| 1 | /* |
| 2 | * Copyright (c) 2012-2017 Apple Inc. All rights reserved. |
| 3 | * |
| 4 | * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ |
| 5 | * |
| 6 | * This file contains Original Code and/or Modifications of Original Code |
| 7 | * as defined in and that are subject to the Apple Public Source License |
| 8 | * Version 2.0 (the 'License'). You may not use this file except in |
| 9 | * compliance with the License. The rights granted to you under the License |
| 10 | * may not be used to create, or enable the creation or redistribution of, |
| 11 | * unlawful or unlicensed copies of an Apple operating system, or to |
| 12 | * circumvent, violate, or enable the circumvention or violation of, any |
| 13 | * terms of an Apple operating system software license agreement. |
| 14 | * |
| 15 | * Please obtain a copy of the License at |
| 16 | * http://www.opensource.apple.com/apsl/ and read it before using this file. |
| 17 | * |
| 18 | * The Original Code and all software distributed under the License are |
| 19 | * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER |
| 20 | * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, |
| 21 | * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, |
| 22 | * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. |
| 23 | * Please see the License for the specific language governing rights and |
| 24 | * limitations under the License. |
| 25 | * |
| 26 | * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ |
| 27 | */ |
| 28 | |
| 29 | /*- |
| 30 | * Copyright (c) 2008 Joerg Sonnenberger <joerg@NetBSD.org>. |
| 31 | * All rights reserved. |
| 32 | * |
| 33 | * Redistribution and use in source and binary forms, with or without |
| 34 | * modification, are permitted provided that the following conditions |
| 35 | * are met: |
| 36 | * |
| 37 | * 1. Redistributions of source code must retain the above copyright |
| 38 | * notice, this list of conditions and the following disclaimer. |
| 39 | * 2. Redistributions in binary form must reproduce the above copyright |
| 40 | * notice, this list of conditions and the following disclaimer in |
| 41 | * the documentation and/or other materials provided with the |
| 42 | * distribution. |
| 43 | * |
| 44 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| 45 | * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| 46 | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS |
| 47 | * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE |
| 48 | * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, |
| 49 | * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, |
| 50 | * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
| 51 | * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED |
| 52 | * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, |
| 53 | * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT |
| 54 | * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 55 | * SUCH DAMAGE. |
| 56 | */ |
| 57 | |
| 58 | #ifdef KERNEL |
| 59 | #include <sys/param.h> |
| 60 | #include <machine/endian.h> |
| 61 | #include <sys/mcache.h> |
| 62 | #include <sys/mbuf.h> |
