Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * This file is subject to the terms and conditions of the GNU General Public |
| 3 | * License. See the file "COPYING" in the main directory of this archive |
| 4 | * for more details. |
| 5 | * |
| 6 | * Unified implementation of memcpy, memmove and the __copy_user backend. |
| 7 | * |
| 8 | * Copyright (C) 1998, 99, 2000, 01, 2002 Ralf Baechle (ralf@gnu.org) |
| 9 | * Copyright (C) 1999, 2000, 01, 2002 Silicon Graphics, Inc. |
| 10 | * Copyright (C) 2002 Broadcom, Inc. |
| 11 | * memcpy/copy_user author: Mark Vandevoorde |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 12 | * Copyright (C) 2007 Maciej W. Rozycki |
Markos Chandras | 5bc0597 | 2014-01-07 12:57:04 +0000 | [diff] [blame] | 13 | * Copyright (C) 2014 Imagination Technologies Ltd. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 14 | * |
| 15 | * Mnemonic names for arguments to memcpy/__copy_user |
| 16 | */ |
Ralf Baechle | e5adb87 | 2005-10-20 22:55:26 +0100 | [diff] [blame] | 17 | |
| 18 | /* |
| 19 | * Hack to resolve longstanding prefetch issue |
| 20 | * |
| 21 | * Prefetching may be fatal on some systems if we're prefetching beyond the |
| 22 | * end of memory on some systems. It's also a seriously bad idea on non |
| 23 | * dma-coherent systems. |
| 24 | */ |
Ralf Baechle | 634286f | 2009-01-28 17:48:40 +0000 | [diff] [blame] | 25 | #ifdef CONFIG_DMA_NONCOHERENT |
Ralf Baechle | e5adb87 | 2005-10-20 22:55:26 +0100 | [diff] [blame] | 26 | #undef CONFIG_CPU_HAS_PREFETCH |
| 27 | #endif |
| 28 | #ifdef CONFIG_MIPS_MALTA |
| 29 | #undef CONFIG_CPU_HAS_PREFETCH |
| 30 | #endif |
| 31 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 32 | #include <asm/asm.h> |
Sam Ravnborg | 048eb58 | 2005-09-09 22:32:31 +0200 | [diff] [blame] | 33 | #include <asm/asm-offsets.h> |
Paul Burton | 576a2f0 | 2016-11-07 11:14:15 +0000 | [diff] [blame] | 34 | #include <asm/export.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 35 | #include <asm/regdef.h> |
| 36 | |
| 37 | #define dst a0 |
| 38 | #define src a1 |
| 39 | #define len a2 |
| 40 | |
| 41 | /* |
| 42 | * Spec |
| 43 | * |
| 44 | * memcpy copies len bytes from src to dst and sets v0 to dst. |
| 45 | * It assumes that |
| 46 | * - src and dst don't overlap |
| 47 | * - src is readable |
| 48 | * - dst is writable |
| 49 | * memcpy uses the standard calling convention |
| 50 | * |
| 51 | * __copy_user copies up to len bytes from src to dst and sets a2 (len) to |
| 52 | * the number of uncopied bytes due to an exception caused by a read or write. |
| 53 | * __copy_user assumes that src and dst don't overlap, and that the call is |
| 54 | * implementing one of the following: |
| 55 | * copy_to_user |
| 56 | * - src is readable (no exceptions when reading src) |
| 57 | * copy_from_user |
| 58 | * - dst is writable (no exceptions when writing dst) |
| 59 | * __copy_user uses a non-standard calling convention; see |
| 60 | * include/asm-mips/uaccess.h |
| 61 | * |
| 62 | * When an exception happens on a load, the handler must |
| 63 | # ensure that all of the destination buffer is overwritten to prevent |
| 64 | * leaking information to user mode programs. |
| 65 | */ |
| 66 | |
| 67 | /* |
| 68 | * Implementation |
| 69 | */ |
| 70 | |
| 71 | /* |
| 72 | * The exception handler for loads requires that: |
| 73 | * 1- AT contain the address of the byte just past the end of the source |
| 74 | * of the copy, |
| 75 | * 2- src_entry <= src < AT, and |
| 76 | * 3- (dst - src) == (dst_entry - src_entry), |
| 77 | * The _entry suffix denotes values when __copy_user was called. |
| 78 | * |
| 79 | * (1) is set up up by uaccess.h and maintained by not writing AT in copy_user |
| 80 | * (2) is met by incrementing src by the number of bytes copied |
| 81 | * (3) is met by not doing loads between a pair of increments of dst and src |
| 82 | * |
| 83 | * The exception handlers for stores adjust len (if necessary) and return. |
| 84 | * These handlers do not need to overwrite any data. |
| 85 | * |
| 86 | * For __rmemcpy and memmove an exception is always a kernel bug, therefore |
| 87 | * they're not protected. |
| 88 | */ |
| 89 | |
Markos Chandras | 5bc0597 | 2014-01-07 12:57:04 +0000 | [diff] [blame] | 90 | /* Instruction type */ |
| 91 | #define LD_INSN 1 |
| 92 | #define ST_INSN 2 |
Markos Chandras | bda4d98 | 2014-01-07 15:59:03 +0000 | [diff] [blame] | 93 | /* Pretech type */ |
| 94 | #define SRC_PREFETCH 1 |
| 95 | #define DST_PREFETCH 2 |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 96 | #define LEGACY_MODE 1 |
| 97 | #define EVA_MODE 2 |
| 98 | #define USEROP 1 |
| 99 | #define KERNELOP 2 |
Markos Chandras | 5bc0597 | 2014-01-07 12:57:04 +0000 | [diff] [blame] | 100 | |
