2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1990 The Regents of the University of California.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * from tahoe: in_cksum.c 1.2 86/01/05
32 * from: @(#)in_cksum.c 1.3 (Berkeley) 1/19/91
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
38 #include <sys/param.h>
39 #include <sys/systm.h>
42 #include <netinet/in.h>
43 #include <netinet/in_systm.h>
44 #include <netinet/ip.h>
46 #include <machine/in_cksum.h>
49 * Checksum routine for Internet Protocol family headers.
51 * This routine is very heavily used in the network
52 * code and should be modified for each CPU to be as fast as possible.
54 * This implementation is 386 version.
58 #define ADDCARRY(x) if ((x) > 0xffff) (x) -= 0xffff
60 * icc needs to be special cased here, as the asm code below results
61 * in broken code if compiled with icc.
63 #if !defined(__GNUCLIKE_ASM) || defined(__INTEL_COMPILER)
64 /* non gcc parts stolen from sys/alpha/alpha/in_cksum.c */
68 sum = q_util.s[0] + q_util.s[1] + q_util.s[2] + q_util.s[3]; \
73 l_util.l = q_util.s[0] + q_util.s[1] + q_util.s[2] + q_util.s[3]; \
74 sum = l_util.s[0] + l_util.s[1]; \
78 #define REDUCE {sum = (sum & 0xffff) + (sum >> 16); ADDCARRY(sum);}
80 #if !defined(__GNUCLIKE_ASM) || defined(__INTEL_COMPILER)
81 static const u_int32_t in_masks[] = {
82 /*0 bytes*/ /*1 byte*/ /*2 bytes*/ /*3 bytes*/
83 0x00000000, 0x000000FF, 0x0000FFFF, 0x00FFFFFF, /* offset 0 */
84 0x00000000, 0x0000FF00, 0x00FFFF00, 0xFFFFFF00, /* offset 1 */
85 0x00000000, 0x00FF0000, 0xFFFF0000, 0xFFFF0000, /* offset 2 */
86 0x00000000, 0xFF000000, 0xFF000000, 0xFF000000, /* offset 3 */
100 in_cksumdata(const u_int32_t *lw, int len)
107 if ((3 & (long) lw) == 0 && len == 20) {
108 sum = (u_int64_t) lw[0] + lw[1] + lw[2] + lw[3] + lw[4];
113 if ((offset = 3 & (long) lw) != 0) {
114 const u_int32_t *masks = in_masks + (offset << 2);
115 lw = (u_int32_t *) (((long) lw) - offset);
116 sum = *lw++ & masks[len >= 3 ? 3 : len];
125 * Force to cache line boundary.
127 offset = 32 - (0x1f & (long) lw);
128 if (offset < 32 && len > offset) {
131 sum += (u_int64_t) lw[0];
135 sum += (u_int64_t) lw[0] + lw[1];
139 sum += (u_int64_t) lw[0] + lw[1] + lw[2] + lw[3];
145 * access prefilling to start load of next cache line.
146 * then add current cache line
147 * save result of prefilling for loop iteration.
150 while ((len -= 32) >= 4) {
151 u_int64_t prefilling = lw[8];
152 sum += prefilled + lw[1] + lw[2] + lw[3]
153 + lw[4] + lw[5] + lw[6] + lw[7];
155 prefilled = prefilling;
158 sum += prefilled + lw[1] + lw[2] + lw[3]
159 + lw[4] + lw[5] + lw[6] + lw[7];
164 while ((len -= 16) >= 0) {
165 sum += (u_int64_t) lw[0] + lw[1] + lw[2] + lw[3];
169 while ((len -= 4) >= 0) {
170 sum += (u_int64_t) *lw++;
174 sum += (u_int64_t) (in_masks[len] & *lw);
180 in_addword(u_short a, u_short b)
182 u_int64_t sum = a + b;
189 in_pseudo(u_int32_t a, u_int32_t b, u_int32_t c)
195 sum = (u_int64_t) a + b + c;
201 in_cksum_skip(struct mbuf *m, int len, int skip)
211 for (; skip && m; m = m->m_next) {
212 if (m->m_len > skip) {
213 mlen = m->m_len - skip;
214 addr = mtod(m, caddr_t) + skip;
221 for (; m && len; m = m->m_next) {
225 addr = mtod(m, caddr_t);
229 if ((clen ^ (long) addr) & 1)
230 sum += in_cksumdata((const u_int32_t *)addr, mlen) << 8;
232 sum += in_cksumdata((const u_int32_t *)addr, mlen);
238 return (~sum & 0xffff);
241 u_int in_cksum_hdr(const struct ip *ip)
243 u_int64_t sum = in_cksumdata((const u_int32_t *)ip, sizeof(struct ip));
248 return (~sum & 0xffff);
253 * These asm statements require __volatile because they pass information
254 * via the condition codes. GCC does not currently provide a way to specify
255 * the condition codes as an input or output operand.