| 63 | #include <kern/debug.h> |
| 64 | #include <libkern/libkern.h> |
| 65 | #include <mach/boolean.h> |
| 66 | #include <pexpert/pexpert.h> |
| 67 | #define CKSUM_ERR(fmt, args...) kprintf(fmt, ## args) |
| 68 | #else /* !KERNEL */ |
| 69 | #ifndef LIBSYSCALL_INTERFACE |
| 70 | #error "LIBSYSCALL_INTERFACE not defined" |
| 71 | #endif /* !LIBSYSCALL_INTERFACE */ |
| 72 | #include <stdlib.h> |
| 73 | #include <stddef.h> |
| 74 | #include <stdint.h> |
| 75 | #include <unistd.h> |
| 76 | #include <strings.h> |
| 77 | #include <mach/boolean.h> |
| 78 | #endif /* !KERNEL */ |
| 79 | |
| 80 | /* compile time assert */ |
| 81 | #ifndef _CASSERT |
| 82 | #define _CASSERT(x) _Static_assert(x, "compile-time assertion failed") |
| 83 | #endif /* !_CASSERT */ |
| 84 | |
| 85 | #ifndef VERIFY |
| 86 | #define VERIFY(EX) ((void)0) |
| 87 | #endif /* !VERIFY */ |
| 88 | |
| 89 | #ifndef CKSUM_ERR |
| 90 | #define CKSUM_ERR(fmt, args...) ((void)0) |
| 91 | #endif /* !CKSUM_ERR */ |
| 92 | |
| 93 | #define PREDICT_TRUE(x) __builtin_expect(!!((long)(x)), 1L) |
| 94 | #define PREDICT_FALSE(x) __builtin_expect(!!((long)(x)), 0L) |
| 95 | |
| 96 | /* fake mbuf struct used only for calling os_cpu_in_cksum_mbuf() */ |
| 97 | struct _mbuf { |
| 98 | struct _mbuf *_m_next; |
| 99 | void *_m_pad; |
| 100 | uint8_t *_m_data; |
| 101 | int32_t _m_len; |
| 102 | }; |
| 103 | |
| 104 | extern uint32_t os_cpu_in_cksum(const void *, uint32_t, uint32_t); |
| 105 | extern uint32_t os_cpu_in_cksum_mbuf(struct _mbuf *, int, int, uint32_t); |
| 106 | |
| 107 | uint32_t |
| 108 | os_cpu_in_cksum(const void *data, uint32_t len, uint32_t initial_sum) |
| 109 | { |
| 110 | /* |
| 111 | * If data is 4-bytes aligned, length is multiple of 4-bytes, |
| 112 | * and the amount to checksum is small, this would be quicker; |
| 113 | * this is suitable for IPv4 header. |
| 114 | */ |
| 115 | if (IS_P2ALIGNED(data, sizeof (uint32_t)) && |
| 116 | len <= 64 && (len & 3) == 0) { |
| 117 | uint8_t *p = __DECONST(uint8_t *, data); |
| 118 | uint64_t sum = initial_sum; |
| 119 | |
| 120 | if (PREDICT_TRUE(len == 20)) { /* simple IPv4 header */ |
| 121 | sum += *(uint32_t *)(void *)p; |
| 122 | sum += *(uint32_t *)(void *)(p + 4); |
| 123 | sum += *(uint32_t *)(void *)(p + 8); |
| 124 | sum += *(uint32_t *)(void *)(p + 12); |
| 125 | sum += *(uint32_t *)(void *)(p + 16); |
| 126 | } else { |
| 127 | while (len) { |
| 128 | sum += *(uint32_t *)(void *)p; |
| 129 | p += 4; |
| 130 | len -= 4; |
| 131 | } |
| 132 | } |
| 133 | |
| 134 | /* fold 64-bit to 16-bit (deferred carries) */ |
| 135 | sum = (sum >> 32) + (sum & 0xffffffff); /* 33-bit */ |
| 136 | sum = (sum >> 16) + (sum & 0xffff); /* 17-bit + carry */ |
| 137 | sum = (sum >> 16) + (sum & 0xffff); /* 16-bit + carry */ |
| 138 | sum = (sum >> 16) + (sum & 0xffff); /* final carry */ |
| 139 | |
| 140 | return (sum & 0xffff); |
| 141 | } |
| 142 | |
| 143 | /* |
| 144 | * Otherwise, let os_cpu_in_cksum_mbuf() handle it; it only looks |
| 145 | * at 3 fields: {next,data,len}, and since it doesn't care about |
| 146 | * the authenticity of the mbuf, we use a fake one here. Make |