| 101 | /* |
| 102 | * Wrapper to add an entry in the exception table |
| 103 | * in case the insn causes a memory exception. |
| 104 | * Arguments: |
| 105 | * insn : Load/store instruction |
| 106 | * type : Instruction type |
| 107 | * reg : Register |
| 108 | * addr : Address |
| 109 | * handler : Exception handler |
| 110 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 111 | |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 112 | #define EXC(insn, type, reg, addr, handler) \ |
| 113 | .if \mode == LEGACY_MODE; \ |
| 114 | 9: insn reg, addr; \ |
| 115 | .section __ex_table,"a"; \ |
| 116 | PTR 9b, handler; \ |
| 117 | .previous; \ |
Markos Chandras | cd26cb4 | 2014-01-07 16:20:22 +0000 | [diff] [blame] | 118 | /* This is assembled in EVA mode */ \ |
| 119 | .else; \ |
| 120 | /* If loading from user or storing to user */ \ |
| 121 | .if ((\from == USEROP) && (type == LD_INSN)) || \ |
| 122 | ((\to == USEROP) && (type == ST_INSN)); \ |
| 123 | 9: __BUILD_EVA_INSN(insn##e, reg, addr); \ |
| 124 | .section __ex_table,"a"; \ |
| 125 | PTR 9b, handler; \ |
| 126 | .previous; \ |
| 127 | .else; \ |
| 128 | /* \ |
| 129 | * Still in EVA, but no need for \ |
| 130 | * exception handler or EVA insn \ |
| 131 | */ \ |
| 132 | insn reg, addr; \ |
| 133 | .endif; \ |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 134 | .endif |
Markos Chandras | cd26cb4 | 2014-01-07 16:20:22 +0000 | [diff] [blame] | 135 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 136 | /* |
| 137 | * Only on the 64-bit kernel we can made use of 64-bit registers. |
| 138 | */ |
Ralf Baechle | 875d43e | 2005-09-03 15:56:16 -0700 | [diff] [blame] | 139 | #ifdef CONFIG_64BIT |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 140 | #define USE_DOUBLE |
| 141 | #endif |
| 142 | |
| 143 | #ifdef USE_DOUBLE |
| 144 | |
Markos Chandras | 5bc0597 | 2014-01-07 12:57:04 +0000 | [diff] [blame] | 145 | #define LOADK ld /* No exception */ |
| 146 | #define LOAD(reg, addr, handler) EXC(ld, LD_INSN, reg, addr, handler) |
| 147 | #define LOADL(reg, addr, handler) EXC(ldl, LD_INSN, reg, addr, handler) |
| 148 | #define LOADR(reg, addr, handler) EXC(ldr, LD_INSN, reg, addr, handler) |
| 149 | #define STOREL(reg, addr, handler) EXC(sdl, ST_INSN, reg, addr, handler) |
| 150 | #define STORER(reg, addr, handler) EXC(sdr, ST_INSN, reg, addr, handler) |
| 151 | #define STORE(reg, addr, handler) EXC(sd, ST_INSN, reg, addr, handler) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 152 | #define ADD daddu |
| 153 | #define SUB dsubu |
| 154 | #define SRL dsrl |
| 155 | #define SRA dsra |
| 156 | #define SLL dsll |
| 157 | #define SLLV dsllv |
| 158 | #define SRLV dsrlv |
| 159 | #define NBYTES 8 |
| 160 | #define LOG_NBYTES 3 |
| 161 | |
Ralf Baechle | 42a3b4f | 2005-09-03 15:56:17 -0700 | [diff] [blame] | 162 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 163 | * As we are sharing code base with the mips32 tree (which use the o32 ABI |
| 164 | * register definitions). We need to redefine the register definitions from |
| 165 | * the n64 ABI register naming to the o32 ABI register naming. |
| 166 | */ |
| 167 | #undef t0 |
| 168 | #undef t1 |
| 169 | #undef t2 |
| 170 | #undef t3 |
| 171 | #define t0 $8 |
| 172 | #define t1 $9 |
| 173 | #define t2 $10 |
| 174 | #define t3 $11 |
| 175 | #define t4 $12 |
| 176 | #define t5 $13 |
| 177 | #define t6 $14 |
| 178 | #define t7 $15 |
Ralf Baechle | 42a3b4f | 2005-09-03 15:56:17 -0700 | [diff] [blame] | 179 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 180 | #else |
| 181 | |
Markos Chandras | 5bc0597 | 2014-01-07 12:57:04 +0000 | [diff] [blame] | 182 | #define LOADK lw /* No exception */ |
| 183 | #define LOAD(reg, addr, handler) EXC(lw, LD_INSN, reg, addr, handler) |
| 184 | #define LOADL(reg, addr, handler) EXC(lwl, LD_INSN, reg, addr, handler) |
| 185 | #define LOADR(reg, addr, handler) EXC(lwr, LD_INSN, reg, addr, handler) |
| 186 | #define STOREL(reg, addr, handler) EXC(swl, ST_INSN, reg, addr, handler) |
| 187 | #define STORER(reg, addr, handler) EXC(swr, ST_INSN, reg, addr, handler) |
| 188 | #define STORE(reg, addr, handler) EXC(sw, ST_INSN, reg, addr, handler) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 189 | #define ADD addu |
| 190 | #define SUB subu |
| 191 | #define SRL srl |
| 192 | #define SLL sll |
| 193 | #define SRA sra |
| 194 | #define SLLV sllv |
| 195 | #define SRLV srlv |
| 196 | #define NBYTES 4 |
| 197 | #define LOG_NBYTES 2 |
| 198 | |
| 199 | #endif /* USE_DOUBLE */ |
| 200 | |
Markos Chandras | 5bc0597 | 2014-01-07 12:57:04 +0000 | [diff] [blame] | 201 | #define LOADB(reg, addr, handler) EXC(lb, LD_INSN, reg, addr, handler) |
