Greg Ungerer | 66d857b | 2011-03-22 13:39:27 +1000 | [diff] [blame] | 1 | /* |
| 2 | * INET An implementation of the TCP/IP protocol suite for the LINUX |
| 3 | * operating system. INET is implemented using the BSD Socket |
| 4 | * interface as the means of communication with the user level. |
| 5 | * |
| 6 | * IP/TCP/UDP checksumming routines |
| 7 | * |
| 8 | * Authors: Jorge Cwik, <jorge@laser.satlink.net> |
| 9 | * Arnt Gulbrandsen, <agulbra@nvg.unit.no> |
| 10 | * Tom May, <ftom@netcom.com> |
| 11 | * Andreas Schwab, <schwab@issan.informatik.uni-dortmund.de> |
| 12 | * Lots of code moved from tcp.c and ip.c; see those files |
| 13 | * for more names. |
| 14 | * |
| 15 | * 03/02/96 Jes Sorensen, Andreas Schwab, Roman Hodek: |
| 16 | * Fixed some nasty bugs, causing some horrible crashes. |
| 17 | * A: At some points, the sum (%0) was used as |
| 18 | * length-counter instead of the length counter |
| 19 | * (%1). Thanks to Roman Hodek for pointing this out. |
| 20 | * B: GCC seems to mess up if one uses too many |
| 21 | * data-registers to hold input values and one tries to |
| 22 | * specify d0 and d1 as scratch registers. Letting gcc |
| 23 | * choose these registers itself solves the problem. |
| 24 | * |
| 25 | * This program is free software; you can redistribute it and/or |
| 26 | * modify it under the terms of the GNU General Public License |
| 27 | * as published by the Free Software Foundation; either version |
| 28 | * 2 of the License, or (at your option) any later version. |
| 29 | * |
| 30 | * 1998/8/31 Andreas Schwab: |
| 31 | * Zero out rest of buffer on exception in |
| 32 | * csum_partial_copy_from_user. |
| 33 | */ |
| 34 | |
| 35 | #include <linux/module.h> |
| 36 | #include <net/checksum.h> |
| 37 | |
| 38 | /* |
| 39 | * computes a partial checksum, e.g. for TCP/UDP fragments |
| 40 | */ |
| 41 | |
| 42 | __wsum csum_partial(const void *buff, int len, __wsum sum) |
| 43 | { |
| 44 | unsigned long tmp1, tmp2; |
| 45 | /* |
| 46 | * Experiments with ethernet and slip connections show that buff |
| 47 | * is aligned on either a 2-byte or 4-byte boundary. |
| 48 | */ |
| 49 | __asm__("movel %2,%3\n\t" |
| 50 | "btst #1,%3\n\t" /* Check alignment */ |
| 51 | "jeq 2f\n\t" |
| 52 | "subql #2,%1\n\t" /* buff%4==2: treat first word */ |
| 53 | "jgt 1f\n\t" |
| 54 | "addql #2,%1\n\t" /* len was == 2, treat only rest */ |
| 55 | "jra 4f\n" |
| 56 | "1:\t" |
| 57 | "addw %2@+,%0\n\t" /* add first word to sum */ |
| 58 | "clrl %3\n\t" |
| 59 | "addxl %3,%0\n" /* add X bit */ |
| 60 | "2:\t" |
| 61 | /* unrolled loop for the main part: do 8 longs at once */ |
| 62 | "movel %1,%3\n\t" /* save len in tmp1 */ |
| 63 | "lsrl #5,%1\n\t" /* len/32 */ |
| 64 | "jeq 2f\n\t" /* not enough... */ |
