1 /* SPDX-License-Identifier: GPL-2.0-only */ !! 1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* !! 2 /* linux/arch/sparc/lib/memset.S: Sparc optimized memset, bzero and clear_user code 3 * Copyright (c) 2011, The Linux Foundation. A !! 3 * Copyright (C) 1991,1996 Free Software Foundation >> 4 * Copyright (C) 1996,1997 Jakub Jelinek (jj@sunsite.mff.cuni.cz) >> 5 * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu) >> 6 * >> 7 * Calls to memset returns initial %o0. Calls to bzero returns 0, if ok, and >> 8 * number of bytes not yet set if exception occurs and we were called as >> 9 * clear_user. 4 */ 10 */ 5 11 >> 12 #include <linux/export.h> >> 13 #include <asm/ptrace.h> 6 14 7 /* HEXAGON assembly optimized memset */ !! 15 /* Work around cpp -rob */ 8 /* Replaces the standard library function mems !! 16 #define ALLOC #alloc >> 17 #define EXECINSTR #execinstr >> 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 you change the logic >> 47 * in the .fixup section below as well. >> 48 * Store 64 bytes at (BASE + OFFSET) using value SOURCE. */ >> 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); 9 68 10 << 11 .macro HEXAGON_OPT_FUNC_BEGIN name << 12 .text 69 .text 13 .p2align 4 !! 70 .align 4 14 .globl \name !! 71 15 .type \name, @function !! 72 .globl __bzero_begin 16 \name: !! 73 __bzero_begin: 17 .endm !! 74 18 !! 75 .globl __bzero 19 .macro HEXAGON_OPT_FUNC_FINISH name !! 76 .type __bzero,#function 20 .size \name, . - \name !! 77 .globl memset 21 .endm !! 78 EXPORT_SYMBOL(__bzero) 22 !! 79 EXPORT_SYMBOL(memset) 23 /* FUNCTION: memset (v2 version) */ !! 80 memset: 24 #if __HEXAGON_ARCH__ < 3 !! 81 mov %o0, %g1 25 HEXAGON_OPT_FUNC_BEGIN memset !! 82 mov 1, %g4 26 { !! 83 and %o1, 0xff, %g3 27 r6 = #8 !! 84 sll %g3, 8, %g2 28 r7 = extractu(r0, #3 , #0) !! 85 or %g3, %g2, %g3 29 p0 = cmp.eq(r2, #0) !! 86 sll %g3, 16, %g2 30 p1 = cmp.gtu(r2, #7) !! 87 or %g3, %g2, %g3 31 } !! 88 b 1f 32 { !! 89 mov %o2, %o1 33 r4 = vsplatb(r1) !! 90 3: 34 r8 = r0 /* leave r0 !! 91 cmp %o2, 3 35 r9 = sub(r6, r7) /* bytes unt !! 92 be 2f 36 if p0 jumpr r31 /* count == !! 93 EX(stb %g3, [%o0], sub %o1, 0) 37 } !! 94 38 { !! 95 cmp %o2, 2 39 r3 = #0 !! 96 be 2f 40 r7 = #0 !! 97 EX(stb %g3, [%o0 + 0x01], sub %o1, 1) 41 p0 = tstbit(r9, #0) !! 98 42 if p1 jump 2f /* skip byte loo !! 99 EX(stb %g3, [%o0 + 0x02], sub %o1, 2) 43 } !! 100 2: 44 !! 101 sub %o2, 4, %o2 45 /* less than 8 bytes to set, so just set a byt !! 102 add %o1, %o2, %o1 46 !! 103 b 4f 47 loop0(1f, r2) /* byte loop */ !! 104 sub %o0, %o2, %o0 48 .falign !! 105 49 1: /* byte loop */ !! 106 __bzero: 50 { !! 107 clr %g4 51 memb(r8++#1) = r4 !! 108 mov %g0, %g3 52 }:endloop0 !! 109 1: 53 jumpr r31 !! 110 cmp %o1, 7 54 .falign !! 111 bleu 7f 55 2: /* skip byte loop */ !! 112 andcc %o0, 3, %o2 56 { !! 113 57 r6 = #1 !! 114 bne 3b 58 p0 = tstbit(r9, #1) !! 115 4: 59 p1 = cmp.eq(r2, #1) !! 116 andcc %o0, 4, %g0 60 if !p0 jump 3f /* skip initial !! 117 61 } !! 118 be 2f 62 { !! 119 mov %g3, %g2 63 memb(r8++#1) = r4 !! 