| 202 | #define STOREB(reg, addr, handler) EXC(sb, ST_INSN, reg, addr, handler) |
| 203 | |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 204 | #define _PREF(hint, addr, type) \ |
| 205 | .if \mode == LEGACY_MODE; \ |
| 206 | PREF(hint, addr); \ |
Markos Chandras | cd26cb4 | 2014-01-07 16:20:22 +0000 | [diff] [blame] | 207 | .else; \ |
| 208 | .if ((\from == USEROP) && (type == SRC_PREFETCH)) || \ |
| 209 | ((\to == USEROP) && (type == DST_PREFETCH)); \ |
| 210 | /* \ |
| 211 | * PREFE has only 9 bits for the offset \ |
| 212 | * compared to PREF which has 16, so it may \ |
| 213 | * need to use the $at register but this \ |
| 214 | * register should remain intact because it's \ |
| 215 | * used later on. Therefore use $v1. \ |
| 216 | */ \ |
| 217 | .set at=v1; \ |
| 218 | PREFE(hint, addr); \ |
| 219 | .set noat; \ |
| 220 | .else; \ |
| 221 | PREF(hint, addr); \ |
| 222 | .endif; \ |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 223 | .endif |
Markos Chandras | bda4d98 | 2014-01-07 15:59:03 +0000 | [diff] [blame] | 224 | |
| 225 | #define PREFS(hint, addr) _PREF(hint, addr, SRC_PREFETCH) |
| 226 | #define PREFD(hint, addr) _PREF(hint, addr, DST_PREFETCH) |
| 227 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 228 | #ifdef CONFIG_CPU_LITTLE_ENDIAN |
| 229 | #define LDFIRST LOADR |
Ralf Baechle | 7034228 | 2013-01-22 12:59:30 +0100 | [diff] [blame] | 230 | #define LDREST LOADL |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 231 | #define STFIRST STORER |
Ralf Baechle | 7034228 | 2013-01-22 12:59:30 +0100 | [diff] [blame] | 232 | #define STREST STOREL |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 233 | #define SHIFT_DISCARD SLLV |
| 234 | #else |
| 235 | #define LDFIRST LOADL |
Ralf Baechle | 7034228 | 2013-01-22 12:59:30 +0100 | [diff] [blame] | 236 | #define LDREST LOADR |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 237 | #define STFIRST STOREL |
Ralf Baechle | 7034228 | 2013-01-22 12:59:30 +0100 | [diff] [blame] | 238 | #define STREST STORER |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 239 | #define SHIFT_DISCARD SRLV |
| 240 | #endif |
| 241 | |
| 242 | #define FIRST(unit) ((unit)*NBYTES) |
| 243 | #define REST(unit) (FIRST(unit)+NBYTES-1) |
| 244 | #define UNIT(unit) FIRST(unit) |
| 245 | |
| 246 | #define ADDRMASK (NBYTES-1) |
| 247 | |
| 248 | .text |
| 249 | .set noreorder |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 250 | #ifndef CONFIG_CPU_DADDI_WORKAROUNDS |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 251 | .set noat |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 252 | #else |
| 253 | .set at=v1 |
| 254 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 255 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 256 | .align 5 |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 257 | |
| 258 | /* |
| 259 | * Macro to build the __copy_user common code |
Andrea Gelmini | 209b877 | 2016-05-21 14:01:36 +0200 | [diff] [blame] | 260 | * Arguments: |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 261 | * mode : LEGACY_MODE or EVA_MODE |
| 262 | * from : Source operand. USEROP or KERNELOP |
| 263 | * to : Destination operand. USEROP or KERNELOP |
| 264 | */ |
| 265 | .macro __BUILD_COPY_USER mode, from, to |
| 266 | |
| 267 | /* initialize __memcpy if this the first time we execute this macro */ |
| 268 | .ifnotdef __memcpy |
| 269 | .set __memcpy, 1 |
| 270 | .hidden __memcpy /* make sure it does not leak */ |
| 271 | .endif |
| 272 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 273 | /* |
| 274 | * Note: dst & src may be unaligned, len may be 0 |
| 275 | * Temps |
| 276 | */ |
| 277 | #define rem t8 |
| 278 | |
Thomas Bogendoerfer | 930bff8 | 2007-11-25 11:47:56 +0100 | [diff] [blame] | 279 | R10KCBARRIER(0(ra)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 280 | /* |
| 281 | * The "issue break"s below are very approximate. |
| 282 | * Issue delays for dcache fills will perturb the schedule, as will |
| 283 | * load queue full replay traps, etc. |
| 284 | * |
| 285 | * If len < NBYTES use byte operations. |
| 286 | */ |
Markos Chandras | bda4d98 | 2014-01-07 15:59:03 +0000 | [diff] [blame] | 287 | PREFS( 0, 0(src) ) |
| 288 | PREFD( 1, 0(dst) ) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 289 | sltu t2, len, NBYTES |
| 290 | and t1, dst, ADDRMASK |
Markos Chandras | bda4d98 | 2014-01-07 15:59:03 +0000 | [diff] [blame] | 291 | PREFS( 0, 1*32(src) ) |
| 292 | PREFD( 1, 1*32(dst) ) |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 293 | bnez t2, .Lcopy_bytes_checklen\@ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 294 | and t0, src, ADDRMASK |