257 * The LOAD macro below is effectively a prefetch into cache. GCC will
258 * load the value into a register but will not use it. Since modern CPUs
259 * reorder operations, this will generally take place in parallel with
260 * other calculations.
263 in_cksum_skip(m, len, skip)
271 int byte_swapped = 0;
272 union { char c[2]; u_short s; } su;
275 for (; skip && m; m = m->m_next) {
276 if (m->m_len > skip) {
277 mlen = m->m_len - skip;
278 w = (u_short *)(mtod(m, u_char *) + skip);
285 for (;m && len; m = m->m_next) {
288 w = mtod(m, u_short *);
291 * The first byte of this mbuf is the continuation
292 * of a word spanning between this mbuf and the
296 /* su.c[0] is already saved when scanning previous
297 * mbuf. sum was REDUCEd when we found mlen == -1
299 su.c[1] = *(u_char *)w;
301 w = (u_short *)((char *)w + 1);
311 * Force to long boundary so we do longword aligned
316 if ((1 & (int) w) && (mlen > 0)) {
318 su.c[0] = *(char *)w;
319 w = (u_short *)((char *)w + 1);
323 if ((2 & (int) w) && (mlen >= 2)) {
329 * Advance to a 486 cache line boundary.
331 if (4 & (int) w && mlen >= 4) {
336 : "g" (((const u_int32_t *)w)[0])
341 if (8 & (int) w && mlen >= 8) {
347 : "g" (((const u_int32_t *)w)[0]),
348 "g" (((const u_int32_t *)w)[1])
354 * Do as much of the checksum as possible 32 bits at at time.
355 * In fact, this loop is unrolled to make overhead from
359 while ((mlen -= 32) >= 0) {
361 * Add with carry 16 words and fold in the last
362 * carry by adding a 0 with carry.
364 * The early ADD(16) and the LOAD(32) are to load
365 * the next 2 cache lines in advance on 486's. The
366 * 486 has a penalty of 2 clock cycles for loading
367 * a cache line, plus whatever time the external
368 * memory takes to load the first word(s) addressed.
369 * These penalties are unavoidable. Subsequent
370 * accesses to a cache line being loaded (and to
371 * other external memory?) are delayed until the
372 * whole load finishes. These penalties are mostly
373 * avoided by not accessing external memory for
374 * 8 cycles after the ADD(16) and 12 cycles after
375 * the LOAD(32). The loop terminates when mlen
376 * is initially 33 (not 32) to guaranteed that
377 * the LOAD(32) is within bounds.
391 : "g" (((const u_int32_t *)w)[4]),
392 "g" (((const u_int32_t *)w)[0]),
393 "g" (((const u_int32_t *)w)[1]),
394 "g" (((const u_int32_t *)w)[2]),
395 "g" (((const u_int32_t *)w)[3]),
396 "g" (((const u_int32_t *)w)[8]),
397 "g" (((const u_int32_t *)w)[5]),
398 "g" (((const u_int32_t *)w)[6]),
399 "g" (((const u_int32_t *)w)[7])
417 : "g" (((const u_int32_t *)w)[4]),
418 "g" (((const u_int32_t *)w)[0]),
419 "g" (((const u_int32_t *)w)[1]),
420 "g" (((const u_int32_t *)w)[2]),
421 "g" (((const u_int32_t *)w)[3]),
422 "g" (((const u_int32_t *)w)[5]),
423 "g" (((const u_int32_t *)w)[6]),
424 "g" (((const u_int32_t *)w)[7])
437 : "g" (((const u_int32_t *)w)[0]),
438 "g" (((const u_int32_t *)w)[1]),
439 "g" (((const u_int32_t *)w)[2]),
440 "g" (((const u_int32_t *)w)[3])
451 : "g" (((const u_int32_t *)w)[0]),
452 "g" (((const u_int32_t *)w)[1])
457 if (mlen == 0 && byte_swapped == 0)
458 continue; /* worth 1% maybe ?? */
460 while ((mlen -= 2) >= 0) {
467 su.c[1] = *(char *)w;
472 } else if (mlen == -1)
474 * This mbuf has odd number of bytes.
475 * There could be a word split betwen
476 * this mbuf and the next mbuf.
477 * Save the last byte (to prepend to next mbuf).
479 su.c[0] = *(char *)w;
483 printf("%s: out of data by %d\n", __func__, len);
485 /* The last mbuf has odd # of bytes. Follow the
486 standard (the odd byte is shifted left by 8 bits) */
491 return (~sum & 0xffff);