1 /* SPDX-License-Identifier: GPL-2.0 */ !! 1 /*
2 /* linux/arch/sparc/lib/memset.S: Sparc optimi !! 2 * This file is subject to the terms and conditions of the GNU General Public
3 * Copyright (C) 1991,1996 Free Software Found !! 3 * License. See the file "COPYING" in the main directory of this archive
4 * Copyright (C) 1996,1997 Jakub Jelinek (jj@s !! 4 * for more details.
5 * Copyright (C) 1996 David S. Miller (davem@c <<
6 * 5 *
7 * Calls to memset returns initial %o0. Calls !! 6 * Copyright (C) 1998, 1999, 2000 by Ralf Baechle
8 * number of bytes not yet set if exception oc !! 7 * Copyright (C) 1999, 2000 Silicon Graphics, Inc.
9 * clear_user. !! 8 * Copyright (C) 2007 by Maciej W. Rozycki
>> 9 * Copyright (C) 2011, 2012 MIPS Technologies, Inc.
10 */ 10 */
>> 11 #include <asm/asm.h>
>> 12 #include <asm/asm-offsets.h>
>> 13 #include <asm/regdef.h>
>> 14
>> 15 #if LONGSIZE == 4
>> 16 #define LONG_S_L swl
>> 17 #define LONG_S_R swr
>> 18 #else
>> 19 #define LONG_S_L sdl
>> 20 #define LONG_S_R sdr
>> 21 #endif
>> 22
>> 23 #ifdef CONFIG_CPU_MICROMIPS
>> 24 #define STORSIZE (LONGSIZE * 2)
>> 25 #define STORMASK (STORSIZE - 1)
>> 26 #define FILL64RG t8
>> 27 #define FILLPTRG t7
>> 28 #undef LONG_S
>> 29 #define LONG_S LONG_SP
>> 30 #else
>> 31 #define STORSIZE LONGSIZE
>> 32 #define STORMASK LONGMASK
>> 33 #define FILL64RG a1
>> 34 #define FILLPTRG t0
>> 35 #endif
>> 36
>> 37 #define LEGACY_MODE 1
>> 38 #define EVA_MODE 2
>> 39
>> 40 /*
>> 41 * No need to protect it with EVA #ifdefery. The generated block of code
>> 42 * will never be assembled if EVA is not enabled.
>> 43 */
>> 44 #define __EVAFY(insn, reg, addr) __BUILD_EVA_INSN(insn##e, reg, addr)
>> 45 #define ___BUILD_EVA_INSN(insn, reg, addr) __EVAFY(insn, reg, addr)
11 46
12 #include <linux/export.h> !! 47 #define EX(insn,reg,addr,handler) \
13 #include <asm/ptrace.h> !! 48 .if \mode == LEGACY_MODE; \
>> 49 9: insn reg, addr; \
>> 50 .else; \
>> 51 9: ___BUILD_EVA_INSN(insn, reg, addr); \
>> 52 .endif; \
>> 53 .section __ex_table,"a"; \
>> 54 PTR 9b, handler; \
>> 55 .previous
>> 56
>> 57 .macro f_fill64 dst, offset, val, fixup, mode
>> 58 EX(LONG_S, \val, (\offset + 0 * STORSIZE)(\dst), \fixup)
>> 59 EX(LONG_S, \val, (\offset + 1 * STORSIZE)(\dst), \fixup)
>> 60 EX(LONG_S, \val, (\offset + 2 * STORSIZE)(\dst), \fixup)
>> 61 EX(LONG_S, \val, (\offset + 3 * STORSIZE)(\dst), \fixup)
>> 62 #if ((defined(CONFIG_CPU_MICROMIPS) && (LONGSIZE == 4)) || !defined(CONFIG_CPU_MICROMIPS))
>> 63 EX(LONG_S, \val, (\offset + 4 * STORSIZE)(\dst), \fixup)
>> 64 EX(LONG_S, \val, (\offset + 5 * STORSIZE)(\dst), \fixup)