| 147 | * sure the offsets are as expected. |
| 148 | */ |
| 149 | #if defined(__LP64__) |
| 150 | _CASSERT(offsetof(struct _mbuf, _m_next) == 0); |
| 151 | _CASSERT(offsetof(struct _mbuf, _m_data) == 16); |
| 152 | _CASSERT(offsetof(struct _mbuf, _m_len) == 24); |
| 153 | #else /* !__LP64__ */ |
| 154 | _CASSERT(offsetof(struct _mbuf, _m_next) == 0); |
| 155 | _CASSERT(offsetof(struct _mbuf, _m_data) == 8); |
| 156 | _CASSERT(offsetof(struct _mbuf, _m_len) == 12); |
| 157 | #endif /* !__LP64__ */ |
| 158 | #ifdef KERNEL |
| 159 | _CASSERT(offsetof(struct _mbuf, _m_next) == |
| 160 | offsetof(struct mbuf, m_next)); |
| 161 | _CASSERT(offsetof(struct _mbuf, _m_data) == |
| 162 | offsetof(struct mbuf, m_data)); |
| 163 | _CASSERT(offsetof(struct _mbuf, _m_len) == |
| 164 | offsetof(struct mbuf, m_len)); |
| 165 | #endif /* KERNEL */ |
| 166 | struct _mbuf m = { |
| 167 | ._m_next = NULL, |
| 168 | ._m_data = __DECONST(uint8_t *, data), |
| 169 | ._m_len = len, |
| 170 | }; |
| 171 | |
| 172 | return (os_cpu_in_cksum_mbuf(&m, len, 0, initial_sum)); |
| 173 | } |
| 174 | |
| 175 | #if defined(__i386__) || defined(__x86_64__) |
| 176 | |
| 177 | /* |
| 178 | * Checksum routine for Internet Protocol family headers (Portable Version). |
| 179 | * |
| 180 | * This routine is very heavily used in the network |
| 181 | * code and should be modified for each CPU to be as fast as possible. |
| 182 | * |
| 183 | * A discussion of different implementation techniques can be found in |
| 184 | * RFC 1071. |
| 185 | * |
| 186 | * The default implementation for 32-bit architectures is using |
| 187 | * a 32-bit accumulator and operating on 16-bit operands. |
| 188 | * |
| 189 | * The default implementation for 64-bit architectures is using |
| 190 | * a 64-bit accumulator and operating on 32-bit operands. |
| 191 | * |
| 192 | * Both versions are unrolled to handle 32 Byte / 64 Byte fragments as core |
| 193 | * of the inner loop. After each iteration of the inner loop, a partial |
| 194 | * reduction is done to avoid carry in long packets. |
| 195 | */ |
| 196 | |
| 197 | #if !defined(__LP64__) |
| 198 | /* 32-bit version */ |
| 199 | uint32_t |
| 200 | os_cpu_in_cksum_mbuf(struct _mbuf *m, int len, int off, uint32_t initial_sum) |
| 201 | { |
| 202 | int mlen; |
| 203 | uint32_t sum, partial; |
| 204 | unsigned int final_acc; |
| 205 | uint8_t *data; |
| 206 | boolean_t needs_swap, started_on_odd; |
| 207 | |
| 208 | VERIFY(len >= 0); |
| 209 | VERIFY(off >= 0); |
| 210 | |
| 211 | needs_swap = FALSE; |
| 212 | started_on_odd = FALSE; |
| 213 | sum = (initial_sum >> 16) + (initial_sum & 0xffff); |
| 214 | |
| 215 | for (;;) { |
| 216 | if (PREDICT_FALSE(m == NULL)) { |
| 217 | CKSUM_ERR("%s: out of data\n" , __func__); |
| 218 | return ((uint32_t)-1); |
| 219 | } |
| 220 | mlen = m->_m_len; |
| 221 | if (mlen > off) { |
| 222 | mlen -= off; |
| 223 | data = m->_m_data + off; |
| 224 | goto post_initial_offset; |
| 225 | } |
| 226 | off -= mlen; |
| 227 | if (len == 0) |
| 228 | break; |
| 229 | m = m->_m_next; |
| 230 | } |
| 231 | |
| 232 | for (; len > 0; m = m->_m_next) { |
