1 /*-
2 * Copyright (c) 1990 The Regents of the University of California.
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * from tahoe: in_cksum.c 1.2 86/01/05
30 * from: @(#)in_cksum.c 1.3 (Berkeley) 1/19/91
31 */
32
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35
36 /*
37 * MPsafe: alfred
38 */
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/mbuf.h>
42
43 #include <netinet/in.h>
44 #include <netinet/in_systm.h>
45 #include <netinet/ip.h>
46
47 #include <machine/in_cksum.h>
48
49 /*
50 * Checksum routine for Internet Protocol family headers.
51 *
52 * This routine is very heavily used in the network
53 * code and should be modified for each CPU to be as fast as possible.
54 *
55 * This implementation is 386 version.
56 */
57
58 #undef ADDCARRY
59 #define ADDCARRY(x) if ((x) > 0xffff) (x) -= 0xffff
60 #if !defined(__GNUC__) || defined(__INTEL_COMPILER)
61 /* non gcc parts stolen from sys/alpha/alpha/in_cksum.c */
62 #define REDUCE32 \
63 { \
64 q_util.q = sum; \
65 sum = q_util.s[0] + q_util.s[1] + q_util.s[2] + q_util.s[3]; \
66 }
67 #define REDUCE16 \
68 { \
69 q_util.q = sum; \
70 l_util.l = q_util.s[0] + q_util.s[1] + q_util.s[2] + q_util.s[3]; \
71 sum = l_util.s[0] + l_util.s[1]; \
72 ADDCARRY(sum); \
73 }
74 #endif
75 #define REDUCE {sum = (sum & 0xffff) + (sum >> 16); ADDCARRY(sum);}
76
77 #if !defined(__GNUC__) || defined(__INTEL_COMPILER)
78 static const u_int32_t in_masks[] = {
79 /*0 bytes*/ /*1 byte*/ /*2 bytes*/ /*3 bytes*/
80 0x00000000, 0x000000FF, 0x0000FFFF, 0x00FFFFFF, /* offset 0 */
81 0x00000000, 0x0000FF00, 0x00FFFF00, 0xFFFFFF00, /* offset 1 */
82 0x00000000, 0x00FF0000, 0xFFFF0000, 0xFFFF0000, /* offset 2 */
83 0x00000000, 0xFF000000, 0xFF000000, 0xFF000000, /* offset 3 */
84 };
85
86 union l_util {
87 u_int16_t s[2];
88 u_int32_t l;
89 };
90 union q_util {
91 u_int16_t s[4];
92 u_int32_t l[2];
93 u_int64_t q;
94 };
95
96 static u_int64_t
97 in_cksumdata(const u_int32_t *lw, int len)
98 {
99 u_int64_t sum = 0;
100 u_int64_t prefilled;
101 int offset;
102 union q_util q_util;
103
104 if ((3 & (long) lw) == 0 && len == 20) {
105 sum = (u_int64_t) lw[0] + lw[1] + lw[2] + lw[3] + lw[4];
106 REDUCE32;
107 return sum;
108 }
109
110 if ((offset = 3 & (long) lw) != 0) {
111 const u_int32_t *masks = in_masks + (offset << 2);
112 lw = (u_int32_t *) (((long) lw) - offset);
113 sum = *lw++ & masks[len >= 3 ? 3 : len];
114 len -= 4 - offset;
115 if (len <= 0) {
116 REDUCE32;
117 return sum;
118 }
119 }
120 #if 0
121 /*
122 * Force to cache line boundary.
123 */
124 offset = 32 - (0x1f & (long) lw);
125 if (offset < 32 && len > offset) {
126 len -= offset;
127 if (4 & offset) {
128 sum += (u_int64_t) lw[0];
129 lw += 1;
130 }
131 if (8 & offset) {
132 sum += (u_int64_t) lw[0] + lw[1];
133 lw += 2;
134 }
135 if (16 & offset) {
136 sum += (u_int64_t) lw[0] + lw[1] + lw[2] + lw[3];
137 lw += 4;
138 }
139 }
140 #endif
141 /*
142 * access prefilling to start load of next cache line.
143 * then add current cache line
144 * save result of prefilling for loop iteration.