>> 65 EX(LONG_S, \val, (\offset + 6 * STORSIZE)(\dst), \fixup)
>> 66 EX(LONG_S, \val, (\offset + 7 * STORSIZE)(\dst), \fixup)
>> 67 #endif
>> 68 #if (!defined(CONFIG_CPU_MICROMIPS) && (LONGSIZE == 4))
>> 69 EX(LONG_S, \val, (\offset + 8 * STORSIZE)(\dst), \fixup)
>> 70 EX(LONG_S, \val, (\offset + 9 * STORSIZE)(\dst), \fixup)
>> 71 EX(LONG_S, \val, (\offset + 10 * STORSIZE)(\dst), \fixup)
>> 72 EX(LONG_S, \val, (\offset + 11 * STORSIZE)(\dst), \fixup)
>> 73 EX(LONG_S, \val, (\offset + 12 * STORSIZE)(\dst), \fixup)
>> 74 EX(LONG_S, \val, (\offset + 13 * STORSIZE)(\dst), \fixup)
>> 75 EX(LONG_S, \val, (\offset + 14 * STORSIZE)(\dst), \fixup)
>> 76 EX(LONG_S, \val, (\offset + 15 * STORSIZE)(\dst), \fixup)
>> 77 #endif
>> 78 .endm
>> 79
>> 80 .set noreorder
>> 81 .align 5
>> 82
>> 83 /*
>> 84 * Macro to generate the __bzero{,_user} symbol
>> 85 * Arguments:
>> 86 * mode: LEGACY_MODE or EVA_MODE
>> 87 */
>> 88 .macro __BUILD_BZERO mode
>> 89 /* Initialize __memset if this is the first time we call this macro */
>> 90 .ifnotdef __memset
>> 91 .set __memset, 1
>> 92 .hidden __memset /* Make sure it does not leak */
>> 93 .endif
>> 94
>> 95 sltiu t0, a2, STORSIZE /* very small region? */
>> 96 bnez t0, .Lsmall_memset\@
>> 97 andi t0, a0, STORMASK /* aligned? */
>> 98
>> 99 #ifdef CONFIG_CPU_MICROMIPS
>> 100 move t8, a1 /* used by 'swp' instruction */
>> 101 move t9, a1
>> 102 #endif
>> 103 #ifndef CONFIG_CPU_DADDI_WORKAROUNDS
>> 104 beqz t0, 1f
>> 105 PTR_SUBU t0, STORSIZE /* alignment in bytes */
>> 106 #else
>> 107 .set noat
>> 108 li AT, STORSIZE
>> 109 beqz t0, 1f
>> 110 PTR_SUBU t0, AT /* alignment in bytes */
>> 111 .set at
>> 112 #endif
>> 113
>> 114 #ifndef CONFIG_CPU_MIPSR6
>> 115 R10KCBARRIER(0(ra))
>> 116 #ifdef __MIPSEB__
>> 117 EX(LONG_S_L, a1, (a0), .Lfirst_fixup\@) /* make word/dword aligned */
>> 118 #else
>> 119 EX(LONG_S_R, a1, (a0), .Lfirst_fixup\@) /* make word/dword aligned */
>> 120 #endif
>> 121 PTR_SUBU a0, t0 /* long align ptr */
>> 122 PTR_ADDU a2, t0 /* correct size */
>> 123
>> 124 #else /* CONFIG_CPU_MIPSR6 */
>> 125 #define STORE_BYTE(N) \
>> 126 EX(sb, a1, N(a0), .Lbyte_fixup\@); \
>> 127 beqz t0, 0f; \
>> 128 PTR_ADDU t0, 1;
>> 129
>> 130 PTR_ADDU a2, t0 /* correct size */
>> 131 PTR_ADDU t0, 1
>> 132 STORE_BYTE(0)
>> 133 STORE_BYTE(1)
>> 134 #if LONGSIZE == 4
>> 135 EX(sb, a1, 2(a0), .Lbyte_fixup\@)
>> 136 #else
>> 137 STORE_BYTE(2)
>> 138 STORE_BYTE(3)
>> 139 STORE_BYTE(4)
>> 140 STORE_BYTE(5)
>> 141 EX(sb, a1, 6(a0), .Lbyte_fixup\@)
>> 142 #endif
>> 143 0:
>> 144 ori a0, STORMASK
>> 145 xori a0, STORMASK