| 233 | if (PREDICT_FALSE(m == NULL)) { |
| 234 | CKSUM_ERR("%s: out of data\n" , __func__); |
| 235 | return ((uint32_t)-1); |
| 236 | } |
| 237 | mlen = m->_m_len; |
| 238 | data = m->_m_data; |
| 239 | post_initial_offset: |
| 240 | if (mlen == 0) |
| 241 | continue; |
| 242 | if (mlen > len) |
| 243 | mlen = len; |
| 244 | len -= mlen; |
| 245 | |
| 246 | partial = 0; |
| 247 | if ((uintptr_t)data & 1) { |
| 248 | /* Align on word boundary */ |
| 249 | started_on_odd = !started_on_odd; |
| 250 | #if BYTE_ORDER == LITTLE_ENDIAN |
| 251 | partial = *data << 8; |
| 252 | #else |
| 253 | partial = *data; |
| 254 | #endif |
| 255 | ++data; |
| 256 | --mlen; |
| 257 | } |
| 258 | needs_swap = started_on_odd; |
| 259 | while (mlen >= 32) { |
| 260 | __builtin_prefetch(data + 32); |
| 261 | partial += *(uint16_t *)(void *)data; |
| 262 | partial += *(uint16_t *)(void *)(data + 2); |
| 263 | partial += *(uint16_t *)(void *)(data + 4); |
| 264 | partial += *(uint16_t *)(void *)(data + 6); |
| 265 | partial += *(uint16_t *)(void *)(data + 8); |
| 266 | partial += *(uint16_t *)(void *)(data + 10); |
| 267 | partial += *(uint16_t *)(void *)(data + 12); |
| 268 | partial += *(uint16_t *)(void *)(data + 14); |
| 269 | partial += *(uint16_t *)(void *)(data + 16); |
| 270 | partial += *(uint16_t *)(void *)(data + 18); |
| 271 | partial += *(uint16_t *)(void *)(data + 20); |
| 272 | partial += *(uint16_t *)(void *)(data + 22); |
| 273 | partial += *(uint16_t *)(void *)(data + 24); |
| 274 | partial += *(uint16_t *)(void *)(data + 26); |
| 275 | partial += *(uint16_t *)(void *)(data + 28); |
| 276 | partial += *(uint16_t *)(void *)(data + 30); |
| 277 | data += 32; |
| 278 | mlen -= 32; |
| 279 | if (PREDICT_FALSE(partial & 0xc0000000)) { |
| 280 | if (needs_swap) |
| 281 | partial = (partial << 8) + |
| 282 | (partial >> 24); |
| 283 | sum += (partial >> 16); |
| 284 | sum += (partial & 0xffff); |
| 285 | partial = 0; |
| 286 | } |
| 287 | } |
| 288 | if (mlen & 16) { |
| 289 | partial += *(uint16_t *)(void *)data; |
| 290 | partial += *(uint16_t *)(void *)(data + 2); |
| 291 | partial += *(uint16_t *)(void *)(data + 4); |
| 292 | partial += *(uint16_t *)(void *)(data + 6); |
| 293 | partial += *(uint16_t *)(void *)(data + 8); |
| 294 | partial += *(uint16_t *)(void *)(data + 10); |
| 295 | partial += *(uint16_t *)(void *)(data + 12); |
| 296 | partial += *(uint16_t *)(void *)(data + 14); |
| 297 | data += 16; |
| 298 | mlen -= 16; |
| 299 | } |
| 300 | /* |
| 301 | * mlen is not updated below as the remaining tests |
| 302 | * are using bit masks, which are not affected. |
| 303 | */ |
| 304 | if (mlen & 8) { |
| 305 | partial += *(uint16_t *)(void *)data; |
| 306 | partial += *(uint16_t *)(void *)(data + 2); |
| 307 | partial += *(uint16_t *)(void *)(data + 4); |
| 308 | partial += *(uint16_t *)(void *)(data + 6); |
| 309 | data += 8; |
| 310 | } |
| 311 | if (mlen & 4) { |
| 312 | partial += *(uint16_t *)(void *)data; |
| 313 | partial += *(uint16_t *)(void *)(data + 2); |
| 314 | data += 4; |
| 315 | } |
| 316 | if (mlen & 2) { |
| 317 | partial += *(uint16_t *)(void *)data; |
