2 * Copyright 2011 Tilera Corporation. All Rights Reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation, version 2.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
11 * NON INFRINGEMENT. See the GNU General Public License for
15 #include <linux/types.h>
16 #include <linux/string.h>
17 #include <linux/module.h>
18 #define __memcpy memcpy
19 /* EXPORT_SYMBOL() is in arch/tile/lib/exports.c since this should be asm. */
21 /* Must be 8 bytes in size. */
22 #define word_t uint64_t
24 #if CHIP_L2_LINE_SIZE() != 64 && CHIP_L2_LINE_SIZE() != 128
25 #error "Assumes 64 or 128 byte line size"
28 /* How many cache lines ahead should we prefetch? */
29 #define PREFETCH_LINES_AHEAD 3
32 * Provide "base versions" of load and store for the normal code path.
33 * The kernel provides other versions for userspace copies.
35 #define ST(p, v) (*(p) = (v))
48 void *memcpy(void *__restrict dstv, const void *__restrict srcv, size_t n)
51 * Special kernel version will provide implementation of the LDn/STn
52 * macros to return a count of uncopied bytes due to mm fault.
55 int USERCOPY_FUNC(void *__restrict dstv, const void *__restrict srcv, size_t n)
58 char *__restrict dst1 = (char *)dstv;
59 const char *__restrict src1 = (const char *)srcv;
60 const char *__restrict src1_end;
61 const char *__restrict prefetch;
62 word_t *__restrict dst8; /* 8-byte pointer to destination memory. */
63 word_t final; /* Final bytes to write to trailing word, if any */
68 ST1(dst1++, LD1(src1++));
73 * Locate the end of source memory we will copy. Don't
76 src1_end = src1 + n - 1;
78 /* Prefetch ahead a few cache lines, but not past the end. */
80 for (i = 0; i < PREFETCH_LINES_AHEAD; i++) {
81 __insn_prefetch(prefetch);
82 prefetch += CHIP_L2_LINE_SIZE();
83 prefetch = (prefetch > src1_end) ? prefetch : src1;
86 /* Copy bytes until dst is word-aligned. */
87 for (; (uintptr_t)dst1 & (sizeof(word_t) - 1); n--)
88 ST1(dst1++, LD1(src1++));
90 /* 8-byte pointer to destination memory. */
91 dst8 = (word_t *)dst1;
93 if (__builtin_expect((uintptr_t)src1 & (sizeof(word_t) - 1), 0)) {
95 * Misaligned copy. Copy 8 bytes at a time, but don't
96 * bother with other fanciness.
98 * TODO: Consider prefetching and using wh64 as well.
101 /* Create an aligned src8. */
102 const word_t *__restrict src8 =
103 (const word_t *)((uintptr_t)src1 & -sizeof(word_t));
106 word_t a = LD8(src8++);
107 for (; n >= sizeof(word_t); n -= sizeof(word_t)) {
109 a = __insn_dblalign(a, b, src1);
117 b = ((const char *)src8 <= src1_end) ? *src8 : 0;
120 * Final source bytes to write to trailing partial
123 final = __insn_dblalign(a, b, src1);
127 const word_t* __restrict src8 = (const word_t *)src1;
129 /* src8 and dst8 are both word-aligned. */
130 if (n >= CHIP_L2_LINE_SIZE()) {
131 /* Copy until 'dst' is cache-line-aligned. */
132 for (; (uintptr_t)dst8 & (CHIP_L2_LINE_SIZE() - 1);
134 ST8(dst8++, LD8(src8++));
136 for (; n >= CHIP_L2_LINE_SIZE(); ) {
140 * Prefetch and advance to next line
141 * to prefetch, but don't go past the end
143 __insn_prefetch(prefetch);
144 prefetch += CHIP_L2_LINE_SIZE();
145 prefetch = (prefetch > src1_end) ? prefetch :
149 * Copy an entire cache line. Manually
150 * unrolled to avoid idiosyncracies of
151 * compiler unrolling.
153 #define COPY_WORD(offset) ({ ST8(dst8+offset, LD8(src8+offset)); n -= 8; })
162 #if CHIP_L2_LINE_SIZE() == 128
171 #elif CHIP_L2_LINE_SIZE() != 64
172 # error Fix code that assumes particular L2 cache line sizes
175 dst8 += CHIP_L2_LINE_SIZE() / sizeof(word_t);
176 src8 += CHIP_L2_LINE_SIZE() / sizeof(word_t);
180 for (; n >= sizeof(word_t); n -= sizeof(word_t))
181 ST8(dst8++, LD8(src8++));
183 if (__builtin_expect(n == 0, 1))
189 /* n != 0 if we get here. Write out any trailing bytes. */
192 ST4((uint32_t *)dst1, final);
198 ST2((uint16_t *)dst1, final);
204 ST1((uint8_t *)dst1, final);