| 65 | "subql #1,%1\n" |
| 66 | "1:\t" |
| 67 | "movel %2@+,%4\n\t" |
| 68 | "addxl %4,%0\n\t" |
| 69 | "movel %2@+,%4\n\t" |
| 70 | "addxl %4,%0\n\t" |
| 71 | "movel %2@+,%4\n\t" |
| 72 | "addxl %4,%0\n\t" |
| 73 | "movel %2@+,%4\n\t" |
| 74 | "addxl %4,%0\n\t" |
| 75 | "movel %2@+,%4\n\t" |
| 76 | "addxl %4,%0\n\t" |
| 77 | "movel %2@+,%4\n\t" |
| 78 | "addxl %4,%0\n\t" |
| 79 | "movel %2@+,%4\n\t" |
| 80 | "addxl %4,%0\n\t" |
| 81 | "movel %2@+,%4\n\t" |
| 82 | "addxl %4,%0\n\t" |
| 83 | "dbra %1,1b\n\t" |
| 84 | "clrl %4\n\t" |
| 85 | "addxl %4,%0\n\t" /* add X bit */ |
| 86 | "clrw %1\n\t" |
| 87 | "subql #1,%1\n\t" |
| 88 | "jcc 1b\n" |
| 89 | "2:\t" |
| 90 | "movel %3,%1\n\t" /* restore len from tmp1 */ |
| 91 | "andw #0x1c,%3\n\t" /* number of rest longs */ |
| 92 | "jeq 4f\n\t" |
| 93 | "lsrw #2,%3\n\t" |
| 94 | "subqw #1,%3\n" |
| 95 | "3:\t" |
| 96 | /* loop for rest longs */ |
| 97 | "movel %2@+,%4\n\t" |
| 98 | "addxl %4,%0\n\t" |
| 99 | "dbra %3,3b\n\t" |
| 100 | "clrl %4\n\t" |
| 101 | "addxl %4,%0\n" /* add X bit */ |
| 102 | "4:\t" |
| 103 | /* now check for rest bytes that do not fit into longs */ |
| 104 | "andw #3,%1\n\t" |
| 105 | "jeq 7f\n\t" |
| 106 | "clrl %4\n\t" /* clear tmp2 for rest bytes */ |
| 107 | "subqw #2,%1\n\t" |
| 108 | "jlt 5f\n\t" |
| 109 | "movew %2@+,%4\n\t" /* have rest >= 2: get word */ |
| 110 | "swap %4\n\t" /* into bits 16..31 */ |
| 111 | "tstw %1\n\t" /* another byte? */ |
| 112 | "jeq 6f\n" |
| 113 | "5:\t" |
| 114 | "moveb %2@,%4\n\t" /* have odd rest: get byte */ |
| 115 | "lslw #8,%4\n\t" /* into bits 8..15; 16..31 untouched */ |
| 116 | "6:\t" |
| 117 | "addl %4,%0\n\t" /* now add rest long to sum */ |
| 118 | "clrl %4\n\t" |
| 119 | "addxl %4,%0\n" /* add X bit */ |
| 120 | "7:\t" |
| 121 | : "=d" (sum), "=d" (len), "=a" (buff), |
| 122 | "=&d" (tmp1), "=&d" (tmp2) |
| 123 | : "0" (sum), "1" (len), "2" (buff) |
| 124 | ); |
| 125 | return(sum); |
| 126 | } |
| 127 | |
| 128 | EXPORT_SYMBOL(csum_partial); |
| 129 | |
| 130 | |
| 131 | /* |
| 132 | * copy from user space while checksumming, with exception handling. |
| 133 | */ |
| 134 | |
| 135 | __wsum |
| 136 | csum_partial_copy_from_user(const void __user *src, void *dst, |
| 137 | int len, __wsum sum, int *csum_err) |
| 138 | { |
| 139 | /* |
| 140 | * GCC doesn't like more than 10 operands for the asm |
| 141 | * statements so we have to use tmp2 for the error |
| 142 | * code. |
| 143 | */ |
| 144 | unsigned long tmp1, tmp2; |
| 145 | |
| 146 | __asm__("movel %2,%4\n\t" |
| 147 | "btst #1,%4\n\t" /* Check alignment */ |
| 148 | "jeq 2f\n\t" |
| 149 | "subql #2,%1\n\t" /* buff%4==2: treat first word */ |
| 150 | "jgt 1f\n\t" |
| 151 | "addql #2,%1\n\t" /* len was == 2, treat only rest */ |
| 152 | "jra 4f\n" |
