1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * Copyright (C) 1991,1992,1993,1997,1998,2003, 2005 Free Software Foundation, Inc.
4  * This file is part of the GNU C Library.
5  * Copyright (c) 2011 The Chromium OS Authors.
6  */
7 
8 /* From glibc-2.14, sysdeps/i386/memset.c */
9 
10 #include <linux/types.h>
11 #include <linux/compiler.h>
12 #include <asm/string.h>
13 
14 typedef uint32_t op_t;
15 
memset(void * dstpp,int c,size_t len)16 void *memset(void *dstpp, int c, size_t len)
17 {
18 	int d0;
19 	unsigned long int dstp = (unsigned long int) dstpp;
20 
21 	/* This explicit register allocation improves code very much indeed. */
22 	register op_t x asm("ax");
23 
24 	x = (unsigned char) c;
25 
26 	/* Clear the direction flag, so filling will move forward.  */
27 	asm volatile("cld");
28 
29 	/* This threshold value is optimal.  */
30 	if (len >= 12) {
31 		/* Fill X with four copies of the char we want to fill with. */
32 		x |= (x << 8);
33 		x |= (x << 16);
34 
35 		/* Adjust LEN for the bytes handled in the first loop.  */
36 		len -= (-dstp) % sizeof(op_t);
37 
38 		/*
39 		 * There are at least some bytes to set. No need to test for
40 		 * LEN == 0 in this alignment loop.
41 		 */
42 
43 		/* Fill bytes until DSTP is aligned on a longword boundary. */
44 		asm volatile(
45 			"rep\n"
46 			"stosb" /* %0, %2, %3 */ :
47 			"=D" (dstp), "=c" (d0) :
48 			"0" (dstp), "1" ((-dstp) % sizeof(op_t)), "a" (x) :
49 			"memory");
50 
51 		/* Fill longwords.  */
52 		asm volatile(
53 			"rep\n"
54 			"stosl" /* %0, %2, %3 */ :
55 			"=D" (dstp), "=c" (d0) :
56 			"0" (dstp), "1" (len / sizeof(op_t)), "a" (x) :
57 			"memory");
58 		len %= sizeof(op_t);
59 	}
60 
61 	/* Write the last few bytes. */
62 	asm volatile(
63 		"rep\n"
64 		"stosb" /* %0, %2, %3 */ :
65 		"=D" (dstp), "=c" (d0) :
66 		"0" (dstp), "1" (len), "a" (x) :
67 		"memory");
68 
69 	return dstpp;
70 }
71 
72 #define	OP_T_THRES	8
73 #define OPSIZ	(sizeof(op_t))
74 
75 #define BYTE_COPY_FWD(dst_bp, src_bp, nbytes)				  \
76 do {									  \
77 	int __d0;							  \
78 	asm volatile(							  \
79 		/* Clear the direction flag, so copying goes forward.  */ \
80 		"cld\n"							  \
81 		/* Copy bytes.  */					  \
82 		"rep\n"							  \
83 		"movsb" :						  \
84 		"=D" (dst_bp), "=S" (src_bp), "=c" (__d0) :		  \
85 		"0" (dst_bp), "1" (src_bp), "2" (nbytes) :		  \
86 		"memory");						  \
87 } while (0)
88 
89 #define WORD_COPY_FWD(dst_bp, src_bp, nbytes_left, nbytes)		  \
90 do {									  \
91 	int __d0;							  \
92 	asm volatile(							  \
93 		/* Clear the direction flag, so copying goes forward.  */ \
94 		"cld\n"							  \
95 		/* Copy longwords.  */					  \
96 		"rep\n"							  \
97 		"movsl" :						  \
98 		"=D" (dst_bp), "=S" (src_bp), "=c" (__d0) :		  \
99 		"0" (dst_bp), "1" (src_bp), "2" ((nbytes) / 4) :	  \
100 		"memory");						  \
101 	(nbytes_left) = (nbytes) % 4;					  \
102 } while (0)
103 
memcpy(void * dstpp,const void * srcpp,size_t len)104 void *memcpy(void *dstpp, const void *srcpp, size_t len)
105 {
106 	unsigned long int dstp = (long int)dstpp;
107 	unsigned long int srcp = (long int)srcpp;
108 
109 	/* Copy from the beginning to the end.  */
110 
111 	/* If there not too few bytes to copy, use word copy.  */
112 	if (len >= OP_T_THRES) {
113 		/* Copy just a few bytes to make DSTP aligned.  */
114 		len -= (-dstp) % OPSIZ;
115 		BYTE_COPY_FWD(dstp, srcp, (-dstp) % OPSIZ);
116 
117 		/* Copy from SRCP to DSTP taking advantage of the known
118 		 * alignment of DSTP.  Number of bytes remaining is put
119 		 * in the third argument, i.e. in LEN.  This number may
120 		 * vary from machine to machine.
