1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
4 *		operating system.  INET is implemented using the  BSD Socket
5 *		interface as the means of communication with the user level.
6 *
7 *		IP/TCP/UDP checksumming routines
8 *
9 * Authors:	Jorge Cwik, <jorge@laser.satlink.net>
10 *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
11 *		Tom May, <ftom@netcom.com>
12 *		Andreas Schwab, <schwab@issan.informatik.uni-dortmund.de>
13 *		Lots of code moved from tcp.c and ip.c; see those files
14 *		for more names.
15 *
16 * 03/02/96	Jes Sorensen, Andreas Schwab, Roman Hodek:
17 *		Fixed some nasty bugs, causing some horrible crashes.
18 *		A: At some points, the sum (%0) was used as
19 *		length-counter instead of the length counter
20 *		(%1). Thanks to Roman Hodek for pointing this out.
21 *		B: GCC seems to mess up if one uses too many
22 *		data-registers to hold input values and one tries to
23 *		specify d0 and d1 as scratch registers. Letting gcc
24 *		choose these registers itself solves the problem.
25 *
26 * 1998/8/31	Andreas Schwab:
27 *		Zero out rest of buffer on exception in
28 *		csum_partial_copy_from_user.
29 */
30
31#include <linux/module.h>
32#include <net/checksum.h>
33
34/*
35 * computes a partial checksum, e.g. for TCP/UDP fragments
36 */
37
38__wsum csum_partial(const void *buff, int len, __wsum sum)
39{
40	unsigned long tmp1, tmp2;
41	  /*
42	   * Experiments with ethernet and slip connections show that buff
43	   * is aligned on either a 2-byte or 4-byte boundary.
44	   */
45	__asm__("movel %2,%3\n\t"
46		"btst #1,%3\n\t"	/* Check alignment */
47		"jeq 2f\n\t"
48		"subql #2,%1\n\t"	/* buff%4==2: treat first word */
49		"jgt 1f\n\t"
50		"addql #2,%1\n\t"	/* len was == 2, treat only rest */
51		"jra 4f\n"
52	     "1:\t"
53		"addw %2@+,%0\n\t"	/* add first word to sum */
54		"clrl %3\n\t"
55		"addxl %3,%0\n"		/* add X bit */
56	     "2:\t"
57		/* unrolled loop for the main part: do 8 longs at once */
58		"movel %1,%3\n\t"	/* save len in tmp1 */
59		"lsrl #5,%1\n\t"	/* len/32 */
60		"jeq 2f\n\t"		/* not enough... */
61		"subql #1,%1\n"
62	     "1:\t"
63		"movel %2@+,%4\n\t"
64		"addxl %4,%0\n\t"
65		"movel %2@+,%4\n\t"
66		"addxl %4,%0\n\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		"dbra %1,1b\n\t"
80		"clrl %4\n\t"
81		"addxl %4,%0\n\t"	/* add X bit */
82		"clrw %1\n\t"
83		"subql #1,%1\n\t"
84		"jcc 1b\n"
85	     "2:\t"
86		"movel %3,%1\n\t"	/* restore len from tmp1 */
87		"andw #0x1c,%3\n\t"	/* number of rest longs */
88		"jeq 4f\n\t"
89		"lsrw #2,%3\n\t"
90		"subqw #1,%3\n"
91	     "3:\t"
92		/* loop for rest longs */
93		"movel %2@+,%4\n\t"
94		"addxl %4,%0\n\t"
95		"dbra %3,3b\n\t"
96		"clrl %4\n\t"
97		"addxl %4,%0\n"		/* add X bit */
98	     "4:\t"
99		/* now check for rest bytes that do not fit into longs */
100		"andw #3,%1\n\t"
101		"jeq 7f\n\t"
102		"clrl %4\n\t"		/* clear tmp2 for rest bytes */
103		"subqw #2,%1\n\t"
104		"jlt 5f\n\t"
105		"movew %2@+,%4\n\t"	/* have rest >= 2: get word */
106		"swap %4\n\t"		/* into bits 16..31 */
107		"tstw %1\n\t"		/* another byte? */
108		"jeq 6f\n"
109	     "5:\t"
110		"moveb %2@,%4\n\t"	/* have odd rest: get byte */
111		"lslw #8,%4\n\t"	/* into bits 8..15; 16..31 untouched */
112	     "6:\t"
113		"addl %4,%0\n\t"	/* now add rest long to sum */
114		"clrl %4\n\t"
115		"addxl %4,%0\n"		/* add X bit */
116	     "7:\t"
117		: "=d" (sum), "=d" (len), "=a" (buff),
118		  "=&d" (tmp1), "=&d" (tmp2)
119		: "0" (sum), "1" (len), "2" (buff)
120	    );
121	return(sum);
122}
123
124EXPORT_SYMBOL(csum_partial);
125
126
127/*
128 * copy from user space while checksumming, with exception handling.
129 */
130
131__wsum
132csum_and_copy_from_user(const void __user *src, void *dst, int len)
133{
134	/*
135	 * GCC doesn't like more than 10 operands for the asm
136	 * statements so we have to use tmp2 for the error
137	 * code.