| 153 | "1:\n" |
| 154 | "10:\t" |
| 155 | "movesw %2@+,%4\n\t" /* add first word to sum */ |
| 156 | "addw %4,%0\n\t" |
| 157 | "movew %4,%3@+\n\t" |
| 158 | "clrl %4\n\t" |
| 159 | "addxl %4,%0\n" /* add X bit */ |
| 160 | "2:\t" |
| 161 | /* unrolled loop for the main part: do 8 longs at once */ |
| 162 | "movel %1,%4\n\t" /* save len in tmp1 */ |
| 163 | "lsrl #5,%1\n\t" /* len/32 */ |
| 164 | "jeq 2f\n\t" /* not enough... */ |
| 165 | "subql #1,%1\n" |
| 166 | "1:\n" |
| 167 | "11:\t" |
| 168 | "movesl %2@+,%5\n\t" |
| 169 | "addxl %5,%0\n\t" |
| 170 | "movel %5,%3@+\n\t" |
| 171 | "12:\t" |
| 172 | "movesl %2@+,%5\n\t" |
| 173 | "addxl %5,%0\n\t" |
| 174 | "movel %5,%3@+\n\t" |
| 175 | "13:\t" |
| 176 | "movesl %2@+,%5\n\t" |
| 177 | "addxl %5,%0\n\t" |
| 178 | "movel %5,%3@+\n\t" |
| 179 | "14:\t" |
| 180 | "movesl %2@+,%5\n\t" |
| 181 | "addxl %5,%0\n\t" |
| 182 | "movel %5,%3@+\n\t" |
| 183 | "15:\t" |
| 184 | "movesl %2@+,%5\n\t" |
| 185 | "addxl %5,%0\n\t" |
| 186 | "movel %5,%3@+\n\t" |
| 187 | "16:\t" |
| 188 | "movesl %2@+,%5\n\t" |
| 189 | "addxl %5,%0\n\t" |
| 190 | "movel %5,%3@+\n\t" |
| 191 | "17:\t" |
| 192 | "movesl %2@+,%5\n\t" |
| 193 | "addxl %5,%0\n\t" |
| 194 | "movel %5,%3@+\n\t" |
| 195 | "18:\t" |
| 196 | "movesl %2@+,%5\n\t" |
| 197 | "addxl %5,%0\n\t" |
| 198 | "movel %5,%3@+\n\t" |
| 199 | "dbra %1,1b\n\t" |
| 200 | "clrl %5\n\t" |
| 201 | "addxl %5,%0\n\t" /* add X bit */ |
| 202 | "clrw %1\n\t" |
| 203 | "subql #1,%1\n\t" |
| 204 | "jcc 1b\n" |
| 205 | "2:\t" |
| 206 | "movel %4,%1\n\t" /* restore len from tmp1 */ |
| 207 | "andw #0x1c,%4\n\t" /* number of rest longs */ |
| 208 | "jeq 4f\n\t" |
| 209 | "lsrw #2,%4\n\t" |
| 210 | "subqw #1,%4\n" |
| 211 | "3:\n" |
| 212 | /* loop for rest longs */ |
| 213 | "19:\t" |
| 214 | "movesl %2@+,%5\n\t" |
| 215 | "addxl %5,%0\n\t" |
| 216 | "movel %5,%3@+\n\t" |
| 217 | "dbra %4,3b\n\t" |
| 218 | "clrl %5\n\t" |
| 219 | "addxl %5,%0\n" /* add X bit */ |
| 220 | "4:\t" |
| 221 | /* now check for rest bytes that do not fit into longs */ |
| 222 | "andw #3,%1\n\t" |
| 223 | "jeq 7f\n\t" |
| 224 | "clrl %5\n\t" /* clear tmp2 for rest bytes */ |
| 225 | "subqw #2,%1\n\t" |
| 226 | "jlt 5f\n\t" |
| 227 | "20:\t" |
| 228 | "movesw %2@+,%5\n\t" /* have rest >= 2: get word */ |
| 229 | "movew %5,%3@+\n\t" |
| 230 | "swap %5\n\t" /* into bits 16..31 */ |
| 231 | "tstw %1\n\t" /* another byte? */ |
| 232 | "jeq 6f\n" |
| 233 | "5:\n" |
| 234 | "21:\t" |
| 235 | "movesb %2@,%5\n\t" /* have odd rest: get byte */ |
| 236 | "moveb %5,%3@+\n\t" |
| 237 | "lslw #8,%5\n\t" /* into bits 8..15; 16..31 untouched */ |
| 238 | "6:\t" |
| 239 | "addl %5,%0\n\t" /* now add rest long to sum */ |
| 240 | "clrl %5\n\t" |
| 241 | "addxl %5,%0\n\t" /* add X bit */ |