Markos Chandras | bda4d98 | 2014-01-07 15:59:03 +0000 | [diff] [blame] | 295 | PREFS( 0, 2*32(src) ) |
| 296 | PREFD( 1, 2*32(dst) ) |
Leonid Yegoshin | b0ce4bd | 2014-11-14 11:55:50 +0000 | [diff] [blame] | 297 | #ifndef CONFIG_CPU_MIPSR6 |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 298 | bnez t1, .Ldst_unaligned\@ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 299 | nop |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 300 | bnez t0, .Lsrc_unaligned_dst_aligned\@ |
Leonid Yegoshin | b0ce4bd | 2014-11-14 11:55:50 +0000 | [diff] [blame] | 301 | #else |
| 302 | or t0, t0, t1 |
| 303 | bnez t0, .Lcopy_unaligned_bytes\@ |
| 304 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 305 | /* |
| 306 | * use delay slot for fall-through |
| 307 | * src and dst are aligned; need to compute rem |
| 308 | */ |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 309 | .Lboth_aligned\@: |
Ralf Baechle | 7034228 | 2013-01-22 12:59:30 +0100 | [diff] [blame] | 310 | SRL t0, len, LOG_NBYTES+3 # +3 for 8 units/iter |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 311 | beqz t0, .Lcleanup_both_aligned\@ # len < 8*NBYTES |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 312 | and rem, len, (8*NBYTES-1) # rem = len % (8*NBYTES) |
Markos Chandras | bda4d98 | 2014-01-07 15:59:03 +0000 | [diff] [blame] | 313 | PREFS( 0, 3*32(src) ) |
| 314 | PREFD( 1, 3*32(dst) ) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 315 | .align 4 |
| 316 | 1: |
Thomas Bogendoerfer | 930bff8 | 2007-11-25 11:47:56 +0100 | [diff] [blame] | 317 | R10KCBARRIER(0(ra)) |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 318 | LOAD(t0, UNIT(0)(src), .Ll_exc\@) |
| 319 | LOAD(t1, UNIT(1)(src), .Ll_exc_copy\@) |
| 320 | LOAD(t2, UNIT(2)(src), .Ll_exc_copy\@) |
| 321 | LOAD(t3, UNIT(3)(src), .Ll_exc_copy\@) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 322 | SUB len, len, 8*NBYTES |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 323 | LOAD(t4, UNIT(4)(src), .Ll_exc_copy\@) |
| 324 | LOAD(t7, UNIT(5)(src), .Ll_exc_copy\@) |
| 325 | STORE(t0, UNIT(0)(dst), .Ls_exc_p8u\@) |
| 326 | STORE(t1, UNIT(1)(dst), .Ls_exc_p7u\@) |
| 327 | LOAD(t0, UNIT(6)(src), .Ll_exc_copy\@) |
| 328 | LOAD(t1, UNIT(7)(src), .Ll_exc_copy\@) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 329 | ADD src, src, 8*NBYTES |
| 330 | ADD dst, dst, 8*NBYTES |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 331 | STORE(t2, UNIT(-6)(dst), .Ls_exc_p6u\@) |
| 332 | STORE(t3, UNIT(-5)(dst), .Ls_exc_p5u\@) |
| 333 | STORE(t4, UNIT(-4)(dst), .Ls_exc_p4u\@) |
| 334 | STORE(t7, UNIT(-3)(dst), .Ls_exc_p3u\@) |
| 335 | STORE(t0, UNIT(-2)(dst), .Ls_exc_p2u\@) |
| 336 | STORE(t1, UNIT(-1)(dst), .Ls_exc_p1u\@) |
Markos Chandras | bda4d98 | 2014-01-07 15:59:03 +0000 | [diff] [blame] | 337 | PREFS( 0, 8*32(src) ) |
| 338 | PREFD( 1, 8*32(dst) ) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 339 | bne len, rem, 1b |
| 340 | nop |
| 341 | |
| 342 | /* |
| 343 | * len == rem == the number of bytes left to copy < 8*NBYTES |
| 344 | */ |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 345 | .Lcleanup_both_aligned\@: |
| 346 | beqz len, .Ldone\@ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 347 | sltu t0, len, 4*NBYTES |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 348 | bnez t0, .Lless_than_4units\@ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 349 | and rem, len, (NBYTES-1) # rem = len % NBYTES |
| 350 | /* |
| 351 | * len >= 4*NBYTES |
| 352 | */ |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 353 | LOAD( t0, UNIT(0)(src), .Ll_exc\@) |
| 354 | LOAD( t1, UNIT(1)(src), .Ll_exc_copy\@) |
| 355 | LOAD( t2, UNIT(2)(src), .Ll_exc_copy\@) |
| 356 | LOAD( t3, UNIT(3)(src), .Ll_exc_copy\@) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 357 | SUB len, len, 4*NBYTES |
| 358 | ADD src, src, 4*NBYTES |
Thomas Bogendoerfer | 930bff8 | 2007-11-25 11:47:56 +0100 | [diff] [blame] | 359 | R10KCBARRIER(0(ra)) |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 360 | STORE(t0, UNIT(0)(dst), .Ls_exc_p4u\@) |
| 361 | STORE(t1, UNIT(1)(dst), .Ls_exc_p3u\@) |
| 362 | STORE(t2, UNIT(2)(dst), .Ls_exc_p2u\@) |
| 363 | STORE(t3, UNIT(3)(dst), .Ls_exc_p1u\@) |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 364 | .set reorder /* DADDI_WAR */ |
| 365 | ADD dst, dst, 4*NBYTES |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 366 | beqz len, .Ldone\@ |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 367 | .set noreorder |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 368 | .Lless_than_4units\@: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 369 | /* |
| 370 | * rem = len % NBYTES |