145 */
146 prefilled = lw[0];
147 while ((len -= 32) >= 4) {
148 u_int64_t prefilling = lw[8];
149 sum += prefilled + lw[1] + lw[2] + lw[3]
150 + lw[4] + lw[5] + lw[6] + lw[7];
151 lw += 8;
152 prefilled = prefilling;
153 }
154 if (len >= 0) {
155 sum += prefilled + lw[1] + lw[2] + lw[3]
156 + lw[4] + lw[5] + lw[6] + lw[7];
157 lw += 8;
158 } else {
159 len += 32;
160 }
161 while ((len -= 16) >= 0) {
162 sum += (u_int64_t) lw[0] + lw[1] + lw[2] + lw[3];
163 lw += 4;
164 }
165 len += 16;
166 while ((len -= 4) >= 0) {
167 sum += (u_int64_t) *lw++;
168 }
169 len += 4;
170 if (len > 0)
171 sum += (u_int64_t) (in_masks[len] & *lw);
172 REDUCE32;
173 return sum;
174 }
175
176 u_short
177 in_addword(u_short a, u_short b)
178 {
179 u_int64_t sum = a + b;
180
181 ADDCARRY(sum);
182 return (sum);
183 }
184
185 u_short
186 in_pseudo(u_int32_t a, u_int32_t b, u_int32_t c)
187 {
188 u_int64_t sum;
189 union q_util q_util;
190 union l_util l_util;
191
192 sum = (u_int64_t) a + b + c;
193 REDUCE16;
194 return (sum);
195 }
196
197 u_short
198 in_cksum_skip(struct mbuf *m, int len, int skip)
199 {
200 u_int64_t sum = 0;
201 int mlen = 0;
202 int clen = 0;
203 caddr_t addr;
204 union q_util q_util;
205 union l_util l_util;
206
207 len -= skip;
208 for (; skip && m; m = m->m_next) {
209 if (m->m_len > skip) {
210 mlen = m->m_len - skip;
211 addr = mtod(m, caddr_t) + skip;
212 goto skip_start;
213 } else {
214 skip -= m->m_len;
215 }
216 }
217
218 for (; m && len; m = m->m_next) {
219 if (m->m_len == 0)
220 continue;
221 mlen = m->m_len;
222 addr = mtod(m, caddr_t);
223 skip_start:
224 if (len < mlen)
225 mlen = len;
226 if ((clen ^ (long) addr) & 1)
227 sum += in_cksumdata((const u_int32_t *)addr, mlen) << 8;
228 else
229 sum += in_cksumdata((const u_int32_t *)addr, mlen);
230
231 clen += mlen;
232 len -= mlen;
233 }
234 REDUCE16;
235 return (~sum & 0xffff);
236 }
237
238 u_int in_cksum_hdr(const struct ip *ip)
239 {
240 u_int64_t sum = in_cksumdata((const u_int32_t *)ip, sizeof(struct ip));
241 union q_util q_util;
242 union l_util l_util;
243
244 REDUCE16;
245 return (~sum & 0xffff);
246 }
247 #else
248
249 /*
250 * These asm statements require __volatile because they pass information
251 * via the condition codes. GCC does not currently provide a way to specify
252 * the condition codes as an input or output operand.
253 *
254 * The LOAD macro below is effectively a prefetch into cache. GCC will
255 * load the value into a register but will not use it. Since modern CPUs
256 * reorder operations, this will generally take place in parallel with
257 * other calculations.
258 */
259 #define ADD(n) __asm __volatile \
260 ("addl %1, %0" : "+r" (sum) : \
261 "g" (((const u_int32_t *)w)[n / 4]))
262 #define ADDC(n) __asm __volatile \
263 ("adcl %1, %0" : "+r" (sum) : \
264 "g" (((const u_int32_t *)w)[n / 4]))
265 #define LOAD(n) __asm __volatile \
266 ("" : : "r" (((const u_int32_t *)w)[n / 4]))
267 #define MOP __asm __volatile \
268 ("adcl $0, %0" : "+r" (sum))
269
270 u_short
271 in_cksum_skip(m, len, skip)
272 struct mbuf *m;
273 int len;
274 int skip;
275 {
276 register u_short *w;
277 register unsigned sum = 0;
278 register int mlen = 0;
279 int byte_swapped = 0;
280 union { char c[2]; u_short s; } su;
281
282 len -= skip;
283 for (; skip && m; m = m->m_next) {
284 if (m->m_len > skip) {
285 mlen = m->m_len - skip;
286 w = (u_short *)(mtod(m, u_char *) + skip);
287 goto skip_start;
288 } else {
289 skip -= m->m_len;
290 }
291 }
292
293 for (;m && len; m = m->m_next) {
294 if (m->m_len == 0)
295 continue;
296 w = mtod(m, u_short *);
297 if (mlen == -1) {
298 /*
299 * The first byte of this mbuf is the continuation
300 * of a word spanning between this mbuf and the
301 * last mbuf.