| 242 | "7:\t" |
| 243 | "clrl %5\n" /* no error - clear return value */ |
| 244 | "8:\n" |
| 245 | ".section .fixup,\"ax\"\n" |
| 246 | ".even\n" |
| 247 | /* If any exception occurs zero out the rest. |
| 248 | Similarities with the code above are intentional :-) */ |
| 249 | "90:\t" |
| 250 | "clrw %3@+\n\t" |
| 251 | "movel %1,%4\n\t" |
| 252 | "lsrl #5,%1\n\t" |
| 253 | "jeq 1f\n\t" |
| 254 | "subql #1,%1\n" |
| 255 | "91:\t" |
| 256 | "clrl %3@+\n" |
| 257 | "92:\t" |
| 258 | "clrl %3@+\n" |
| 259 | "93:\t" |
| 260 | "clrl %3@+\n" |
| 261 | "94:\t" |
| 262 | "clrl %3@+\n" |
| 263 | "95:\t" |
| 264 | "clrl %3@+\n" |
| 265 | "96:\t" |
| 266 | "clrl %3@+\n" |
| 267 | "97:\t" |
| 268 | "clrl %3@+\n" |
| 269 | "98:\t" |
| 270 | "clrl %3@+\n\t" |
| 271 | "dbra %1,91b\n\t" |
| 272 | "clrw %1\n\t" |
| 273 | "subql #1,%1\n\t" |
| 274 | "jcc 91b\n" |
| 275 | "1:\t" |
| 276 | "movel %4,%1\n\t" |
| 277 | "andw #0x1c,%4\n\t" |
| 278 | "jeq 1f\n\t" |
| 279 | "lsrw #2,%4\n\t" |
| 280 | "subqw #1,%4\n" |
| 281 | "99:\t" |
| 282 | "clrl %3@+\n\t" |
| 283 | "dbra %4,99b\n\t" |
| 284 | "1:\t" |
| 285 | "andw #3,%1\n\t" |
| 286 | "jeq 9f\n" |
| 287 | "100:\t" |
| 288 | "clrw %3@+\n\t" |
| 289 | "tstw %1\n\t" |
| 290 | "jeq 9f\n" |
| 291 | "101:\t" |
| 292 | "clrb %3@+\n" |
| 293 | "9:\t" |
| 294 | #define STR(X) STR1(X) |
| 295 | #define STR1(X) #X |
| 296 | "moveq #-" STR(EFAULT) ",%5\n\t" |
| 297 | "jra 8b\n" |
| 298 | ".previous\n" |
| 299 | ".section __ex_table,\"a\"\n" |
| 300 | ".long 10b,90b\n" |
| 301 | ".long 11b,91b\n" |
| 302 | ".long 12b,92b\n" |
| 303 | ".long 13b,93b\n" |
| 304 | ".long 14b,94b\n" |
| 305 | ".long 15b,95b\n" |
| 306 | ".long 16b,96b\n" |
| 307 | ".long 17b,97b\n" |
| 308 | ".long 18b,98b\n" |
| 309 | ".long 19b,99b\n" |
| 310 | ".long 20b,100b\n" |
| 311 | ".long 21b,101b\n" |
| 312 | ".previous" |
| 313 | : "=d" (sum), "=d" (len), "=a" (src), "=a" (dst), |
| 314 | "=&d" (tmp1), "=d" (tmp2) |
| 315 | : "0" (sum), "1" (len), "2" (src), "3" (dst) |
| 316 | ); |
| 317 | |
| 318 | *csum_err = tmp2; |
| 319 | |
| 320 | return(sum); |
| 321 | } |
| 322 | |
| 323 | EXPORT_SYMBOL(csum_partial_copy_from_user); |
| 324 | |
| 325 | |
| 326 | /* |
| 327 | * copy from kernel space while checksumming, otherwise like csum_partial |
| 328 | */ |
| 329 | |
| 330 | __wsum |
| 331 | csum_partial_copy_nocheck(const void *src, void *dst, int len, __wsum sum) |
| 332 | { |
| 333 | unsigned long tmp1, tmp2; |
| 334 | __asm__("movel %2,%4\n\t" |
| 335 | "btst #1,%4\n\t" /* Check alignment */ |
| 336 | "jeq 2f\n\t" |
| 337 | "subql #2,%1\n\t" /* buff%4==2: treat first word */ |
| 338 | "jgt 1f\n\t" |
| 339 | "addql #2,%1\n\t" /* len was == 2, treat only rest */ |
| 340 | "jra 4f\n" |
| 341 | "1:\t" |