| 371 | */ |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 372 | beq rem, len, .Lcopy_bytes\@ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 373 | nop |
| 374 | 1: |
Thomas Bogendoerfer | 930bff8 | 2007-11-25 11:47:56 +0100 | [diff] [blame] | 375 | R10KCBARRIER(0(ra)) |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 376 | LOAD(t0, 0(src), .Ll_exc\@) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 377 | ADD src, src, NBYTES |
| 378 | SUB len, len, NBYTES |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 379 | STORE(t0, 0(dst), .Ls_exc_p1u\@) |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 380 | .set reorder /* DADDI_WAR */ |
| 381 | ADD dst, dst, NBYTES |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 382 | bne rem, len, 1b |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 383 | .set noreorder |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 384 | |
Leonid Yegoshin | b0ce4bd | 2014-11-14 11:55:50 +0000 | [diff] [blame] | 385 | #ifndef CONFIG_CPU_MIPSR6 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 386 | /* |
| 387 | * src and dst are aligned, need to copy rem bytes (rem < NBYTES) |
| 388 | * A loop would do only a byte at a time with possible branch |
Ralf Baechle | 7034228 | 2013-01-22 12:59:30 +0100 | [diff] [blame] | 389 | * mispredicts. Can't do an explicit LOAD dst,mask,or,STORE |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 390 | * because can't assume read-access to dst. Instead, use |
| 391 | * STREST dst, which doesn't require read access to dst. |
| 392 | * |
| 393 | * This code should perform better than a simple loop on modern, |
| 394 | * wide-issue mips processors because the code has fewer branches and |
| 395 | * more instruction-level parallelism. |
| 396 | */ |
| 397 | #define bits t2 |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 398 | beqz len, .Ldone\@ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 399 | ADD t1, dst, len # t1 is just past last byte of dst |
| 400 | li bits, 8*NBYTES |
| 401 | SLL rem, len, 3 # rem = number of bits to keep |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 402 | LOAD(t0, 0(src), .Ll_exc\@) |
Ralf Baechle | 7034228 | 2013-01-22 12:59:30 +0100 | [diff] [blame] | 403 | SUB bits, bits, rem # bits = number of bits to discard |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 404 | SHIFT_DISCARD t0, t0, bits |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 405 | STREST(t0, -1(t1), .Ls_exc\@) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 406 | jr ra |
| 407 | move len, zero |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 408 | .Ldst_unaligned\@: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 409 | /* |
| 410 | * dst is unaligned |
| 411 | * t0 = src & ADDRMASK |
| 412 | * t1 = dst & ADDRMASK; T1 > 0 |
| 413 | * len >= NBYTES |
| 414 | * |
| 415 | * Copy enough bytes to align dst |
| 416 | * Set match = (src and dst have same alignment) |
| 417 | */ |
| 418 | #define match rem |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 419 | LDFIRST(t3, FIRST(0)(src), .Ll_exc\@) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 420 | ADD t2, zero, NBYTES |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 421 | LDREST(t3, REST(0)(src), .Ll_exc_copy\@) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 422 | SUB t2, t2, t1 # t2 = number of bytes copied |
| 423 | xor match, t0, t1 |
Thomas Bogendoerfer | 930bff8 | 2007-11-25 11:47:56 +0100 | [diff] [blame] | 424 | R10KCBARRIER(0(ra)) |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 425 | STFIRST(t3, FIRST(0)(dst), .Ls_exc\@) |
| 426 | beq len, t2, .Ldone\@ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 427 | SUB len, len, t2 |
| 428 | ADD dst, dst, t2 |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 429 | beqz match, .Lboth_aligned\@ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 430 | ADD src, src, t2 |
| 431 | |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 432 | .Lsrc_unaligned_dst_aligned\@: |
Ralf Baechle | 7034228 | 2013-01-22 12:59:30 +0100 | [diff] [blame] | 433 | SRL t0, len, LOG_NBYTES+2 # +2 for 4 units/iter |
Markos Chandras | bda4d98 | 2014-01-07 15:59:03 +0000 | [diff] [blame] | 434 | PREFS( 0, 3*32(src) ) |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 435 | beqz t0, .Lcleanup_src_unaligned\@ |
Ralf Baechle | 7034228 | 2013-01-22 12:59:30 +0100 | [diff] [blame] | 436 | and rem, len, (4*NBYTES-1) # rem = len % 4*NBYTES |
Markos Chandras | bda4d98 | 2014-01-07 15:59:03 +0000 | [diff] [blame] | 437 | PREFD( 1, 3*32(dst) ) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 438 | 1: |
| 439 | /* |
| 440 | * Avoid consecutive LD*'s to the same register since some mips |
| 441 | * implementations can't issue them in the same cycle. |