>> 146 PTR_ADDIU a0, STORSIZE
>> 147 #endif /* CONFIG_CPU_MIPSR6 */
>> 148 1: ori t1, a2, 0x3f /* # of full blocks */
>> 149 xori t1, 0x3f
>> 150 beqz t1, .Lmemset_partial\@ /* no block to fill */
>> 151 andi t0, a2, 0x40-STORSIZE
>> 152
>> 153 PTR_ADDU t1, a0 /* end address */
>> 154 .set reorder
>> 155 1: PTR_ADDIU a0, 64
>> 156 R10KCBARRIER(0(ra))
>> 157 f_fill64 a0, -64, FILL64RG, .Lfwd_fixup\@, \mode
>> 158 bne t1, a0, 1b
>> 159 .set noreorder
>> 160
>> 161 .Lmemset_partial\@:
>> 162 R10KCBARRIER(0(ra))
>> 163 PTR_LA t1, 2f /* where to start */
>> 164 #ifdef CONFIG_CPU_MICROMIPS
>> 165 LONG_SRL t7, t0, 1
>> 166 #endif
>> 167 #if LONGSIZE == 4
>> 168 PTR_SUBU t1, FILLPTRG
>> 169 #else
>> 170 .set noat
>> 171 LONG_SRL AT, FILLPTRG, 1
>> 172 PTR_SUBU t1, AT
>> 173 .set at
>> 174 #endif
>> 175 jr t1
>> 176 PTR_ADDU a0, t0 /* dest ptr */
>> 177
>> 178 .set push
>> 179 .set noreorder
>> 180 .set nomacro
>> 181 /* ... but first do longs ... */
>> 182 f_fill64 a0, -64, FILL64RG, .Lpartial_fixup\@, \mode
>> 183 2: .set pop
>> 184 andi a2, STORMASK /* At most one long to go */
>> 185
>> 186 beqz a2, 1f
>> 187 #ifndef CONFIG_CPU_MIPSR6
>> 188 PTR_ADDU a0, a2 /* What's left */
>> 189 R10KCBARRIER(0(ra))
>> 190 #ifdef __MIPSEB__
>> 191 EX(LONG_S_R, a1, -1(a0), .Llast_fixup\@)
>> 192 #else
>> 193 EX(LONG_S_L, a1, -1(a0), .Llast_fixup\@)
>> 194 #endif
>> 195 #else
>> 196 PTR_SUBU t0, $0, a2
>> 197 PTR_ADDIU t0, 1
>> 198 STORE_BYTE(0)
>> 199 STORE_BYTE(1)
>> 200 #if LONGSIZE == 4
>> 201 EX(sb, a1, 2(a0), .Lbyte_fixup\@)
>> 202 #else
>> 203 STORE_BYTE(2)
>> 204 STORE_BYTE(3)
>> 205 STORE_BYTE(4)
>> 206 STORE_BYTE(5)
>> 207 EX(sb, a1, 6(a0), .Lbyte_fixup\@)
>> 208 #endif
>> 209 0:
>> 210 #endif
>> 211 1: jr ra
>> 212 move a2, zero
>> 213
>> 214 .Lsmall_memset\@:
>> 215 beqz a2, 2f
>> 216 PTR_ADDU t1, a0, a2
>> 217
>> 218 1: PTR_ADDIU a0, 1 /* fill bytewise */
>> 219 R10KCBARRIER(0(ra))
>> 220 bne t1, a0, 1b
>> 221 EX(sb, a1, -1(a0), .Lsmall_fixup\@)
>> 222
>> 223 2: jr ra /* done */
>> 224 move a2, zero
>> 225 .if __memset == 1
>> 226 END(memset)
>> 227 .set __memset, 0
>> 228 .hidden __memset
>> 229 .endif
>> 230
>> 231 #ifdef CONFIG_CPU_MIPSR6
>> 232 .Lbyte_fixup\@:
>> 233 PTR_SUBU a2, $0, t0
>> 234 jr ra
>> 235 PTR_ADDIU a2, 1
>> 236 #endif /* CONFIG_CPU_MIPSR6 */
>> 237
>> 238 .Lfirst_fixup\@:
>> 239 jr ra
>> 240 nop
>> 241
>> 242 .Lfwd_fixup\@:
>> 243 PTR_L t0, TI_TASK($28)
>> 244 andi a2, 0x3f
>> 245 LONG_L t0, THREAD_BUADDR(t0)
>> 246 LONG_ADDU a2, t1
>> 247 jr ra
>> 248 LONG_SUBU a2, t0
>> 249
>> 250 .Lpartial_fixup\@:
>> 251 PTR_L t0, TI_TASK($28)