| 318 | data += 2; |
| 319 | } |
| 320 | if (mlen & 1) { |
| 321 | #if BYTE_ORDER == LITTLE_ENDIAN |
| 322 | partial += *data; |
| 323 | #else |
| 324 | partial += *data << 8; |
| 325 | #endif |
| 326 | started_on_odd = !started_on_odd; |
| 327 | } |
| 328 | |
| 329 | if (needs_swap) |
| 330 | partial = (partial << 8) + (partial >> 24); |
| 331 | sum += (partial >> 16) + (partial & 0xffff); |
| 332 | /* |
| 333 | * Reduce sum to allow potential byte swap |
| 334 | * in the next iteration without carry. |
| 335 | */ |
| 336 | sum = (sum >> 16) + (sum & 0xffff); |
| 337 | } |
| 338 | final_acc = ((sum >> 16) & 0xffff) + (sum & 0xffff); |
| 339 | final_acc = (final_acc >> 16) + (final_acc & 0xffff); |
| 340 | return (final_acc & 0xffff); |
| 341 | } |
| 342 | |
| 343 | #else /* __LP64__ */ |
| 344 | /* 64-bit version */ |
| 345 | uint32_t |
| 346 | os_cpu_in_cksum_mbuf(struct _mbuf *m, int len, int off, uint32_t initial_sum) |
| 347 | { |
| 348 | int mlen; |
| 349 | uint64_t sum, partial; |
| 350 | unsigned int final_acc; |
| 351 | uint8_t *data; |
| 352 | boolean_t needs_swap, started_on_odd; |
| 353 | |
| 354 | VERIFY(len >= 0); |
| 355 | VERIFY(off >= 0); |
| 356 | |
| 357 | needs_swap = FALSE; |
| 358 | started_on_odd = FALSE; |
| 359 | sum = initial_sum; |
| 360 | |
| 361 | for (;;) { |
| 362 | if (PREDICT_FALSE(m == NULL)) { |
| 363 | CKSUM_ERR("%s: out of data\n" , __func__); |
| 364 | return ((uint32_t)-1); |
| 365 | } |
| 366 | mlen = m->_m_len; |
| 367 | if (mlen > off) { |
| 368 | mlen -= off; |
| 369 | data = m->_m_data + off; |
| 370 | goto post_initial_offset; |
| 371 | } |
| 372 | off -= mlen; |
| 373 | if (len == 0) |
| 374 | break; |
| 375 | m = m->_m_next; |
| 376 | } |
| 377 | |
| 378 | for (; len > 0; m = m->_m_next) { |
| 379 | if (PREDICT_FALSE(m == NULL)) { |
| 380 | CKSUM_ERR("%s: out of data\n" , __func__); |
| 381 | return ((uint32_t)-1); |
| 382 | } |
| 383 | mlen = m->_m_len; |
| 384 | data = m->_m_data; |
| 385 | post_initial_offset: |
| 386 | if (mlen == 0) |
| 387 | continue; |
| 388 | if (mlen > len) |
| 389 | mlen = len; |
| 390 | len -= mlen; |
| 391 | |
| 392 | partial = 0; |
| 393 | if ((uintptr_t)data & 1) { |
| 394 | /* Align on word boundary */ |
| 395 | started_on_odd = !started_on_odd; |
| 396 | #if BYTE_ORDER == LITTLE_ENDIAN |
| 397 | partial = *data << 8; |
| 398 | #else |
| 399 | partial = *data; |
| 400 | #endif |
| 401 | ++data; |
| 402 | --mlen; |
| 403 | } |
| 404 | needs_swap = started_on_odd; |
| 405 | if ((uintptr_t)data & 2) { |
| 406 | if (mlen < 2) |
| 407 | goto trailing_bytes; |
| 408 | partial += *(uint16_t *)(void *)data; |
| 409 | data += 2; |
| 410 | mlen -= 2; |
| 411 | } |
| 412 | while (mlen >= 64) { |
| 413 | __builtin_prefetch(data + 32); |
| 414 | __builtin_prefetch(data + 64); |
| 415 | partial += *(uint32_t *)(void *)data; |
| 416 | partial += *(uint32_t *)(void *)(data + 4); |
| 417 | partial += *(uint32_t *)(void *)(data + 8); |
| 418 | partial += *(uint32_t *)(void *)(data + 12); |
| 419 | partial += *(uint32_t *)(void *)(data + 16); |
| 420 | partial += *(uint32_t *)(void *)(data + 20); |