| 442 | * It's OK to load FIRST(N+1) before REST(N) because the two addresses |
| 443 | * are to the same unit (unless src is aligned, but it's not). |
| 444 | */ |
Thomas Bogendoerfer | 930bff8 | 2007-11-25 11:47:56 +0100 | [diff] [blame] | 445 | R10KCBARRIER(0(ra)) |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 446 | LDFIRST(t0, FIRST(0)(src), .Ll_exc\@) |
| 447 | LDFIRST(t1, FIRST(1)(src), .Ll_exc_copy\@) |
Ralf Baechle | 7034228 | 2013-01-22 12:59:30 +0100 | [diff] [blame] | 448 | SUB len, len, 4*NBYTES |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 449 | LDREST(t0, REST(0)(src), .Ll_exc_copy\@) |
| 450 | LDREST(t1, REST(1)(src), .Ll_exc_copy\@) |
| 451 | LDFIRST(t2, FIRST(2)(src), .Ll_exc_copy\@) |
| 452 | LDFIRST(t3, FIRST(3)(src), .Ll_exc_copy\@) |
| 453 | LDREST(t2, REST(2)(src), .Ll_exc_copy\@) |
| 454 | LDREST(t3, REST(3)(src), .Ll_exc_copy\@) |
Markos Chandras | bda4d98 | 2014-01-07 15:59:03 +0000 | [diff] [blame] | 455 | PREFS( 0, 9*32(src) ) # 0 is PREF_LOAD (not streamed) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 456 | ADD src, src, 4*NBYTES |
| 457 | #ifdef CONFIG_CPU_SB1 |
| 458 | nop # improves slotting |
| 459 | #endif |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 460 | STORE(t0, UNIT(0)(dst), .Ls_exc_p4u\@) |
| 461 | STORE(t1, UNIT(1)(dst), .Ls_exc_p3u\@) |
| 462 | STORE(t2, UNIT(2)(dst), .Ls_exc_p2u\@) |
| 463 | STORE(t3, UNIT(3)(dst), .Ls_exc_p1u\@) |
Markos Chandras | bda4d98 | 2014-01-07 15:59:03 +0000 | [diff] [blame] | 464 | PREFD( 1, 9*32(dst) ) # 1 is PREF_STORE (not streamed) |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 465 | .set reorder /* DADDI_WAR */ |
| 466 | ADD dst, dst, 4*NBYTES |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 467 | bne len, rem, 1b |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 468 | .set noreorder |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 469 | |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 470 | .Lcleanup_src_unaligned\@: |
| 471 | beqz len, .Ldone\@ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 472 | and rem, len, NBYTES-1 # rem = len % NBYTES |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 473 | beq rem, len, .Lcopy_bytes\@ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 474 | nop |
| 475 | 1: |
Thomas Bogendoerfer | 930bff8 | 2007-11-25 11:47:56 +0100 | [diff] [blame] | 476 | R10KCBARRIER(0(ra)) |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 477 | LDFIRST(t0, FIRST(0)(src), .Ll_exc\@) |
| 478 | LDREST(t0, REST(0)(src), .Ll_exc_copy\@) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 479 | ADD src, src, NBYTES |
| 480 | SUB len, len, NBYTES |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 481 | STORE(t0, 0(dst), .Ls_exc_p1u\@) |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 482 | .set reorder /* DADDI_WAR */ |
| 483 | ADD dst, dst, NBYTES |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 484 | bne len, rem, 1b |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 485 | .set noreorder |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 486 | |
Leonid Yegoshin | b0ce4bd | 2014-11-14 11:55:50 +0000 | [diff] [blame] | 487 | #endif /* !CONFIG_CPU_MIPSR6 */ |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 488 | .Lcopy_bytes_checklen\@: |
| 489 | beqz len, .Ldone\@ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 490 | nop |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 491 | .Lcopy_bytes\@: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 492 | /* 0 < len < NBYTES */ |
Thomas Bogendoerfer | 930bff8 | 2007-11-25 11:47:56 +0100 | [diff] [blame] | 493 | R10KCBARRIER(0(ra)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 494 | #define COPY_BYTE(N) \ |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 495 | LOADB(t0, N(src), .Ll_exc\@); \ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 496 | SUB len, len, 1; \ |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 497 | beqz len, .Ldone\@; \ |
| 498 | STOREB(t0, N(dst), .Ls_exc_p1\@) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 499 | |
| 500 | COPY_BYTE(0) |
| 501 | COPY_BYTE(1) |
| 502 | #ifdef USE_DOUBLE |
| 503 | COPY_BYTE(2) |
| 504 | COPY_BYTE(3) |
| 505 | COPY_BYTE(4) |
| 506 | COPY_BYTE(5) |
| 507 | #endif |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 508 | LOADB(t0, NBYTES-2(src), .Ll_exc\@) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 509 | SUB len, len, 1 |
| 510 | jr ra |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 511 | STOREB(t0, NBYTES-2(dst), .Ls_exc_p1\@) |
| 512 | .Ldone\@: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 513 | jr ra |
Markos Chandras | 51b1029 | 2014-11-17 09:32:38 +0000 | [diff] [blame] | 514 | nop |