121 		 */
122 		WORD_COPY_FWD(dstp, srcp, len, len);
123 
124 		/* Fall out and copy the tail.  */
125 	}
126 
127 	/* There are just a few bytes to copy. Use byte memory operations. */
128 	BYTE_COPY_FWD(dstp, srcp, len);
129 
130 	return dstpp;
131 }
132 
memmove(void * dest,const void * src,size_t n)133 void *memmove(void *dest, const void *src, size_t n)
134 {
135 	int d0, d1, d2, d3, d4, d5;
136 	char *ret = dest;
137 
138 	__asm__ __volatile__(
139 		/* Handle more 16 bytes in loop */
140 		"cmp $0x10, %0\n\t"
141 		"jb	1f\n\t"
142 
143 		/* Decide forward/backward copy mode */
144 		"cmp %2, %1\n\t"
145 		"jb	2f\n\t"
146 
147 		/*
148 		 * movs instruction have many startup latency
149 		 * so we handle small size by general register.
150 		 */
151 		"cmp  $680, %0\n\t"
152 		"jb 3f\n\t"
153 		/* movs instruction is only good for aligned case */
154 		"mov %1, %3\n\t"
155 		"xor %2, %3\n\t"
156 		"and $0xff, %3\n\t"
157 		"jz 4f\n\t"
158 		"3:\n\t"
159 		"sub $0x10, %0\n\t"
160 
161 		/* We gobble 16 bytes forward in each loop */
162 		"3:\n\t"
163 		"sub $0x10, %0\n\t"
164 		"mov 0*4(%1), %3\n\t"
165 		"mov 1*4(%1), %4\n\t"
166 		"mov  %3, 0*4(%2)\n\t"
167 		"mov  %4, 1*4(%2)\n\t"
168 		"mov 2*4(%1), %3\n\t"
169 		"mov 3*4(%1), %4\n\t"
170 		"mov  %3, 2*4(%2)\n\t"
171 		"mov  %4, 3*4(%2)\n\t"
172 		"lea  0x10(%1), %1\n\t"
173 		"lea  0x10(%2), %2\n\t"
174 		"jae 3b\n\t"
175 		"add $0x10, %0\n\t"
176 		"jmp 1f\n\t"
177 
178 		/* Handle data forward by movs */
179 		".p2align 4\n\t"
180 		"4:\n\t"
181 		"mov -4(%1, %0), %3\n\t"
182 		"lea -4(%2, %0), %4\n\t"
183 		"shr $2, %0\n\t"
184 		"rep movsl\n\t"
185 		"mov %3, (%4)\n\t"
186 		"jmp 11f\n\t"
187 		/* Handle data backward by movs */
188 		".p2align 4\n\t"
189 		"6:\n\t"
190 		"mov (%1), %3\n\t"
191 		"mov %2, %4\n\t"
192 		"lea -4(%1, %0), %1\n\t"
193 		"lea -4(%2, %0), %2\n\t"
194 		"shr $2, %0\n\t"
195 		"std\n\t"
196 		"rep movsl\n\t"
197 		"mov %3,(%4)\n\t"
198 		"cld\n\t"
199 		"jmp 11f\n\t"
200 
201 		/* Start to prepare for backward copy */
202 		".p2align 4\n\t"
203 		"2:\n\t"
204 		"cmp  $680, %0\n\t"
205 		"jb 5f\n\t"
