, base + c) >= 0)
86
-				break;
87
-			swap_func(base + r, base + c, size);
88
-		}
89
-	}
220
+	/*
221
+	 * Loop invariants:
222
+	 * 1. elements [a,n) satisfy the heap property (compare greater than
223
+	 *    all of their children),
224
+	 * 2. elements [n,num*size) are sorted, and
225
+	 * 3. a <= b <= c <= d <= n (whenever they are valid).
226
+	 */
227
+	for (;;) {
228
+		size_t b, c, d;
90
229
 
91
-	/* sort */
92
-	for (i = n - size; i > 0; i -= size) {
93
-		swap_func(base, base + i, size);
94
-		for (r = 0; r * 2 + size < i; r = c) {
95
-			c = r * 2 + size;
96
-			if (c < i - size &&
97
-					cmp_func(base + c, base + c + size) < 0)
98
-				c += size;
99
-			if (cmp_func(base + r, base + c) >= 0)
100
-				break;
101
-			swap_func(base + r, base + c, size);
230
+		if (a)			/* Building heap: sift down --a */
231
+			a -= size;
232
+		else if (n -= size)	/* Sorting: Extract root to --n */
233
+			do_swap(base, base + n, size, swap_func);
234
+		else			/* Sort complete */
235
+			break;
236
+
237
+		/*
238
+		 * Sift element at "a" down into heap.  This is the
239
+		 * "bottom-up" variant, which significantly reduces
240
+		 * calls to cmp_func(): we find the sift-down path all
241
+		 * the way to the leaves (one compare per level), then
242
+		 * backtrack to find where to insert the target element.
243
+		 *
244
+		 * Because elements tend to sift down close to the leaves,
245
+		 * this uses fewer compares than doing two per level
246
+		 * on the way down.  (A bit more than half as many on
247
+		 * average, 3/4 worst-case.)
248
+		 */
249
+		for (b = a; c = 2*b + size, (d = c + size) < n;)
250
+			b = do_cmp(base + c, base + d, cmp_func, priv) >= 0 ? c : d;
251
+		if (d == n)	/* Special case last leaf with no sibling */
252
+			b = c;
253
+
254
+		/* Now backtrack from "b" to the correct location for "a" */
255
+		while (b != a && do_cmp(base + a, base + b, cmp_func, priv) >= 0)
256
+			b = parent(b, lsbit, size);
257
+		c = b;			/* Where "a" belongs */
258
+		while (b != a) {	/* Shift it into place */
259
+			b = parent(b, lsbit, size);
260
+			do_swap(base + b, base + c, size, swap_func);
102
261
 		}
103
262
 	}
104
263
 }
264
+EXPORT_SYMBOL(sort_r);
105
265
 
266
+void sort(void *base, size_t num, size_t size,
267
+	  cmp_func_t cmp_func,
268
+	  swap_func_t swap_func)
269
+{
270
+	return sort_r(base, num, size, _CMP_WRAPPER, swap_func, cmp_func);
271
+}
106
272
 EXPORT_SYMBOL(sort);