Leonid Yegoshin | b0ce4bd | 2014-11-14 11:55:50 +0000 | [diff] [blame] | 515 | |
| 516 | #ifdef CONFIG_CPU_MIPSR6 |
| 517 | .Lcopy_unaligned_bytes\@: |
| 518 | 1: |
| 519 | COPY_BYTE(0) |
| 520 | COPY_BYTE(1) |
| 521 | COPY_BYTE(2) |
| 522 | COPY_BYTE(3) |
| 523 | COPY_BYTE(4) |
| 524 | COPY_BYTE(5) |
| 525 | COPY_BYTE(6) |
| 526 | COPY_BYTE(7) |
| 527 | ADD src, src, 8 |
| 528 | b 1b |
| 529 | ADD dst, dst, 8 |
| 530 | #endif /* CONFIG_CPU_MIPSR6 */ |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 531 | .if __memcpy == 1 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 532 | END(memcpy) |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 533 | .set __memcpy, 0 |
| 534 | .hidden __memcpy |
| 535 | .endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 536 | |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 537 | .Ll_exc_copy\@: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 538 | /* |
| 539 | * Copy bytes from src until faulting load address (or until a |
| 540 | * lb faults) |
| 541 | * |
| 542 | * When reached by a faulting LDFIRST/LDREST, THREAD_BUADDR($28) |
| 543 | * may be more than a byte beyond the last address. |
| 544 | * Hence, the lb below may get an exception. |
| 545 | * |
| 546 | * Assumes src < THREAD_BUADDR($28) |
| 547 | */ |
Markos Chandras | 5bc0597 | 2014-01-07 12:57:04 +0000 | [diff] [blame] | 548 | LOADK t0, TI_TASK($28) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 549 | nop |
Markos Chandras | 5bc0597 | 2014-01-07 12:57:04 +0000 | [diff] [blame] | 550 | LOADK t0, THREAD_BUADDR(t0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 551 | 1: |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 552 | LOADB(t1, 0(src), .Ll_exc\@) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 553 | ADD src, src, 1 |
| 554 | sb t1, 0(dst) # can't fault -- we're copy_from_user |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 555 | .set reorder /* DADDI_WAR */ |
| 556 | ADD dst, dst, 1 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 557 | bne src, t0, 1b |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 558 | .set noreorder |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 559 | .Ll_exc\@: |
Markos Chandras | 5bc0597 | 2014-01-07 12:57:04 +0000 | [diff] [blame] | 560 | LOADK t0, TI_TASK($28) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 561 | nop |
Markos Chandras | 5bc0597 | 2014-01-07 12:57:04 +0000 | [diff] [blame] | 562 | LOADK t0, THREAD_BUADDR(t0) # t0 is just past last good address |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 563 | nop |
| 564 | SUB len, AT, t0 # len number of uncopied bytes |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 565 | jr ra |
| 566 | nop |
| 567 | |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 568 | #define SEXC(n) \ |
Ralf Baechle | 7034228 | 2013-01-22 12:59:30 +0100 | [diff] [blame] | 569 | .set reorder; /* DADDI_WAR */ \ |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 570 | .Ls_exc_p ## n ## u\@: \ |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 571 | ADD len, len, n*NBYTES; \ |
| 572 | jr ra; \ |
| 573 | .set noreorder |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 574 | |
| 575 | SEXC(8) |
| 576 | SEXC(7) |
| 577 | SEXC(6) |
| 578 | SEXC(5) |
| 579 | SEXC(4) |
| 580 | SEXC(3) |
| 581 | SEXC(2) |
| 582 | SEXC(1) |
| 583 | |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 584 | .Ls_exc_p1\@: |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 585 | .set reorder /* DADDI_WAR */ |
| 586 | ADD len, len, 1 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 587 | jr ra |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 588 | .set noreorder |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 589 | .Ls_exc\@: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 590 | jr ra |
| 591 | nop |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 592 | .endm |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 593 | |
| 594 | .align 5 |
| 595 | LEAF(memmove) |
Paul Burton | 576a2f0 | 2016-11-07 11:14:15 +0000 | [diff] [blame] | 596 | EXPORT_SYMBOL(memmove) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 597 | ADD t0, a0, a2 |
| 598 | ADD t1, a1, a2 |
| 599 | sltu t0, a1, t0 # dst + len <= src -> memcpy |
| 600 | sltu t1, a0, t1 # dst >= src + len -> memcpy |
| 601 | and t0, t1 |
Ralf Baechle | c5ec198 | 2008-01-29 10:14:59 +0000 | [diff] [blame] | 602 | beqz t0, .L__memcpy |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 603 | move v0, a0 /* return value */ |
Ralf Baechle | c5ec198 | 2008-01-29 10:14:59 +0000 | [diff] [blame] | 604 | beqz a2, .Lr_out |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 605 | END(memmove) |