| 421 | partial += *(uint32_t *)(void *)(data + 24); |
| 422 | partial += *(uint32_t *)(void *)(data + 28); |
| 423 | partial += *(uint32_t *)(void *)(data + 32); |
| 424 | partial += *(uint32_t *)(void *)(data + 36); |
| 425 | partial += *(uint32_t *)(void *)(data + 40); |
| 426 | partial += *(uint32_t *)(void *)(data + 44); |
| 427 | partial += *(uint32_t *)(void *)(data + 48); |
| 428 | partial += *(uint32_t *)(void *)(data + 52); |
| 429 | partial += *(uint32_t *)(void *)(data + 56); |
| 430 | partial += *(uint32_t *)(void *)(data + 60); |
| 431 | data += 64; |
| 432 | mlen -= 64; |
| 433 | if (PREDICT_FALSE(partial & (3ULL << 62))) { |
| 434 | if (needs_swap) |
| 435 | partial = (partial << 8) + |
| 436 | (partial >> 56); |
| 437 | sum += (partial >> 32); |
| 438 | sum += (partial & 0xffffffff); |
| 439 | partial = 0; |
| 440 | } |
| 441 | } |
| 442 | /* |
| 443 | * mlen is not updated below as the remaining tests |
| 444 | * are using bit masks, which are not affected. |
| 445 | */ |
| 446 | if (mlen & 32) { |
| 447 | partial += *(uint32_t *)(void *)data; |
| 448 | partial += *(uint32_t *)(void *)(data + 4); |
| 449 | partial += *(uint32_t *)(void *)(data + 8); |
| 450 | partial += *(uint32_t *)(void *)(data + 12); |
| 451 | partial += *(uint32_t *)(void *)(data + 16); |
| 452 | partial += *(uint32_t *)(void *)(data + 20); |
| 453 | partial += *(uint32_t *)(void *)(data + 24); |
| 454 | partial += *(uint32_t *)(void *)(data + 28); |
| 455 | data += 32; |
| 456 | } |
| 457 | if (mlen & 16) { |
| 458 | partial += *(uint32_t *)(void *)data; |
| 459 | partial += *(uint32_t *)(void *)(data + 4); |
| 460 | partial += *(uint32_t *)(void *)(data + 8); |
| 461 | partial += *(uint32_t *)(void *)(data + 12); |
| 462 | data += 16; |
| 463 | } |
| 464 | if (mlen & 8) { |
| 465 | partial += *(uint32_t *)(void *)data; |
| 466 | partial += *(uint32_t *)(void *)(data + 4); |
| 467 | data += 8; |
| 468 | } |
| 469 | if (mlen & 4) { |
| 470 | partial += *(uint32_t *)(void *)data; |
| 471 | data += 4; |
| 472 | } |
| 473 | if (mlen & 2) { |
| 474 | partial += *(uint16_t *)(void *)data; |
| 475 | data += 2; |
| 476 | } |
| 477 | trailing_bytes: |
| 478 | if (mlen & 1) { |
| 479 | #if BYTE_ORDER == LITTLE_ENDIAN |
| 480 | partial += *data; |
| 481 | #else |
| 482 | partial += *data << 8; |
| 483 | #endif |
| 484 | started_on_odd = !started_on_odd; |
| 485 | } |
| 486 | |
| 487 | if (needs_swap) |
| 488 | partial = (partial << 8) + (partial >> 56); |
| 489 | sum += (partial >> 32) + (partial & 0xffffffff); |
| 490 | /* |
| 491 | * Reduce sum to allow potential byte swap |
| 492 | * in the next iteration without carry. |
| 493 | */ |
| 494 | sum = (sum >> 32) + (sum & 0xffffffff); |
| 495 | } |
| 496 | final_acc = (sum >> 48) + ((sum >> 32) & 0xffff) + |
| 497 | ((sum >> 16) & 0xffff) + (sum & 0xffff); |
| 498 | final_acc = (final_acc >> 16) + (final_acc & 0xffff); |
| 499 | final_acc = (final_acc >> 16) + (final_acc & 0xffff); |
| 500 | return (final_acc & 0xffff); |
| 501 | } |
| 502 | #endif /* __LP64 */ |
| 503 | |
| 504 | #endif /* __i386__ || __x86_64__ */ |
| 505 | |