>> 252 andi a2, STORMASK
>> 253 LONG_L t0, THREAD_BUADDR(t0)
>> 254 LONG_ADDU a2, a0
>> 255 jr ra
>> 256 LONG_SUBU a2, t0
14 257
15 /* Work around cpp -rob */ !! 258 .Llast_fixup\@:
16 #define ALLOC #alloc !! 259 jr ra
17 #define EXECINSTR #execinstr !! 260 nop
18 #define EX(x,y,a,b) <<
19 98: x,y; <<
20 .section .fixup,ALLOC,EXECINSTR; <<
21 .align 4; <<
22 99: retl; <<
23 a, b, %o0; <<
24 .section __ex_table,ALLOC; <<
25 .align 4; <<
26 .word 98b, 99b; <<
27 .text; <<
28 .align 4 <<
29 <<
30 #define STORE(source, base, offset, n) <<
31 98: std source, [base + offset + n]; <<
32 .section .fixup,ALLOC,EXECINSTR; <<
33 .align 4; <<
34 99: ba 30f; <<
35 sub %o3, n - offset, %o3; <<
36 .section __ex_table,ALLOC; <<
37 .align 4; <<
38 .word 98b, 99b; <<
39 .text; <<
40 .align 4; <<
41 <<
42 #define STORE_LAST(source, base, offset, n) <<
43 EX(std source, [base - offset - n], <<
44 add %o1, offset + n); <<
45 <<
46 /* Please don't change these macros, unless yo <<
47 * in the .fixup section below as well. <<
48 * Store 64 bytes at (BASE + OFFSET) using val <<
49 #define ZERO_BIG_BLOCK(base, offset, source) <<
50 STORE(source, base, offset, 0x00); <<
51 STORE(source, base, offset, 0x08); <<
52 STORE(source, base, offset, 0x10); <<
53 STORE(source, base, offset, 0x18); <<
54 STORE(source, base, offset, 0x20); <<
55 STORE(source, base, offset, 0x28); <<
56 STORE(source, base, offset, 0x30); <<
57 STORE(source, base, offset, 0x38); <<
58 <<
59 #define ZERO_LAST_BLOCKS(base, offset, source) <<
60 STORE_LAST(source, base, offset, 0x38) <<
61 STORE_LAST(source, base, offset, 0x30) <<
62 STORE_LAST(source, base, offset, 0x28) <<
63 STORE_LAST(source, base, offset, 0x20) <<
64 STORE_LAST(source, base, offset, 0x18) <<
65 STORE_LAST(source, base, offset, 0x10) <<
66 STORE_LAST(source, base, offset, 0x08) <<
67 STORE_LAST(source, base, offset, 0x00) <<
68 <<
69 .text <<
70 .align 4 <<
71 <<
72 .globl __bzero_begin <<
73 __bzero_begin: <<
74 <<
75 .globl __bzero <<
76 .type __bzero,#function <<
77 .globl memset <<
78 EXPORT_SYMBOL(__bzero) <<
79 EXPORT_SYMBOL(memset) <<
80 memset: <<
81 mov %o0, %g1 <<
82 mov 1, %g4 <<
83 and %o1, 0xff, %g3 <<
84 sll %g3, 8, %g2 <<
85 or %g3, %g2, %g3 <<
86 sll %g3, 16, %g2 <<
87 or %g3, %g2, %g3 <<
88 b 1f <<
89 mov %o2, %o1 <<
90 3: <<
91 cmp %o2, 3 <<
92 be 2f <<
93 EX(stb %g3, [%o0], sub %o1, 0) <<
94 <<
95 cmp %o2, 2 <<
96 be 2f <<
97 EX(stb %g3, [%o0 + 0x01], sub %o1, 1) <<
98 <<
99 EX(stb %g3, [%o0 + 0x02], sub %o1, 2) <<
100 2: <<
101 sub %o2, 4, %o2 <<
102 add %o1, %o2, %o1 <<
103 b 4f <<