138	 */
139	unsigned long tmp1, tmp2;
140	__wsum sum = ~0U;
141
142	__asm__("movel %2,%4\n\t"
143		"btst #1,%4\n\t"	/* Check alignment */
144		"jeq 2f\n\t"
145		"subql #2,%1\n\t"	/* buff%4==2: treat first word */
146		"jgt 1f\n\t"
147		"addql #2,%1\n\t"	/* len was == 2, treat only rest */
148		"jra 4f\n"
149	     "1:\n"
150	     "10:\t"
151		"movesw %2@+,%4\n\t"	/* add first word to sum */
152		"addw %4,%0\n\t"
153		"movew %4,%3@+\n\t"
154		"clrl %4\n\t"
155		"addxl %4,%0\n"		/* add X bit */
156	     "2:\t"
157		/* unrolled loop for the main part: do 8 longs at once */
158		"movel %1,%4\n\t"	/* save len in tmp1 */
159		"lsrl #5,%1\n\t"	/* len/32 */
160		"jeq 2f\n\t"		/* not enough... */
161		"subql #1,%1\n"
162	     "1:\n"
163	     "11:\t"
164		"movesl %2@+,%5\n\t"
165		"addxl %5,%0\n\t"
166		"movel %5,%3@+\n\t"
167	     "12:\t"
168		"movesl %2@+,%5\n\t"
169		"addxl %5,%0\n\t"
170		"movel %5,%3@+\n\t"
171	     "13:\t"
172		"movesl %2@+,%5\n\t"
173		"addxl %5,%0\n\t"
174		"movel %5,%3@+\n\t"
175	     "14:\t"
176		"movesl %2@+,%5\n\t"
177		"addxl %5,%0\n\t"
178		"movel %5,%3@+\n\t"
179	     "15:\t"
180		"movesl %2@+,%5\n\t"
181		"addxl %5,%0\n\t"
182		"movel %5,%3@+\n\t"
183	     "16:\t"
184		"movesl %2@+,%5\n\t"
185		"addxl %5,%0\n\t"
186		"movel %5,%3@+\n\t"
187	     "17:\t"
188		"movesl %2@+,%5\n\t"
189		"addxl %5,%0\n\t"
190		"movel %5,%3@+\n\t"
191	     "18:\t"
192		"movesl %2@+,%5\n\t"
193		"addxl %5,%0\n\t"
194		"movel %5,%3@+\n\t"
195		"dbra %1,1b\n\t"
196		"clrl %5\n\t"
197		"addxl %5,%0\n\t"	/* add X bit */
198		"clrw %1\n\t"
199		"subql #1,%1\n\t"
200		"jcc 1b\n"
201	     "2:\t"
202		"movel %4,%1\n\t"	/* restore len from tmp1 */
203		"andw #0x1c,%4\n\t"	/* number of rest longs */
204		"jeq 4f\n\t"
205		"lsrw #2,%4\n\t"
206		"subqw #1,%4\n"
207	     "3:\n"
208		/* loop for rest longs */
209	     "19:\t"
210		"movesl %2@+,%5\n\t"
211		"addxl %5,%0\n\t"
212		"movel %5,%3@+\n\t"
213		"dbra %4,3b\n\t"
214		"clrl %5\n\t"
215		"addxl %5,%0\n"		/* add X bit */
216	     "4:\t"
217		/* now check for rest bytes that do not fit into longs */
218		"andw #3,%1\n\t"
219		"jeq 7f\n\t"
220		"clrl %5\n\t"		/* clear tmp2 for rest bytes */
221		"subqw #2,%1\n\t"
222		"jlt 5f\n\t"
223	     "20:\t"
224		"movesw %2@+,%5\n\t"	/* have rest >= 2: get word */
225		"movew %5,%3@+\n\t"
226		"swap %5\n\t"		/* into bits 16..31 */
227		"tstw %1\n\t"		/* another byte? */
228		"jeq 6f\n"
229	     "5:\n"
230	     "21:\t"
231		"movesb %2@,%5\n\t"	/* have odd rest: get byte */
232		"moveb %5,%3@+\n\t"
233		"lslw #8,%5\n\t"	/* into bits 8..15; 16..31 untouched */
234	     "6:\t"
235		"addl %5,%0\n\t"	/* now add rest long to sum */
236		"clrl %5\n\t"
237		"addxl %5,%0\n\t"	/* add X bit */
238	     "7:\t"
239		".section .fixup,\"ax\"\n"
240		".even\n"
241		/* If any exception occurs, return 0 */
242	     "90:\t"
243		"clrl %0\n"
244		"jra 7b\n"
245		".previous\n"
246		".section __ex_table,\"a\"\n"
247		".long 10b,90b\n"
248		".long 11b,90b\n"
249		".long 12b,90b\n"
250		".long 13b,90b\n"
251		".long 14b,90b\n"
252		".long 15b,90b\n"
253		".long 16b,90b\n"
254		".long 17b,90b\n"
255		".long 18b,90b\n"
256		".long 19b,90b\n"