206 		"mov %1, %3\n\t"
207 		"xor %2, %3\n\t"
208 		"and $0xff, %3\n\t"
209 		"jz 6b\n\t"
210 
211 		/* Calculate copy position to tail */
212 		"5:\n\t"
213 		"add %0, %1\n\t"
214 		"add %0, %2\n\t"
215 		"sub $0x10, %0\n\t"
216 
217 		/* We gobble 16 bytes backward in each loop */
218 		"7:\n\t"
219 		"sub $0x10, %0\n\t"
220 
221 		"mov -1*4(%1), %3\n\t"
222 		"mov -2*4(%1), %4\n\t"
223 		"mov  %3, -1*4(%2)\n\t"
224 		"mov  %4, -2*4(%2)\n\t"
225 		"mov -3*4(%1), %3\n\t"
226 		"mov -4*4(%1), %4\n\t"
227 		"mov  %3, -3*4(%2)\n\t"
228 		"mov  %4, -4*4(%2)\n\t"
229 		"lea  -0x10(%1), %1\n\t"
230 		"lea  -0x10(%2), %2\n\t"
231 		"jae 7b\n\t"
232 		/* Calculate copy position to head */
233 		"add $0x10, %0\n\t"
234 		"sub %0, %1\n\t"
235 		"sub %0, %2\n\t"
236 
237 		/* Move data from 8 bytes to 15 bytes */
238 		".p2align 4\n\t"
239 		"1:\n\t"
240 		"cmp $8, %0\n\t"
241 		"jb 8f\n\t"
242 		"mov 0*4(%1), %3\n\t"
243 		"mov 1*4(%1), %4\n\t"
244 		"mov -2*4(%1, %0), %5\n\t"
245 		"mov -1*4(%1, %0), %1\n\t"
246 
247 		"mov  %3, 0*4(%2)\n\t"
248 		"mov  %4, 1*4(%2)\n\t"
249 		"mov  %5, -2*4(%2, %0)\n\t"
250 		"mov  %1, -1*4(%2, %0)\n\t"
251 		"jmp 11f\n\t"
252 
253 		/* Move data from 4 bytes to 7 bytes */
254 		".p2align 4\n\t"
255 		"8:\n\t"
256 		"cmp $4, %0\n\t"
257 		"jb 9f\n\t"
258 		"mov 0*4(%1), %3\n\t"
259 		"mov -1*4(%1, %0), %4\n\t"
260 		"mov  %3, 0*4(%2)\n\t"
261 		"mov  %4, -1*4(%2, %0)\n\t"
262 		"jmp 11f\n\t"
263 
264 		/* Move data from 2 bytes to 3 bytes */
265 		".p2align 4\n\t"
266 		"9:\n\t"
267 		"cmp $2, %0\n\t"
268 		"jb 10f\n\t"
269 		"movw 0*2(%1), %%dx\n\t"
270 		"movw -1*2(%1, %0), %%bx\n\t"
271 		"movw %%dx, 0*2(%2)\n\t"
272 		"movw %%bx, -1*2(%2, %0)\n\t"
273 		"jmp 11f\n\t"
274 
275 		/* Move data for 1 byte */
276 		".p2align 4\n\t"
277 		"10:\n\t"
278 		"cmp $1, %0\n\t"
279 		"jb 11f\n\t"
280 		"movb (%1), %%cl\n\t"
281 		"movb %%cl, (%2)\n\t"
282 		".p2align 4\n\t"
283 		"11:"
284 		: "=&c" (d0), "=&S" (d1), "=&D" (d2),
285 		  "=r" (d3), "=r" (d4), "=r"(d5)
286 		: "0" (n),
287 		 "1" (src),
288 		 "2" (dest)
289 		: "memory");
290 
291 	return ret;
292 }
293