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 /* EXPORT_SYMBOL() is in arch/tile/lib/exports.c since this should be asm. */
20 /* Must be 8 bytes in size. */
21 #define word_t uint64_t
23 #if CHIP_L2_LINE_SIZE() != 64 && CHIP_L2_LINE_SIZE() != 128
24 #error "Assumes 64 or 128 byte line size"
27 /* How many cache lines ahead should we prefetch? */
28 #define PREFETCH_LINES_AHEAD 3
31 * Provide "base versions" of load and store for the normal code path.
32 * The kernel provides other versions for userspace copies.
34 #define ST(p, v) (*(p) = (v))
47 void *memcpy(void *__restrict dstv
, const void *__restrict srcv
, size_t n
)
50 * Special kernel version will provide implementation of the LDn/STn
51 * macros to return a count of uncopied bytes due to mm fault.
54 int USERCOPY_FUNC(void *__restrict dstv
, const void *__restrict srcv
, size_t n
)
57 char *__restrict dst1
= (char *)dstv
;
58 const char *__restrict src1
= (const char *)srcv
;
59 const char *__restrict src1_end
;
60 const char *__restrict prefetch
;
61 word_t
*__restrict dst8
; /* 8-byte pointer to destination memory. */
62 word_t final
; /* Final bytes to write to trailing word, if any */
67 ST1(dst1
++, LD1(src1
++));
72 * Locate the end of source memory we will copy. Don't
75 src1_end
= src1
+ n
- 1;
77 /* Prefetch ahead a few cache lines, but not past the end. */
79 for (i
= 0; i
< PREFETCH_LINES_AHEAD
; i
++) {
80 __insn_prefetch(prefetch
);
81 prefetch
+= CHIP_L2_LINE_SIZE();
82 prefetch
= (prefetch
> src1_end
) ? prefetch
: src1
;
85 /* Copy bytes until dst is word-aligned. */
86 for (; (uintptr_t)dst1
& (sizeof(word_t
) - 1); n
--)
87 ST1(dst1
++, LD1(src1
++));
89 /* 8-byte pointer to destination memory. */
90 dst8
= (word_t
*)dst1
;
92 if (__builtin_expect((uintptr_t)src1
& (sizeof(word_t
) - 1), 0)) {
94 * Misaligned copy. Copy 8 bytes at a time, but don't
95 * bother with other fanciness.
97 * TODO: Consider prefetching and using wh64 as well.
100 /* Create an aligned src8. */
101 const word_t
*__restrict src8
=
102 (const word_t
*)((uintptr_t)src1
& -sizeof(word_t
));
105 word_t a
= LD8(src8
++);
106 for (; n
>= sizeof(word_t
); n
-= sizeof(word_t
)) {
108 a
= __insn_dblalign(a
, b
, src1
);
116 b
= ((const char *)src8
<= src1_end
) ? *src8
: 0;
119 * Final source bytes to write to trailing partial
122 final
= __insn_dblalign(a
, b
, src1
);
126 const word_t
* __restrict src8
= (const word_t
*)src1
;
128 /* src8 and dst8 are both word-aligned. */
129 if (n
>= CHIP_L2_LINE_SIZE()) {
130 /* Copy until 'dst' is cache-line-aligned. */
131 for (; (uintptr_t)dst8
& (CHIP_L2_LINE_SIZE() - 1);
133 ST8(dst8
++, LD8(src8
++));
135 for (; n
>= CHIP_L2_LINE_SIZE(); ) {
139 * Prefetch and advance to next line
140 * to prefetch, but don't go past the end
142 __insn_prefetch(prefetch
);
143 prefetch
+= CHIP_L2_LINE_SIZE();
144 prefetch
= (prefetch
> src1_end
) ? prefetch
:
148 * Copy an entire cache line. Manually
149 * unrolled to avoid idiosyncracies of
150 * compiler unrolling.
152 #define COPY_WORD(offset) ({ ST8(dst8+offset, LD8(src8+offset)); n -= 8; })
161 #if CHIP_L2_LINE_SIZE() == 128
170 #elif CHIP_L2_LINE_SIZE() != 64
171 # error Fix code that assumes particular L2 cache line sizes
174 dst8
+= CHIP_L2_LINE_SIZE() / sizeof(word_t
);
175 src8
+= CHIP_L2_LINE_SIZE() / sizeof(word_t
);
179 for (; n
>= sizeof(word_t
); n
-= sizeof(word_t
))
180 ST8(dst8
++, LD8(src8
++));
182 if (__builtin_expect(n
== 0, 1))
188 /* n != 0 if we get here. Write out any trailing bytes. */
190 #ifndef __BIG_ENDIAN__
192 ST4((uint32_t *)dst1
, final
);
198 ST2((uint16_t *)dst1
, final
);
204 ST1((uint8_t *)dst1
, final
);
207 ST4((uint32_t *)dst1
, final
>> 32);
215 ST2((uint16_t *)dst1
, final
>> 16);
223 ST1((uint8_t *)dst1
, final
>> 8);