120 64 r3:2 = sub(r3:2, r7:6) !! 121 EX(st %g3, [%o0], sub %o1, 0) 65 if p1 jumpr r31 !! 122 sub %o1, 4, %o1 66 } !! 123 add %o0, 4, %o0 67 .falign !! 124 2: 68 3: /* skip initial byte store */ !! 125 andcc %o1, 0xffffff80, %o3 ! Now everything is 8 aligned and o1 is len to run 69 { !! 126 be 9f 70 r6 = #2 !! 127 andcc %o1, 0x78, %o2 71 p0 = tstbit(r9, #2) !! 128 10: 72 p1 = cmp.eq(r2, #2) !! 129 ZERO_BIG_BLOCK(%o0, 0x00, %g2) 73 if !p0 jump 4f /* skip initial !! 130 subcc %o3, 128, %o3 74 } !! 131 ZERO_BIG_BLOCK(%o0, 0x40, %g2) 75 { !! 132 bne 10b 76 memh(r8++#2) = r4 !! 133 add %o0, 128, %o0 77 r3:2 = sub(r3:2, r7:6) !! 134 78 if p1 jumpr r31 !! 135 orcc %o2, %g0, %g0 79 } !! 136 9: 80 .falign !! 137 be 13f 81 4: /* skip initial half store */ !! 138 andcc %o1, 7, %o1 82 { !! 139 83 r6 = #4 !! 140 srl %o2, 1, %o3 84 p0 = cmp.gtu(r2, #7) !! 141 set 13f, %o4 85 p1 = cmp.eq(r2, #4) !! 142 sub %o4, %o3, %o4 86 if !p0 jump 5f /* skip initial !! 143 jmp %o4 87 } !! 144 add %o0, %o2, %o0 88 { !! 145 89 memw(r8++#4) = r4 !! 146 ZERO_LAST_BLOCKS(%o0, 0x48, %g2) 90 r3:2 = sub(r3:2, r7:6) !! 147 ZERO_LAST_BLOCKS(%o0, 0x08, %g2) 91 p0 = cmp.gtu(r2, #11) !! 148 13: 92 if p1 jumpr r31 !! 149 be 8f 93 } !! 150 andcc %o1, 4, %g0 94 .falign !! 151 95 5: /* skip initial word store */ !! 152 be 1f 96 { !! 153 andcc %o1, 2, %g0 97 r10 = lsr(r2, #3) !! 154 98 p1 = cmp.eq(r3, #1) !! 155 EX(st %g3, [%o0], and %o1, 7) 99 if !p0 jump 7f /* skip double !! 156 add %o0, 4, %o0 100 } !! 157 1: 101 { !! 158 be 1f 102 r5 = r4 !! 159 andcc %o1, 1, %g0 103 r6 = #8 !! 160 104 loop0(6f, r10) /* double loop !! 161 EX(sth %g3, [%o0], and %o1, 3) 105 } !! 162 add %o0, 2, %o0 106 !! 163 1: 107 /* set bytes a double word at a time */ !! 164 bne,a 8f 108 !! 165 EX(stb %g3, [%o0], and %o1, 1) 109 .falign !! 166 8: 110 6: /* double loop */ !! 167 b 0f 111 { !! 168 nop 112 memd(r8++#8) = r5:4 !! 169 7: 113 r3:2 = sub(r3:2, r7:6) !! 170 be 13b 114 p1 = cmp.eq(r2, #8) !! 171 orcc %o1, 0, %g0 115 }:endloop0 !! 172 116 .falign !! 173 be 0f 117 7: /* skip double loop */ !! 174 8: 118 { !! 175 add %o0, 1, %o0 119 p0 = tstbit(r2, #2) !! 176 subcc %o1, 1, %o1 120 if p1 jumpr r31 !! 177 bne 8b 121 } !! 178 EX(stb %g3, [%o0 - 1], add %o1, 1) 122 { !! 179 0: 123 r6 = #4 !! 180 andcc %g4, 1, %g0 124 p0 = tstbit(r2, #1) !! 181 be 5f 125 p1 = cmp.eq(r2, #4) !! 182 nop 126 if !p0 jump 8f /* skip final w !! 183 retl 127 } !! 184 mov %g1, %o0 128 { !! 185 5: 129 memw(r8++#4) = r4 !! 186 retl 130 r3:2 = sub(r3:2, r7:6) !! 187 clr %o0 131 if p1 jumpr r31 !! 188 132 } !! 189 .section .fixup,#alloc,#execinstr 133 .falign !! 190 .align 4 134 8: /* skip final word store */ !! 191 30: 135 { !! 192 and %o1, 0x7f, %o1 136 p1 = cmp.eq(r2, #2) !! 193 retl 137 if !p0 jump 9f /* skip final h !! 194 add %o3, %o1, %o0 138 } !! 195 139 { !! 196 .globl __bzero_end 140 memh(r8++#2) = r4 !! 