257		".long 20b,90b\n"
258		".long 21b,90b\n"
259		".previous"
260		: "=d" (sum), "=d" (len), "=a" (src), "=a" (dst),
261		  "=&d" (tmp1), "=d" (tmp2)
262		: "0" (sum), "1" (len), "2" (src), "3" (dst)
263	    );
264
265	return sum;
266}
267
268
269/*
270 * copy from kernel space while checksumming, otherwise like csum_partial
271 */
272
273__wsum
274csum_partial_copy_nocheck(const void *src, void *dst, int len)
275{
276	unsigned long tmp1, tmp2;
277	__wsum sum = 0;
278	__asm__("movel %2,%4\n\t"
279		"btst #1,%4\n\t"	/* Check alignment */
280		"jeq 2f\n\t"
281		"subql #2,%1\n\t"	/* buff%4==2: treat first word */
282		"jgt 1f\n\t"
283		"addql #2,%1\n\t"	/* len was == 2, treat only rest */
284		"jra 4f\n"
285	     "1:\t"
286		"movew %2@+,%4\n\t"	/* add first word to sum */
287		"addw %4,%0\n\t"
288		"movew %4,%3@+\n\t"
289		"clrl %4\n\t"
290		"addxl %4,%0\n"		/* add X bit */
291	     "2:\t"
292		/* unrolled loop for the main part: do 8 longs at once */
293		"movel %1,%4\n\t"	/* save len in tmp1 */
294		"lsrl #5,%1\n\t"	/* len/32 */
295		"jeq 2f\n\t"		/* not enough... */
296		"subql #1,%1\n"
297	     "1:\t"
298		"movel %2@+,%5\n\t"
299		"addxl %5,%0\n\t"
300		"movel %5,%3@+\n\t"
301		"movel %2@+,%5\n\t"
302		"addxl %5,%0\n\t"
303		"movel %5,%3@+\n\t"
304		"movel %2@+,%5\n\t"
305		"addxl %5,%0\n\t"
306		"movel %5,%3@+\n\t"
307		"movel %2@+,%5\n\t"
308		"addxl %5,%0\n\t"
309		"movel %5,%3@+\n\t"
310		"movel %2@+,%5\n\t"
311		"addxl %5,%0\n\t"
312		"movel %5,%3@+\n\t"
313		"movel %2@+,%5\n\t"
314		"addxl %5,%0\n\t"
315		"movel %5,%3@+\n\t"
316		"movel %2@+,%5\n\t"
317		"addxl %5,%0\n\t"
318		"movel %5,%3@+\n\t"
319		"movel %2@+,%5\n\t"
320		"addxl %5,%0\n\t"
321		"movel %5,%3@+\n\t"
322		"dbra %1,1b\n\t"
323		"clrl %5\n\t"
324		"addxl %5,%0\n\t"	/* add X bit */
325		"clrw %1\n\t"
326		"subql #1,%1\n\t"
327		"jcc 1b\n"
328	     "2:\t"
329		"movel %4,%1\n\t"	/* restore len from tmp1 */
330		"andw #0x1c,%4\n\t"	/* number of rest longs */
331		"jeq 4f\n\t"
332		"lsrw #2,%4\n\t"
333		"subqw #1,%4\n"
334	     "3:\t"
335		/* loop for rest longs */
336		"movel %2@+,%5\n\t"
337		"addxl %5,%0\n\t"
338		"movel %5,%3@+\n\t"
339		"dbra %4,3b\n\t"
340		"clrl %5\n\t"
341		"addxl %5,%0\n"		/* add X bit */
342	     "4:\t"
343		/* now check for rest bytes that do not fit into longs */
344		"andw #3,%1\n\t"
345		"jeq 7f\n\t"
346		"clrl %5\n\t"		/* clear tmp2 for rest bytes */
347		"subqw #2,%1\n\t"
348		"jlt 5f\n\t"
349		"movew %2@+,%5\n\t"	/* have rest >= 2: get word */
350		"movew %5,%3@+\n\t"
351		"swap %5\n\t"		/* into bits 16..31 */
352		"tstw %1\n\t"		/* another byte? */
353		"jeq 6f\n"
354	     "5:\t"
355		"moveb %2@,%5\n\t"	/* have odd rest: get byte */
356		"moveb %5,%3@+\n\t"
357		"lslw #8,%5\n"		/* into bits 8..15; 16..31 untouched */
358	     "6:\t"
359		"addl %5,%0\n\t"	/* now add rest long to sum */
360		"clrl %5\n\t"
361		"addxl %5,%0\n"		/* add X bit */
362	     "7:\t"
363		: "=d" (sum), "=d" (len), "=a" (src), "=a" (dst),
364		  "=&d" (tmp1), "=&d" (tmp2)
365		: "0" (sum), "1" (len), "2" (src), "3" (dst)
366	    );
367    return(sum);
368}
369EXPORT_SYMBOL(csum_partial_copy_nocheck);
370