| 342 | "movew %2@+,%4\n\t" /* add first word to sum */ |
| 343 | "addw %4,%0\n\t" |
| 344 | "movew %4,%3@+\n\t" |
| 345 | "clrl %4\n\t" |
| 346 | "addxl %4,%0\n" /* add X bit */ |
| 347 | "2:\t" |
| 348 | /* unrolled loop for the main part: do 8 longs at once */ |
| 349 | "movel %1,%4\n\t" /* save len in tmp1 */ |
| 350 | "lsrl #5,%1\n\t" /* len/32 */ |
| 351 | "jeq 2f\n\t" /* not enough... */ |
| 352 | "subql #1,%1\n" |
| 353 | "1:\t" |
| 354 | "movel %2@+,%5\n\t" |
| 355 | "addxl %5,%0\n\t" |
| 356 | "movel %5,%3@+\n\t" |
| 357 | "movel %2@+,%5\n\t" |
| 358 | "addxl %5,%0\n\t" |
| 359 | "movel %5,%3@+\n\t" |
| 360 | "movel %2@+,%5\n\t" |
| 361 | "addxl %5,%0\n\t" |
| 362 | "movel %5,%3@+\n\t" |
| 363 | "movel %2@+,%5\n\t" |
| 364 | "addxl %5,%0\n\t" |
| 365 | "movel %5,%3@+\n\t" |
| 366 | "movel %2@+,%5\n\t" |
| 367 | "addxl %5,%0\n\t" |
| 368 | "movel %5,%3@+\n\t" |
| 369 | "movel %2@+,%5\n\t" |
| 370 | "addxl %5,%0\n\t" |
| 371 | "movel %5,%3@+\n\t" |
| 372 | "movel %2@+,%5\n\t" |
| 373 | "addxl %5,%0\n\t" |
| 374 | "movel %5,%3@+\n\t" |
| 375 | "movel %2@+,%5\n\t" |
| 376 | "addxl %5,%0\n\t" |
| 377 | "movel %5,%3@+\n\t" |
| 378 | "dbra %1,1b\n\t" |
| 379 | "clrl %5\n\t" |
| 380 | "addxl %5,%0\n\t" /* add X bit */ |
| 381 | "clrw %1\n\t" |
| 382 | "subql #1,%1\n\t" |
| 383 | "jcc 1b\n" |
| 384 | "2:\t" |
| 385 | "movel %4,%1\n\t" /* restore len from tmp1 */ |
| 386 | "andw #0x1c,%4\n\t" /* number of rest longs */ |
| 387 | "jeq 4f\n\t" |
| 388 | "lsrw #2,%4\n\t" |
| 389 | "subqw #1,%4\n" |
| 390 | "3:\t" |
| 391 | /* loop for rest longs */ |
| 392 | "movel %2@+,%5\n\t" |
| 393 | "addxl %5,%0\n\t" |
| 394 | "movel %5,%3@+\n\t" |
| 395 | "dbra %4,3b\n\t" |
| 396 | "clrl %5\n\t" |
| 397 | "addxl %5,%0\n" /* add X bit */ |
| 398 | "4:\t" |
| 399 | /* now check for rest bytes that do not fit into longs */ |
| 400 | "andw #3,%1\n\t" |
| 401 | "jeq 7f\n\t" |
| 402 | "clrl %5\n\t" /* clear tmp2 for rest bytes */ |
| 403 | "subqw #2,%1\n\t" |
| 404 | "jlt 5f\n\t" |
| 405 | "movew %2@+,%5\n\t" /* have rest >= 2: get word */ |
| 406 | "movew %5,%3@+\n\t" |
| 407 | "swap %5\n\t" /* into bits 16..31 */ |
| 408 | "tstw %1\n\t" /* another byte? */ |
| 409 | "jeq 6f\n" |
| 410 | "5:\t" |
| 411 | "moveb %2@,%5\n\t" /* have odd rest: get byte */ |
| 412 | "moveb %5,%3@+\n\t" |
| 413 | "lslw #8,%5\n" /* into bits 8..15; 16..31 untouched */ |
| 414 | "6:\t" |
| 415 | "addl %5,%0\n\t" /* now add rest long to sum */ |
| 416 | "clrl %5\n\t" |
| 417 | "addxl %5,%0\n" /* add X bit */ |
| 418 | "7:\t" |
| 419 | : "=d" (sum), "=d" (len), "=a" (src), "=a" (dst), |
| 420 | "=&d" (tmp1), "=&d" (tmp2) |
| 421 | : "0" (sum), "1" (len), "2" (src), "3" (dst) |
| 422 | ); |
| 423 | return(sum); |
| 424 | } |
| 425 | EXPORT_SYMBOL(csum_partial_copy_nocheck); |