302 */
303
304 /* su.c[0] is already saved when scanning previous
305 * mbuf. sum was REDUCEd when we found mlen == -1
306 */
307 su.c[1] = *(u_char *)w;
308 sum += su.s;
309 w = (u_short *)((char *)w + 1);
310 mlen = m->m_len - 1;
311 len--;
312 } else
313 mlen = m->m_len;
314 skip_start:
315 if (len < mlen)
316 mlen = len;
317 len -= mlen;
318 /*
319 * Force to long boundary so we do longword aligned
320 * memory operations
321 */
322 if (3 & (int) w) {
323 REDUCE;
324 if ((1 & (int) w) && (mlen > 0)) {
325 sum <<= 8;
326 su.c[0] = *(char *)w;
327 w = (u_short *)((char *)w + 1);
328 mlen--;
329 byte_swapped = 1;
330 }
331 if ((2 & (int) w) && (mlen >= 2)) {
332 sum += *w++;
333 mlen -= 2;
334 }
335 }
336 /*
337 * Advance to a 486 cache line boundary.
338 */
339 if (4 & (int) w && mlen >= 4) {
340 ADD(0);
341 MOP;
342 w += 2;
343 mlen -= 4;
344 }
345 if (8 & (int) w && mlen >= 8) {
346 ADD(0);
347 ADDC(4);
348 MOP;
349 w += 4;
350 mlen -= 8;
351 }
352 /*
353 * Do as much of the checksum as possible 32 bits at at time.
354 * In fact, this loop is unrolled to make overhead from
355 * branches &c small.
356 */
357 mlen -= 1;
358 while ((mlen -= 32) >= 0) {
359 /*
360 * Add with carry 16 words and fold in the last
361 * carry by adding a 0 with carry.
362 *
363 * The early ADD(16) and the LOAD(32) are to load
364 * the next 2 cache lines in advance on 486's. The
365 * 486 has a penalty of 2 clock cycles for loading
366 * a cache line, plus whatever time the external
367 * memory takes to load the first word(s) addressed.
368 * These penalties are unavoidable. Subsequent
369 * accesses to a cache line being loaded (and to
370 * other external memory?) are delayed until the
371 * whole load finishes. These penalties are mostly
372 * avoided by not accessing external memory for
373 * 8 cycles after the ADD(16) and 12 cycles after
374 * the LOAD(32). The loop terminates when mlen
375 * is initially 33 (not 32) to guaranteed that
376 * the LOAD(32) is within bounds.
377 */
378 ADD(16);
379 ADDC(0);
380 ADDC(4);
381 ADDC(8);
382 ADDC(12);
383 LOAD(32);
384 ADDC(20);
385 ADDC(24);
386 ADDC(28);
387 MOP;
388 w += 16;
389 }
390 mlen += 32 + 1;
391 if (mlen >= 32) {
392 ADD(16);
393 ADDC(0);
394 ADDC(4);
395 ADDC(8);
396 ADDC(12);
397 ADDC(20);
398 ADDC(24);
399 ADDC(28);
400 MOP;
401 w += 16;
402 mlen -= 32;
403 }
404 if (mlen >= 16) {
405 ADD(0);
406 ADDC(4);
407 ADDC(8);
408 ADDC(12);
409 MOP;
410 w += 8;
411 mlen -= 16;
412 }
413 if (mlen >= 8) {
414 ADD(0);
415 ADDC(4);
416 MOP;
417 w += 4;
418 mlen -= 8;
419 }
420 if (mlen == 0 && byte_swapped == 0)
421 continue; /* worth 1% maybe ?? */
422 REDUCE;
423 while ((mlen -= 2) >= 0) {
424 sum += *w++;
425 }
426 if (byte_swapped) {
427 sum <<= 8;
428 byte_swapped = 0;
429 if (mlen == -1) {
430 su.c[1] = *(char *)w;
431 sum += su.s;
432 mlen = 0;
433 } else
434 mlen = -1;
435 } else if (mlen == -1)
436 /*
437 * This mbuf has odd number of bytes.
438 * There could be a word split betwen
439 * this mbuf and the next mbuf.
440 * Save the last byte (to prepend to next mbuf).
441 */
442 su.c[0] = *(char *)w;
443 }
444
445 if (len)
446 printf("%s: out of data by %d\n", __func__, len);
447 if (mlen == -1) {
448 /* The last mbuf has odd # of bytes. Follow the
449 standard (the odd byte is shifted left by 8 bits) */
450 su.c[1] = 0;
451 sum += su.s;
452 }
453 REDUCE;
454 return (~sum & 0xffff);
455 }
456 #endif
Cache object: f16be6cb34b39514fce3f048e321485a
|