104 sub %o0, %o2, %o0 <<
105 <<
106 __bzero: <<
107 clr %g4 <<
108 mov %g0, %g3 <<
109 1: <<
110 cmp %o1, 7 <<
111 bleu 7f <<
112 andcc %o0, 3, %o2 <<
113 <<
114 bne 3b <<
115 4: <<
116 andcc %o0, 4, %g0 <<
117 <<
118 be 2f <<
119 mov %g3, %g2 <<
120 <<
121 EX(st %g3, [%o0], sub %o1, 0) <<
122 sub %o1, 4, %o1 <<
123 add %o0, 4, %o0 <<
124 2: <<
125 andcc %o1, 0xffffff80, %o3 ! Now <<
126 be 9f <<
127 andcc %o1, 0x78, %o2 <<
128 10: <<
129 ZERO_BIG_BLOCK(%o0, 0x00, %g2) <<
130 subcc %o3, 128, %o3 <<
131 ZERO_BIG_BLOCK(%o0, 0x40, %g2) <<
132 bne 10b <<
133 add %o0, 128, %o0 <<
134 <<
135 orcc %o2, %g0, %g0 <<
136 9: <<
137 be 13f <<
138 andcc %o1, 7, %o1 <<
139 <<
140 srl %o2, 1, %o3 <<
141 set 13f, %o4 <<
142 sub %o4, %o3, %o4 <<
143 jmp %o4 <<
144 add %o0, %o2, %o0 <<
145 <<
146 ZERO_LAST_BLOCKS(%o0, 0x48, %g2) <<
147 ZERO_LAST_BLOCKS(%o0, 0x08, %g2) <<
148 13: <<
149 be 8f <<
150 andcc %o1, 4, %g0 <<
151 261
152 be 1f !! 262 .Lsmall_fixup\@:
153 andcc %o1, 2, %g0 !! 263 PTR_SUBU a2, t1, a0
>> 264 jr ra
>> 265 PTR_ADDIU a2, 1
154 266
155 EX(st %g3, [%o0], and %o1, 7) !! 267 .endm
156 add %o0, 4, %o0 <<
157 1: <<
158 be 1f <<
159 andcc %o1, 1, %g0 <<
160 268
161 EX(sth %g3, [%o0], and %o1, 3) !! 269 /*
162 add %o0, 2, %o0 !! 270 * memset(void *s, int c, size_t n)
163 1: !! 271 *
164 bne,a 8f !! 272 * a0: start of area to clear
165 EX(stb %g3, [%o0], and %o1, 1) !! 273 * a1: char to fill with
166 8: !! 274 * a2: size of area to clear
167 b 0f !! 275 */
168 nop <<
169 7: <<
170 be 13b <<
171 orcc %o1, 0, %g0 <<
172 <<
173 be 0f <<
174 8: <<
175 add %o0, 1, %o0 <<
176 subcc %o1, 1, %o1 <<
177 bne 8b <<
178 EX(stb %g3, [%o0 - 1], add %o1, 1) <<
179 0: <<
180 andcc %g4, 1, %g0 <<
181 be 5f <<
182 nop <<
183 retl <<
184 mov %g1, %o0 <<
185 5: <<
186 retl <<
187 clr %o0 <<
188 <<
189 .section .fixup,#alloc,#execinstr <<
190 .align 4 <<
191 30: <<
192 and %o1, 0x7f, %o1 <<
193 retl <<
194 add %o3, %o1, %o0 <<
195 276
196 .globl __bzero_end !! 277 LEAF(memset)
197 __bzero_end: !! 278 beqz a1, 1f
>> 279 move v0, a0 /* result */
>> 280
>> 281 andi a1, 0xff /* spread fillword */
>> 282 LONG_SLL t1, a1, 8
>> 283 or a1, t1
>> 284 LONG_SLL t1, a1, 16
>> 285 #if LONGSIZE == 8
>> 286 or a1, t1
>> 287 LONG_SLL t1, a1, 32
>> 288 #endif
>> 289 or a1, t1
>> 290 1:
>> 291 #ifndef CONFIG_EVA
>> 292 FEXPORT(__bzero)
>> 293 #else
>> 294 FEXPORT(__bzero_kernel)
>> 295 #endif
>> 296 __BUILD_BZERO LEGACY_MODE
>> 297
>> 298 #ifdef CONFIG_EVA
>> 299 LEAF(__bzero)
>> 300 __BUILD_BZERO EVA_MODE
>> 301 END(__bzero)
>> 302 #endif
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.