197 __bzero_end: 141 if p1 jumpr r31 << 142 } << 143 .falign << 144 9: /* skip final half store */ << 145 { << 146 memb(r8++#1) = r4 << 147 jumpr r31 << 148 } << 149 HEXAGON_OPT_FUNC_FINISH memset << 150 #endif << 151 << 152 << 153 /* FUNCTION: memset (v3 and higher version) << 154 #if __HEXAGON_ARCH__ >= 3 << 155 HEXAGON_OPT_FUNC_BEGIN memset << 156 { << 157 r7=vsplatb(r1) << 158 r6 = r0 << 159 if (r2==#0) jump:nt .L1 << 160 } << 161 { << 162 r5:4=combine(r7,r7) << 163 p0 = cmp.gtu(r2,#8) << 164 if (p0.new) jump:nt .L3 << 165 } << 166 { << 167 r3 = r0 << 168 loop0(.L47,r2) << 169 } << 170 .falign << 171 .L47: << 172 { << 173 memb(r3++#1) = r1 << 174 }:endloop0 /* start=.L47 */ << 175 jumpr r31 << 176 .L3: << 177 { << 178 p0 = tstbit(r0,#0) << 179 if (!p0.new) jump:nt .L8 << 180 p1 = cmp.eq(r2, #1) << 181 } << 182 { << 183 r6 = add(r0, #1) << 184 r2 = add(r2,#-1) << 185 memb(r0) = r1 << 186 if (p1) jump .L1 << 187 } << 188 .L8: << 189 { << 190 p0 = tstbit(r6,#1) << 191 if (!p0.new) jump:nt .L10 << 192 } << 193 { << 194 r2 = add(r2,#-2) << 195 memh(r6++#2) = r7 << 196 p0 = cmp.eq(r2, #2) << 197 if (p0.new) jump:nt .L1 << 198 } << 199 .L10: << 200 { << 201 p0 = tstbit(r6,#2) << 202 if (!p0.new) jump:nt .L12 << 203 } << 204 { << 205 r2 = add(r2,#-4) << 206 memw(r6++#4) = r7 << 207 p0 = cmp.eq(r2, #4) << 208 if (p0.new) jump:nt .L1 << 209 } << 210 .L12: << 211 { << 212 p0 = cmp.gtu(r2,#127) << 213 if (!p0.new) jump:nt .L14 << 214 } << 215 r3 = and(r6,#31) << 216 if (r3==#0) jump:nt .L17 << 217 { << 218 memd(r6++#8) = r5:4 << 219 r2 = add(r2,#-8) << 220 } << 221 r3 = and(r6,#31) << 222 if (r3==#0) jump:nt .L17 << 223 { << 224 memd(r6++#8) = r5:4 << 225 r2 = add(r2,#-8) << 226 } << 227 r3 = and(r6,#31) << 228 if (r3==#0) jump:nt .L17 << 229 { << 230 memd(r6++#8) = r5:4 << 231 r2 = add(r2,#-8) << 232 } << 233 .L17: << 234 { << 235 r3 = lsr(r2,#5) << 236 if (r1!=#0) jump:nt .L18 << 237 } << 238 { << 239 r8 = r3 << 240 r3 = r6 << 241 loop0(.L46,r3) << 242 } << 243 .falign << 244 .L46: << 245 { << 246 dczeroa(r6) << 247 r6 = add(r6,#32) << 248 r2 = add(r2,#-32) << 249 }:endloop0 /* start=.L46 */ << 250 .L14: << 251 { << 252 p0 = cmp.gtu(r2,#7) << 253 if (!p0.new) jump:nt .L28 << 254 r8 = lsr(r2,#3) << 255 } << 256 loop0(.L44,r8) << 257 .falign << 258 .L44: << 259 { << 260 memd(r6++#8) = r5:4 << 261 r2 = add(r2,#-8) << 262 }:endloop0 /* start=.L44 */ << 263 .L28: << 264 { << 265 p0 = tstbit(r2,#2) << 266 if (!p0.new) jump:nt .L33 << 267 } << 268 { << 269 r2 = add(r2,#-4) << 270 memw(r6++#4) = r7 << 271 } << 272 .L33: << 273 { << 274 p0 = tstbit(r2,#1) << 275 if (!p0.new) jump:nt .L35 << 276 } << 277 { << 278 r2 = add(r2,#-2) << 279 memh(r6++#2) = r7 << 280 } << 281 .L35: << 282 p0 = cmp.eq(r2,#1) << 283 if (p0) memb(r6) = r1 << 284 .L1: << 285 jumpr r31 << 286 .L18: << 287 loop0(.L45,r3) << 288 .falign << 289 .L45: << 290 dczeroa(r6) << 291 { << 292 memd(r6++#8) = r5:4 << 293 r2 = add(r2,#-32) << 294 } << 295 memd(r6++#8) = r5:4 << 296 memd(r6++#8) = r5:4 << 297 { << 298 memd(r6++#8) = r5:4 << 299 }:endloop0 /* start=.L45 */ << 300 jump .L14 << 301 HEXAGON_OPT_FUNC_FINISH memset << 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.