| 606 | |
| 607 | /* fall through to __rmemcpy */ |
| 608 | LEAF(__rmemcpy) /* a0=dst a1=src a2=len */ |
| 609 | sltu t0, a1, a0 |
Ralf Baechle | c5ec198 | 2008-01-29 10:14:59 +0000 | [diff] [blame] | 610 | beqz t0, .Lr_end_bytes_up # src >= dst |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 611 | nop |
| 612 | ADD a0, a2 # dst = dst + len |
| 613 | ADD a1, a2 # src = src + len |
| 614 | |
Ralf Baechle | c5ec198 | 2008-01-29 10:14:59 +0000 | [diff] [blame] | 615 | .Lr_end_bytes: |
Thomas Bogendoerfer | 930bff8 | 2007-11-25 11:47:56 +0100 | [diff] [blame] | 616 | R10KCBARRIER(0(ra)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 617 | lb t0, -1(a1) |
| 618 | SUB a2, a2, 0x1 |
| 619 | sb t0, -1(a0) |
| 620 | SUB a1, a1, 0x1 |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 621 | .set reorder /* DADDI_WAR */ |
| 622 | SUB a0, a0, 0x1 |
Ralf Baechle | c5ec198 | 2008-01-29 10:14:59 +0000 | [diff] [blame] | 623 | bnez a2, .Lr_end_bytes |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 624 | .set noreorder |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 625 | |
Ralf Baechle | c5ec198 | 2008-01-29 10:14:59 +0000 | [diff] [blame] | 626 | .Lr_out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 627 | jr ra |
| 628 | move a2, zero |
| 629 | |
Ralf Baechle | c5ec198 | 2008-01-29 10:14:59 +0000 | [diff] [blame] | 630 | .Lr_end_bytes_up: |
Thomas Bogendoerfer | 930bff8 | 2007-11-25 11:47:56 +0100 | [diff] [blame] | 631 | R10KCBARRIER(0(ra)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 632 | lb t0, (a1) |
| 633 | SUB a2, a2, 0x1 |
| 634 | sb t0, (a0) |
| 635 | ADD a1, a1, 0x1 |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 636 | .set reorder /* DADDI_WAR */ |
| 637 | ADD a0, a0, 0x1 |
Ralf Baechle | c5ec198 | 2008-01-29 10:14:59 +0000 | [diff] [blame] | 638 | bnez a2, .Lr_end_bytes_up |
Maciej W. Rozycki | 619b6e1 | 2007-10-23 12:43:25 +0100 | [diff] [blame] | 639 | .set noreorder |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 640 | |
| 641 | jr ra |
| 642 | move a2, zero |
| 643 | END(__rmemcpy) |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 644 | |
| 645 | /* |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 646 | * A combined memcpy/__copy_user |
| 647 | * __copy_user sets len to 0 for success; else to an upper bound of |
| 648 | * the number of uncopied bytes. |
| 649 | * memcpy sets v0 to dst. |
| 650 | */ |
| 651 | .align 5 |
| 652 | LEAF(memcpy) /* a0=dst a1=src a2=len */ |
Paul Burton | 576a2f0 | 2016-11-07 11:14:15 +0000 | [diff] [blame] | 653 | EXPORT_SYMBOL(memcpy) |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 654 | move v0, dst /* return value */ |
| 655 | .L__memcpy: |
| 656 | FEXPORT(__copy_user) |
Paul Burton | 576a2f0 | 2016-11-07 11:14:15 +0000 | [diff] [blame] | 657 | EXPORT_SYMBOL(__copy_user) |
Markos Chandras | cf62a8b | 2014-01-07 14:34:05 +0000 | [diff] [blame] | 658 | /* Legacy Mode, user <-> user */ |
| 659 | __BUILD_COPY_USER LEGACY_MODE USEROP USEROP |
Markos Chandras | cd26cb4 | 2014-01-07 16:20:22 +0000 | [diff] [blame] | 660 | |
| 661 | #ifdef CONFIG_EVA |
| 662 | |
| 663 | /* |
| 664 | * For EVA we need distinct symbols for reading and writing to user space. |
| 665 | * This is because we need to use specific EVA instructions to perform the |
| 666 | * virtual <-> physical translation when a virtual address is actually in user |
| 667 | * space |
| 668 | */ |
| 669 | |
Markos Chandras | cd26cb4 | 2014-01-07 16:20:22 +0000 | [diff] [blame] | 670 | /* |
| 671 | * __copy_from_user (EVA) |
| 672 | */ |
| 673 | |
| 674 | LEAF(__copy_from_user_eva) |
Paul Burton | 576a2f0 | 2016-11-07 11:14:15 +0000 | [diff] [blame] | 675 | EXPORT_SYMBOL(__copy_from_user_eva) |
Markos Chandras | cd26cb4 | 2014-01-07 16:20:22 +0000 | [diff] [blame] | 676 | __BUILD_COPY_USER EVA_MODE USEROP KERNELOP |
| 677 | END(__copy_from_user_eva) |
| 678 | |
| 679 | |
| 680 | |
| 681 | /* |
| 682 | * __copy_to_user (EVA) |
| 683 | */ |
| 684 | |
| 685 | LEAF(__copy_to_user_eva) |
Paul Burton | 576a2f0 | 2016-11-07 11:14:15 +0000 | [diff] [blame] | 686 | EXPORT_SYMBOL(__copy_to_user_eva) |
Markos Chandras | cd26cb4 | 2014-01-07 16:20:22 +0000 | [diff] [blame] | 687 | __BUILD_COPY_USER EVA_MODE KERNELOP USEROP |
| 688 | END(__copy_to_user_eva) |
| 689 | |
| 690 | /* |
| 691 | * __copy_in_user (EVA) |
| 692 | */ |
| 693 | |
| 694 | LEAF(__copy_in_user_eva) |
Paul Burton | 576a2f0 | 2016-11-07 11:14:15 +0000 | [diff] [blame] | 695 | EXPORT_SYMBOL(__copy_in_user_eva) |
Markos Chandras | cd26cb4 | 2014-01-07 16:20:22 +0000 | [diff] [blame] | 696 | __BUILD_COPY_USER EVA_MODE USEROP USEROP |
| 697 | END(__copy_in_user_eva) |
| 698 | |
| 699 | #endif |